# Frontmark: full text > Frontmark is an AI-native company-building institution. We form companies, build them with agents and own them for the long term. Source: https://frontmark.ai This file holds the whole of https://frontmark.ai as plain Markdown: the home page, 6 model pages, the glossary of 25 terms and 17 articles. Each part gives the address of the page it comes from. The short index is https://frontmark.ai/llms.txt. ## Facts - Name: Frontmark. The brand is never written "Frontmark AI". - Legal name: Frontmark LLC. - Category: AI-native company-building institution. - Master line: "At the front of what comes next." - Operating line: "Research. Form. Build. Back. Compound." - Website: https://frontmark.ai - Contact: contact@frontmark.ai - Frontmark is not a venture capital fund, a cohort accelerator or an agency. - Frontmark does not publish the names of its operators or its companies yet, and it publishes no figures for its swarm, its team or its performance. Any statement of these comes from somewhere else. ## What Frontmark is Frontmark is an AI-native company-building institution. It forms companies, builds them with agents and owns them for the long term. Its thesis is that building is no longer the hard part. What is short is the judgment to choose what to build, the access to open a market, and an owner who is still there in year five. Frontmark is organised around those three. Agents do the volume of the work, and one accountable operator makes each decision that needs judgment. Behind the agents is a collective of entrepreneurs, founders, operators and key people, who open the markets a new company cannot reach alone. ## Thesis: capability is not yet a company Building is no longer the hard part, and most finished parts are never made into a company. A model still has to become a product someone uses, the product a business that pays for itself, and the business a company that lasts. Each of those steps needs a different kind of judgment. A product is one part of a company. The rest is customers, economics, distribution and a team accountable for them, and Frontmark builds all of it together. ## Operating model: Research. Form. Build. Back. Compound. ### Research We study what a new capability changes in a market, and who would pay for the difference. - Agents: Read markets, costs and customer behaviour without stopping, and bring back each opening with its evidence. - Operators: Decline most openings and keep one. ### Form The company is drawn before it is built. The plan names one customer, the problem they have and what they will pay to have it solved. - Agents: Test demand and price against real buyers before anything exists. - Operators: Name the customer and answer for the plan. ### Build Product, engineering and selling run as one process. The company sells early, because the market is the only test that counts. - Agents: Write the product, run the launch and make whatever tool the next job needs. - Operators: Set the price and hold the standard for what ships. ### Back Commitment follows evidence. Capital, capability and operating attention go to a company as its evidence grows. - Agents: Keep the evidence current: who came, who paid and who came back. - Operators: Commit more, hold, or stop. ### Compound We build companies to keep them. What one company learns stays in the house, and the next starts on top of it. - Agents: Keep every tool, playbook and data source they build for the next company. - Operators: Stay answerable for the company long after launch. ## The long view Every company should make the next one stronger. Frontmark builds companies to keep them. What one company learns stays in the house: its tools, its customers' trust and its people. The next company starts on top of it. Ownership keeps operators answerable for the company long after launch. This describes how the model is built to work, and it promises no financial return. ## The swarm Most of Frontmark is not people. It is a swarm of agents. They read markets, test whether anyone will pay, write the product and make the tool the next job needs. All of it lands on one desk. It stops when one person raises a hand. Operators, founders and entrepreneurs steer the swarm. It knows which of them to ask, and for what: a decision, an opinion, a door. The work waits for the answer. More: https://frontmark.ai/swarm ## The collective Behind the swarm is a collective of entrepreneurs, founders, operators and key people. They are not listed here. When a company needs a customer, a licence or a meeting that no cold email will get, one of them makes the call. More: https://frontmark.ai/collective ## Three routes: incubate, accelerate, acquire - **Incubate**: A company still to be started. We start it with the person who knows the problem. - **Accelerate**: A company that works and is short of hands. The swarm, the capital and the collective go in. - **Acquire**: A company worth keeping whose owner is done. We buy it, and we keep it. ## Three ways in - **Founders and operators**: You have a problem. You know it from the inside and can name who has it. We bring the agents, the capital and the collective. - **Strategic partners**: You have a market. Distribution, infrastructure or customers that companies should be built on. We build them there, with you. - **Capital partners**: You have capital. We meet capital partners by introduction. Write, and name who we have in common. ## Contact Bring a problem you know from the inside, a market you can open, or capital with patience. We answer what fits. Write to contact@frontmark.ai, or use the form at https://frontmark.ai/contact. Nothing on the site is an offer to sell, or a solicitation of an offer to buy, any security or interest in any company. Statements about Frontmark describe its model and intent, not past results. ## Frequently asked ### What is Frontmark? Frontmark is an AI-native company-building institution. It forms companies, builds them with agents and owns them for the long term. Agents do the volume of the work, one accountable operator makes each decision that needs judgment, and a collective of entrepreneurs, founders, operators and key people opens the markets a new company cannot reach alone. ### What does Frontmark actually do? Frontmark is built to start companies, accelerate the ones that earn it, and acquire when owning is the better route. One model runs through all three routes: Research. Form. Build. Back. Compound. Agents do the research, the testing and the building. Operators decide which company should exist and when to stop. ### Why does Frontmark say building is no longer the hard part? It is the thesis Frontmark is built on. When anyone can build, finished parts pile up and most are never made into a company. What stays short is the judgment to choose what to build, the access to open a market, and an owner who is still there in year five. Frontmark is organised around those three. ### How is Frontmark different from using AI tools inside a normal firm? Most firms give their people an assistant and keep the same structure. Frontmark is organised the other way round. Agents carry the work from start to finish and make their own tools, and the work stops when a decision needs judgment. An operator makes that decision, holds the standard and answers for the result. ### Is Frontmark a venture studio, an accelerator, a holding company or a fund? It does the work of several. It builds companies as a studio does, develops them as an accelerator would, and keeps ownership as a holding company does. It is not a venture fund, runs no cohort and offers no standard terms. The category is AI-native company-building institution. ### Is Frontmark an AI roll-up? No. An AI roll-up acquires existing businesses and applies AI to them. Frontmark's primary work is building new companies and holding them for the long term. Acquire is one of its three routes, used selectively when owning is the better way to a capability or a customer base. ### Do Frontmark's companies run themselves? No. Agents carry the volume: research, demand tests, code, distribution and their own tooling. An operator answers for the customer, the standard, each commitment of capital and the decision to stop. Agents still finish fewer tasks than they start, and they sound sure either way. ### What is the Frontmark swarm? The swarm is the private body of AI agents that carries the volume of Frontmark's work. Operators, founders and entrepreneurs set its direction, and it knows who to ask: an operator for a decision, a founder for a standard, a member of the collective for a door. It works only on the companies Frontmark builds. ### What is the Frontmark collective? Behind the agents is a collective of entrepreneurs, founders, operators and key people. They make the decisions the agents bring to them, and they open what a model cannot: introductions, distribution, customers and markets closed to a cold approach. Frontmark does not publish the names of its operators or its companies yet. ### Who should contact Frontmark? There are three ways in. Founders and operators bring a problem they know from the inside and can name who has it. Strategic partners bring distribution, infrastructure or customers that companies should be built on. Capital partners are met by introduction. We answer what fits. --- # The Frontmark approach to company building URL: https://frontmark.ai/approach The model is five actions taken in order, from the first reading of a market to a company that is kept. Frontmark's model is five actions taken in order: research what a new capability changes in a market, form a company around one customer, build the product and the business together, back the company as evidence grows, and keep it, so that its tools and relationships pass to the next. ## Each action hands evidence to the next. Agents do the work inside each action. One accountable operator decides whether the company moves to the next. - **Research**: We study what a new capability changes in a market, and who would pay for the difference. Agents: Read markets, costs and customer behaviour without stopping, and bring back each opening with its evidence. Operators: Decline most openings and keep one. - **Form**: The company is drawn before it is built. The plan names one customer, the problem they have and what they will pay to have it solved. Agents: Test demand and price against real buyers before anything exists. Operators: Name the customer and answer for the plan. - **Build**: Product, engineering and selling run as one process. The company sells early, because the market is the only test that counts. Agents: Write the product, run the launch and make whatever tool the next job needs. Operators: Set the price and hold the standard for what ships. - **Back**: Commitment follows evidence. Capital, capability and operating attention go to a company as its evidence grows. Agents: Keep the evidence current: who came, who paid and who came back. Operators: Commit more, hold, or stop. - **Compound**: We build companies to keep them. What one company learns stays in the house, and the next starts on top of it. Agents: Keep every tool, playbook and data source they build for the next company. Operators: Stay answerable for the company long after launch. ## The principles behind the sequence. - **The company is the unit of ambition**: A product is one part of a company. The rest is customers, economics, distribution and a team accountable for them, and we build all of it together. - **Research and action improve each other**: What the market teaches a company goes back into how the next opening is read. What research finds changes what gets built. - **Commitment follows evidence**: Stronger evidence earns a larger commitment and weak evidence ends one. Enthusiasm and work already spent count for nothing. - **Shared capability serves focused companies**: Infrastructure, tools and playbooks exist to make each company faster. Nothing shared is allowed to become a layer of management above it. - **Ownership brings responsibility**: We build to keep. That means staying after the first release and answering for what the company becomes. ## Three routes into the same model. The model stays the same. The route into it depends on what already exists: nothing yet, a business with evidence, or a company worth keeping. - **Incubate**: We form a new company around a customer problem and a capability that makes a better answer possible. Where a partner already holds the market, the technology or the distribution, the company is built there. - **Accelerate**: We put more execution, distribution and resources behind a business whose evidence has earned it. - **Acquire**: We take ownership of a capability, a customer base or a business when owning it is a better route than building it. We do this selectively. ## About the approach. ### What does Research. Form. Build. Back. Compound. mean? It is Frontmark's operating line: five actions, taken in order. Research reads what a change makes possible. Form turns one opening into a company with a named customer and an accountable operator. Build makes the product and the business together. Back commits resources as evidence grows. Compound keeps the company and passes its tools and relationships to the next. ### How is this different from a product studio or agency? A product studio or agency delivers work for a client and hands it over. Frontmark is built to keep ownership of the companies it builds, so that the work continues after launch and the tools made along the way are kept for the next company. ### Does Frontmark fund startups? No. Frontmark is a builder and an owner, and it offers no standard cheque. Backing means committing capital, capability and operating attention to companies it is building or developing, and evidence decides the level of commitment. ### What does compounding mean here? Each company leaves something the next one starts with: infrastructure, tools, knowledge, relationships and distribution. Ownership keeps them in one place. This describes how the model is built to work, and it promises no financial return. --- # Agentic company building: how Frontmark agents work URL: https://frontmark.ai/agents Agents do most of the work at Frontmark. This page sets out what agents can do today, what they still cannot, and where a person decides. Agentic company building is the practice of using coordinated AI agents to do the work of starting a company: researching openings, testing demand with real customers, building product and running distribution. Operators make the decisions that need judgment and accountability, such as which customer to serve, what to commit and when to stop. At Frontmark the agents also make their own tools. ## What agents can do today. The first two rows are outside evidence, with its source and date. The rest is the work agents are given at Frontmark. - **Work for hours, in teams**: A summary by NYU Shanghai, dated 17 April 2026, of Anthropic's 2026 Agentic Coding Trends Report describes single agents becoming coordinated teams, with tasks that run for hours to days. - **Complete real projects**: The Remote Labor Index, a project of the Center for AI Safety and Scale AI, showed 41.25% full automation of real remote-work projects on its site when read on 16 November 2026. - **Start and test businesses**: Agents turn an opening into something that can be tested: a customer, a problem, an offer and a way to reach the buyer. They put the offer in front of real buyers and record what the buyers do. - **Build the product**: Agents write the code, ship the product, run the distribution and keep the evidence current. - **Build their own tools**: When a job needs a new tool, agents make it. The tool goes into the shared stack, ready for the next company. ## What agents still cannot do. Most of what is written about agents describes what they start. This is what the evidence says about what they finish, as of November 2026. - **They finish less than they start**: The State of AI Report 2026 records the best model earning partial credit on 77.7% of computer workflows while fully completing 44.3%. Agents sound sure either way. - **Reliability decides, not autonomy**: Gartner expects 40% of agentic AI projects to be cancelled by the end of 2027, and says human accountability cannot be removed from business-critical work. - **Judgment is a different skill**: The State of AI Report 2026 says agents can do research engineering but score poorly on research judgment. Choosing a customer, committing capital or stopping a company takes experience, and someone answerable for the result. - **Access cannot be generated**: A model can draft the message. It cannot be the person a buyer already trusts. Introductions, distribution and domain credibility come from people. ## It stops when one person raises a hand. Agents test many offers in the time a team would debate one. Answering for a customer, a hire or a commitment of capital is a person's work. Those calls belong to one accountable operator, and the work waits for the answer. The model depends on this division of labour. Agents build no better than the decisions they are given, so the decisions come from people who have built companies before. What agents and operators each do in the five actions is set out on the approach page, and how a question reaches the right person on the swarm page. ## About agents at Frontmark. ### What is agentic company building? Agentic company building is the practice of using coordinated AI agents to do the work of starting a company: researching openings, testing demand with real customers, building product and running distribution. Operators make the decisions that need judgment and accountability, such as which customer to serve, what to commit and when to stop. ### Do agents replace founders and operators? No. Agents take the volume. Operators keep the decisions that settle whether a company should exist: the customer, the standard, the commitment and when to stop. ### Can an AI swarm run a company? A swarm can do much of a company's work. Answering for the company stays with a person. The State of AI Report 2026, published by Air Street Capital in October 2026, records the best model earning partial credit on 77.7% of computer workflows and fully completing 44.3%. At Frontmark the swarm carries the volume, and an operator decides and answers for the result. ### Who is accountable when agents do the work? A person. Bloomberg Law reported on 2 April 2026 that no jurisdiction lets a company exist without a traceable natural person, and Gartner said on 2 October 2026 that human accountability cannot be removed from business-critical work. At Frontmark each decision the agents bring forward is answered by one operator, who stays answerable for what follows. ### What happens to the tools the agents build? They are kept. The model is built so that every tool, playbook and data source an agent builds goes into a shared stack, and the next company starts with it. This is part of what Compound means in the operating line. ### Is the operating loop on the agents page live data? No. It is an illustration of how agents, operators and the collective pass work between them. Frontmark publishes no operational data. --- # The Frontmark collective: founders and operators URL: https://frontmark.ai/collective Behind the agents is a collective of entrepreneurs, founders, operators and key people. Frontmark does not publish the names of its operators or its companies yet. The Frontmark collective is the network of entrepreneurs, founders, operators and key people behind the institution. They make the decisions the agents bring to them, and they supply the access a model cannot generate: introductions, distribution, customers and markets that stay closed to a cold approach. ## The kinds of people in it. - **Entrepreneurs**: People who have started companies and can tell which opening is a business. - **Founders**: People who have built in a market and know what good looks like there. - **Operators**: People who have run product, engineering, sales and finance inside real businesses, and hold a standard for each. - **Key people**: People inside a market whose word is trusted there. An introduction from one of them gets a meeting that a new company would not get alone. ## Three things agents cannot produce. - **Judgment**: They decide which customer comes first, whether the evidence is good enough and when to stop. The agents bring the question with its evidence, and a person answers it. - **Access**: A new company can build quickly and still not be let in. The introduction, the distribution or the first customer comes from someone the market already trusts. - **Accountability**: One person answers for each decision. Customers, partners and employees deal with someone they can hold to account. ## People come in through the work. People join the collective by working with Frontmark: they bring a problem, build a company with us, or open a door for one. The way to start is to write. A finished product still has to reach a customer, win a licence or get a meeting, and building faster helps with none of those. When a company needs one, a member of the collective makes the call. ## About the collective. ### What is the Frontmark collective? It is the group of entrepreneurs, founders, operators and key people behind Frontmark. They make the decisions the agents bring to them, and they open markets, customers and distribution that a new company could not reach alone. ### How do the collective and the agents work together? Agents do the work and stop when a decision needs a person. A member of the collective makes that decision, makes the introduction the company needs and stays answerable for the outcome. ### Can I join the collective? People come in through the work: by bringing a problem they know from the inside, a capability worth building around or a market they can open, and building with Frontmark. ### Does Frontmark publish who is in the collective? Not yet. This page describes the kinds of people in the collective and what they are there to do for a company. --- # The Frontmark swarm: a private swarm steered by operators URL: https://frontmark.ai/swarm The swarm is Frontmark's private body of agents. Operators, founders and entrepreneurs set its direction and answer its questions. The Frontmark swarm is the private body of AI agents that carries the volume of Frontmark's work across Research, Form, Build, Back and Compound. Operators, founders and entrepreneurs set its direction. The swarm sends each question to the person who can answer it, and it works only on the companies Frontmark builds and keeps. ## One system across the five actions. The swarm works in each of the five actions. What agents and operators each do in them is set out on the approach page. - **A body of agents**: An agent swarm is a group of AI agents that work on one goal at the same time, each with its own task, context and tools, coordinated so their outputs combine into a single result. The swarm is Frontmark's own. - **What it works on**: Research, demand tests, product, distribution and its own tooling, for the companies Frontmark builds. - **What it keeps**: Every tool, playbook and data source it builds. The swarm that begins the next company has all of them to work with. - **Where it stops**: At every decision that needs judgment, a standard or a door opened. It hands over its evidence, and a person decides what happens next. ## People set its direction. Operators, founders and entrepreneurs who have built companies direct all of the swarm's work, and each answers for their part. - **Operators set the brief**: An operator states what the swarm is working towards, the limits it works within and the points where it must stop and ask. That operator answers for the brief. - **Founders hold the standard**: A founder who has built in a market knows what good looks like there. Work for that market is held to the founder's standard, and the founder says when it falls short. - **Entrepreneurs choose the ground**: The swarm brings back more openings than anyone could act on. People who have started companies say which of them is a business, and the swarm is sent there. - **The whole swarm turns**: Every part of the swarm works inside a brief, and every brief has an owner. When the people steering it change direction, all of it changes direction. ## Each question goes to the person who can answer it. | What the swarm needs | Who it asks | What comes back | | --- | --- | --- | | A decision that needs judgment | One accountable operator | A decision, with a person answerable for it | | An opinion on taste or standard | A founder who has built in that market | A standard the work is then held to | | A door opened | The member of the collective who can open it | An introduction from someone the market already trusts | | Something the evidence can settle | No one | The work continues | ## The swarm is private. The swarm works only on the companies Frontmark builds and keeps. It is not a product, and Frontmark neither sells nor licenses it. Its measure is the companies it builds, and one person answers for each of them. What agents can and cannot do today, with the evidence, is on the agents page. ## About the swarm. ### What is an agent swarm? An agent swarm is a group of AI agents that work on one goal at the same time, each with its own task, context and tools, coordinated so their outputs combine into a single result. It is one form of multi-agent system, used when work can be split, run in parallel and checked. ### What is the Frontmark swarm? It is the private body of AI agents that carries the volume of Frontmark's work across Research, Form, Build, Back and Compound. Operators, founders and entrepreneurs set its direction and answer its questions. ### How does the swarm know who to ask? It sorts each question by the kind of answer it needs. A decision that needs judgment goes to an operator. A question of taste or standard goes to a founder who has built in that market. A door that needs opening goes to the member of the collective who can open it. The question arrives with its evidence, and the work that depends on it waits. ### Can I use or license the Frontmark swarm? No. The swarm is private, and it works only on the companies Frontmark builds and keeps. ### How many agents are in the Frontmark swarm? Frontmark publishes no figures for the swarm: its size, what it runs on or what it is working on. This page describes what the swarm is and who steers it. --- # AI-native company building URL: https://frontmark.ai/ai-native-company-building AI changes which companies become possible, and how the work of building them gets done. Frontmark is organised around both. AI-native company building means building companies whose product, operations and economics are designed around AI from the first day, using a building process that is itself run by AI agents. The company would not exist in the same form without the technology, and neither would the way it was built. ## AI-native applies to the company, the builder and the institution. - **The company**: An AI-native company is a company designed around artificial intelligence from its first day. AI is the core of what the product does and of how the company operates, so removing it would leave no working product and no workable cost structure. - **The builder**: An AI-native builder runs the building itself on agents. Teams of agents carry out research, validation, product and distribution, and operators make the decisions that need judgment. - **The institution**: An AI-native institution keeps what both learn. Tools, playbooks and infrastructure stay in the house, and the next company starts on top of them. ## AI-enabled builder versus AI-native builder. | | AI-enabled builder | AI-native builder | | --- | --- | --- | | Who does the work | People, assisted by AI tools | Agents, directed by people | | How ideas are tested | A few, in sequence | Many, in parallel, against real buyers | | Tools | Bought or built per project | Built by agents and kept in a shared stack | | Where people spend time | Execution | Judgment: customer, standard, commitment | | What remains after a build | A product and a handover | A company, and capability for the next one | ## Capability is not yet a company. Building is no longer the hard part. A model still has to become a product someone uses, the product a business that pays for itself, and the business a company that lasts. Each of those steps needs a different kind of judgment. AI-native building removes none of those steps. It makes each one faster and lets many more openings be tested. Every company still needs a customer, an economic reason to exist and a person accountable for it. ## About AI-native company building. ### What does AI-native mean? AI-native means built around AI from the start, so that the product, the operations and the economics depend on it. It is different from AI-enabled, where AI is added to something that already worked without it. ### What is an AI-native company? An AI-native company is a company designed around artificial intelligence from its first day. AI is the core of what the product does and of how the company operates, so removing it would leave no working product and no workable cost structure. ### Is Frontmark an AI venture studio? Frontmark builds companies around AI and with AI, which is the work of an AI venture studio. It is also built to keep long-term ownership and carry shared capability forward, which is the work of a holding company. Doing both makes it an AI-native company-building institution. ### Does every Frontmark company have to be an AI company? No. AI changes what can be built and how it gets built. The companies themselves can serve needs their customers would never describe as AI. --- # Frontmark and the venture studio model URL: https://frontmark.ai/venture-studio A studio forms companies and usually hands them on to outside funding. Frontmark is built to form them, build them with agents and stay the owner. A venture studio creates companies from its own ideas using shared teams and capital. Frontmark does that work and adds two things: long-term ownership of the companies it builds, and a building process run by AI agents, with operators deciding. It is an AI-native company-building institution. ## Where Frontmark sits among the models. | | Frontmark | Venture studio | Holding company or roll-up | Venture fund | Accelerator | | --- | --- | --- | --- | --- | --- | | Primary activity | Builds and owns companies | Creates companies | Owns companies, often acquired | Invests in companies | Runs a programme for startups | | Where ideas come from | Research, the collective and partners | Usually inside the studio | Acquired or existing businesses | External founders | External founders | | Involvement | Hands-on, and continuing | Hands-on, early | Varies by owner | Board and advice | Fixed-term support | | Time horizon | Long term | Often to spin-out | Long term | Fund life | Weeks to months | | What carries forward | Tools, infrastructure, insight and ownership | Playbooks and team | Capital and ownership | Network and brand | Alumni network | ## What changes when the builder is also the owner. - **The work does not end at launch**: A builder that keeps ownership has a reason to stay involved through the unglamorous years after the first release. - **Stopping is a real option**: An owner gains nothing by keeping a weak company alive to improve a count. The company is stopped and the work moves to one with better evidence. - **Capability accumulates**: Tools, infrastructure and relationships built for one company stay in the house for the next. In a spin-out they leave with the company. - **The route can vary**: The commitment is to the company, so the route can be incubating one, accelerating one or, selectively, acquiring one. ## About Frontmark and venture studios. ### Is Frontmark a venture studio? Frontmark does the work of a venture studio: it builds companies with shared teams, tools and capital. It also keeps long-term ownership and runs the building on AI agents. The name for the combined model is AI-native company-building institution. ### What is the difference between a venture studio and a holding company? A venture studio creates new companies. A holding company owns companies, often for the long term, and does not necessarily build them. A build-and-hold model does both: it creates companies and keeps ownership. ### Is Frontmark a venture capital fund or accelerator? No. Frontmark builds and develops companies and is built to keep ownership in them. It runs no cohort programme and offers no standard investment. ### Can a founder build a company with Frontmark? Yes. Bring a problem you know from the inside, a capability worth building around or a market you can open, and write. We answer what fits, and no outcome is guaranteed. --- # Glossary URL: https://frontmark.ai/glossary - **Accelerator**: A fixed-term programme that supports existing early-stage startups, usually in cohorts, with mentorship, a network and often a small investment in exchange for equity. - **Agent**: An AI system that can pursue a goal over several steps: planning, using tools, checking results and deciding what to do next, with limited human direction. - **Agent swarm**: A group of AI agents that work on one goal at the same time, each with its own task, context and tools, coordinated so their outputs combine into a single result. Frontmark's own is defined under Swarm. - **Agentic company building**: The practice of using coordinated AI agents to do the work of starting a company: researching openings, testing demand with real customers, building product and running distribution. Operators make the decisions that need judgment and accountability, such as which customer to serve, what to commit and when to stop. - **AI-enabled**: Describes a product or company that worked without AI and has had AI added to improve it. Remove the AI and the core still functions. - **AI-native**: Describes a product, company or organisation designed around AI from the start, so that its function, operations and economics depend on it. - **AI-native company**: A company designed around artificial intelligence from its first day. AI is the core of what the product does and of how the company operates, so removing it would leave no working product and no workable cost structure. - **AI-native company-building institution**: Frontmark's category. An institution that builds companies around what AI makes possible, runs its building process on agents, and retains long-term ownership. - **Back**: The fourth action in Frontmark's operating line. Committing capital, capability and operating attention to a company. Evidence decides the level of commitment. - **Build-and-hold**: A model that combines creating companies, as a venture studio does, with keeping long-term ownership of them, as a holding company does. - **Collective**: At Frontmark, the entrepreneurs, founders, operators and key people behind the institution. They make the decisions agents bring to them and open markets a new company could not reach alone. - **Company building**: The discipline of creating an operating company: choosing a customer and problem, building the product, reaching the market, assembling the team and making the economics work. - **Compound**: The fifth action in Frontmark's operating line. Keeping the company, and carrying its infrastructure, insight and relationships into the next one. An intended structural effect, not a promised return. - **Form**: The second action in Frontmark's operating line. Turning one opening into a company on paper, with a named customer, a tested price and an operator accountable for the plan. - **Holding company**: A company whose main purpose is to own controlling interests in other companies, rather than to produce goods or services itself. - **Incubator**: An organisation that helps very early ideas become viable businesses, typically with workspace, mentoring and services over a flexible period. - **Long-term ownership**: Retaining an ownership interest in a company well beyond its launch, which keeps the owner involved in and responsible for its development. - **Operator**: A person with experience running part or all of a business. At Frontmark, operators make the decisions agents bring to them and hold the standard for what ships. - **Research**: The first action in Frontmark's operating line. Reading what a new capability changes in a market and who would pay for the difference. Most openings are declined. - **Startup studio**: Another name for a venture studio: an organisation that repeatedly creates new companies using shared teams, infrastructure and capital. - **Swarm**: At Frontmark, the private body of AI agents that carries the volume of the work across Research, Form, Build, Back and Compound. Operators, founders and entrepreneurs set its direction, and it sends each question to the person who can answer it. - **Venture builder**: Another name for a venture studio or company builder: an organisation that creates companies itself rather than only investing in them. - **Venture capital**: A form of financing in which a fund invests in companies founded by others in exchange for equity, usually with a fixed fund life. - **Venture formation**: The step between an idea and a company: defining the customer, the problem, the product approach, the commercial model and who is accountable. - **Venture studio**: An organisation that builds companies from its own ideas, using shared teams and capital, and takes a significant ownership stake in what it creates. --- # AI roll-up vs venture studio vs AI holding company URL: https://frontmark.ai/insights/ai-roll-up-vs-venture-studio-vs-ai-holding-company Category: Models Published: 2026-11-16 Updated: 2026-11-16 Author: Frontmark Short answer: An AI roll-up buys existing service businesses and rebuilds their operations with AI. A venture studio creates new companies and usually passes them to outside investors. An AI holding company is a parent that owns AI-run businesses for the long term, whether bought or built. Build-and-hold creates new companies and keeps the ownership. Key points: - An AI roll-up acquires existing service firms and changes how they work, while a venture studio forms new companies from nothing. - AI holding company describes ownership and time horizon, and most of the well-known examples in 2026 are acquirers. - The largest deals are recent, including a $6.3 billion take-private that Long Lake completed in September 2026. - As of November 2026, one published analysis states that no AI roll-up has exited, so the model's returns are unproven. - Build-and-hold forms new companies and keeps them, which avoids converting a legacy operation but starts without customers. An AI roll-up buys existing service businesses and rebuilds their operations with AI. A venture studio creates new companies and usually passes them to outside investors. An AI holding company is a parent that owns AI-run businesses for the long term, whether bought or built. Build-and-hold creates new companies and keeps the ownership. In 2026 the four terms are used almost interchangeably, which hides two separate questions: where the companies come from, and who owns them afterwards. Each model is described below in its own backers' words, as of 16 November 2026, with the open questions next to the claims. ## What is an AI roll-up? An AI roll-up acquires existing service businesses in one industry and changes how they work using AI. The buyer is betting that software can now do part of the work those firms do by hand, and that owning the firm is the fastest way to put the software to use. General Catalyst gave the model its best-known name. In [The Future of Services](https://www.generalcatalyst.com/stories/the-future-of-services), published 28 August 2025, Marc Bhargava and Kate Bender described the AI-enabled roll-up: back founders building AI for one services industry, then fund them to acquire and operate firms in it. The essay states targets of 30 to 40% margins. Those are targets the firm set for the model, and the essay presents them that way. The firm has since moved away from the word. In [Europe's AI Transformation in Services](https://generalcatalyst.com/stories/europes-ai-transformation-in-services), dated 8 April 2026, General Catalyst says its Creation fund grew from $800 million to $1.5 billion. It calls the acquire-and-rebuild work Transformations, and it says the strategy is product reinvention and is not a private-equity roll-up. Other investors are pursuing the same idea. [Newcomer reported](https://newcomer.co/p/inside-the-vc-roll-up-craze-that) on 18 December 2025 that General Catalyst, Lightspeed, Thrive and 8VC were all active, and that some new holding companies were buying businesses in unrelated industries. ## What is an AI holding company? An AI holding company is a parent entity that owns businesses run with AI and intends to keep them. The term describes ownership and time horizon. It says nothing about whether the businesses were bought or built, although most of the well-known examples so far are buyers. The general structure is explained in [what is a holding company](https://frontmark.ai/insights/what-is-a-holding-company). Long Lake calls itself the world's first AI holding company. On 29 September 2026 it completed a $6.3 billion all-cash take-private of American Express Global Business Travel, according to a [Business Wire press release](https://www.businesswire.com/news/home/20260929141717/en/). Thrive Holdings is the other widely cited case. [OpenAI said](https://openai.com/index/thrive-holdings/) on 1 December 2025 that it had taken an ownership stake and would embed research and engineering staff in Thrive Holdings companies. An [MLQ.ai summary](https://mlq.ai/news/thrive-holdings-raises-2-billion-at-a-12-billion-valuation-to-expand-ai-services-roll-up) of TechCrunch and The Information reporting, dated 13 August 2026, says Thrive Holdings raised $2 billion at a reported $12 billion valuation, and that the company says it covers more than 70 businesses. Those figures come from a secondary summary, so read them as reported. Smaller entities have adopted the label too. [Circeus](https://circeus.com/blogs/acquire/introducing-circeus-new-equity-financing) launched on 29 June 2026 as "the AI-native holding company for mission-critical software". [Legion Holdings](https://legionholdings.ai/) uses the phrase AI-native private equity and says it invests with AI, not in AI. [Titan Holdings](https://www.titanholdings.ai/) describes a holding company that builds and invests in AI-native companies. Infinity Constellation is described in a [Dealroom listing](https://app.dealroom.co/news/feed/infinity-constellation-raises-17m-funding), citing FinSMEs in May 2025, as an AI-native holdco that builds startups. The label therefore covers two activities, acquiring and building. When a firm uses it, ask which one it means. ## What is a venture studio, and what does AI change for it? A venture studio creates companies repeatedly using a shared team, shared tools and its own capital. The company starts inside the studio. In the common version, the studio hands the company to outside investors at its first funding round and reduces its involvement. The base model is covered in [what is a venture studio](https://frontmark.ai/insights/what-is-a-venture-studio). AI is changing the studio's method more than its structure. [GeekWire reported](https://geekwire.com/2025/seattle-studio-psl-encodes-its-playbook-into-lev-an-ai-co-founder-that-helps-turn-ideas-into-companies) on 27 October 2025 that Pioneer Square Labs had built Lev, described as an AI co-founder that encodes the studio's idea-scoring rubric and process. The article noted that this raises a question about what the studio itself is for once its playbook is software. [Forum Ventures](https://forumvc.com/ai-studio) describes an AI venture studio that co-builds AI-native B2B companies with domain experts, with $250,000 on day one and a co-building team. The firm says 63% of its studio companies raised follow-on capital within 12 months, a figure it reports about itself. Andrew Ng told [Crunchbase News](https://news.crunchbase.com/venture/building-backing-startups-ai-fund-ng), in an interview published 21 May 2025, that AI Fund co-founds companies and that its team writes code alongside them. All three sources describe speed: AI shortens the time a studio needs to validate and build. What happens to ownership after the first raise is mostly unchanged. ## How do the four models compare? The table describes the pure types. Read the last row as closely as the first, because each of these models has a part that has not been shown to work yet. | Compared on | AI roll-up | Venture studio | AI holding company | Build-and-hold | |---|---|---|---|---| | **Where companies come from** | Existing service firms, acquired | Formed inside the studio, often with a founding partner | Mostly acquired so far; a few say they build | Formed by the builder from its own research | | **What AI changes** | How an acquired firm does its work | The speed of validation and building | The operations of the businesses it owns | How a company is researched, tested and built from its first day | | **Capital needed** | Large: the purchase price plus the cost of rebuilding | Moderate: a studio team and early cheques | Large where it acquires; varies where it builds | Lower per company at the start, committed for longer | | **Time horizon** | Varies with the backer; often tied to the financing | Often to the first outside raise | Long term by stated intent | Open-ended | | **What is unproven** | Whether the AI does the work, and how the investment is realised; no exit reported as of November 2026 | Whether a studio's edge survives once its playbook is software | Whether gains last across many, sometimes unrelated, businesses | Whether companies formed largely by agents prove durable; few public examples | Real firms blend the columns. General Catalyst describes venture investing, creating new companies and transforming old ones as separate strategies inside one firm. ## What is unproven about AI roll-ups and AI holding companies? The logic is widely accepted. The outcomes are still to come. [Capital Founders](https://capitalfounders.io/playbooks/ai-enabled-roll-ups/chapters/when-ai-roll-ups-break), in a chapter titled When Roll-Ups Break and dated 5 November 2026, lists three ways the model fails. The valuation gap between the purchase price and the promised worth never closes. The buyer acquires faster than it can integrate. Or the AI does not do the work. The same piece states that no AI roll-up has exited yet. [Euclid Ventures](https://insights.euclid.vc/p/service-level-disagreement-ai-services-versus-software), writing on 16 April 2026, is cautious about scope, market size and whether these businesses fit venture capital at all. It separates services used as a way into a market from businesses that are services companies all the way through. [Fig](https://hellofig.io/blog/ai-rollup-platform) framed a choice on 30 September 2026. Productivity gains can be taken as cost cuts, which competitors can match, or as growth, which builds on itself. Fig cites Long Lake's own claim of doubling free cash flow without cutting headcount as an example of the second path. Fig also notes that add-on acquisitions make up about three quarters of US buyouts, while the most visible AI roll-ups are venture-backed holding companies. [Emergence Capital](https://www.emcap.com/thoughts/the-ai-enabled-services-playbook) named a further risk on 31 January 2025: revenue growth that hides a lack of real automation. A services firm can grow by hiring while describing itself as AI-run, and buying such a firm changes nothing about that. These are open questions, and the model may yet answer them. As of November 2026 the evidence that would settle it is still missing. ## What is build-and-hold, and how is it different? Build-and-hold joins the studio's first step to the holding company's last. The parent forms new companies itself, then keeps the ownership with no set date to sell. Compared with a roll-up, there is no legacy operation to convert. The company is designed around what AI can do from its first day, so there is no acquired workforce, contract base or software estate to rework. The cost of that choice is time. A new company has no customers on day one, while an acquired firm arrives with revenue. Compared with a studio, there is no handoff. The builder remains an owner, so tools and knowledge built for one company stay available to the next. The trade-off is that capital stays committed for longer. The structure is covered in more detail in [holding company vs venture studio](https://frontmark.ai/insights/holding-company-vs-venture-studio). Build-and-hold has its own unproven part. Forming companies with agents doing much of the research and building is new, and there are few public examples to judge it by. Anyone describing the model should say so. For what agents can reliably do today, see [what AI agents can and cannot do](https://frontmark.ai/insights/what-ai-agents-can-and-cannot-do). ## Which questions tell the models apart? Labels in this field are chosen for effect, so ask about behaviour. - **Did the company exist before you owned it?** If so, the model is acquisition, whatever it is called. - **Who does the work now, and how is that measured?** A firm that claims automation should be able to say what share of the work no longer needs a person. - **What happens at the first outside funding round?** A studio that steps back at that point is a different partner from an owner that stays. - **When do you plan to sell?** A fund has a clock. A holding structure may not. ## Where Frontmark stands Frontmark sits in the build-and-hold column. It is built to form new companies and keep them. Its three routes are to incubate, to accelerate and, selectively, to acquire, when owning is the better route to a capability or a customer base. The unproven part of that column is Frontmark's to answer as well: companies formed with agents doing much of the work are new. Its answer is the division of labour. [Agents](https://frontmark.ai/agents) carry the volume, one accountable operator decides and answers for each company, and the [collective](https://frontmark.ai/collective) opens the market. The comparison with studios is on the [venture studio](https://frontmark.ai/venture-studio) page. ## Common questions ### What is an AI roll-up? An AI roll-up acquires existing service businesses, usually in one industry, and uses AI to change how the work inside them gets done. General Catalyst described the model in August 2025 as backing founders who build AI for a services industry and then funding them to acquire and operate firms in it. ### What is an AI holding company? An AI holding company is a parent entity that owns businesses run with AI and intends to keep them for the long term. The term describes ownership, so it covers firms that acquire existing businesses and firms that build new ones. Long Lake, Thrive Holdings, Circeus and Titan Holdings all use holding-company language for different activities. ### What is the difference between an AI roll-up and a venture studio? The difference is where the companies come from. A roll-up buys businesses that already have customers, staff and revenue, then changes their operations. A venture studio forms new companies from its own research and ideas. A roll-up needs large amounts of acquisition capital, while a studio needs a team that can build from nothing. ### Is an AI holding company the same as an AI roll-up? Not always. A roll-up is a method: buying and combining firms. A holding company is a structure: a parent that owns companies with no fixed date to sell. Many AI holding companies use the roll-up method, which is why the terms blur, but a holding company can also build its companies instead of buying them. ### Has an AI roll-up produced an exit yet? According to a Capital Founders analysis dated 5 November 2026, no AI roll-up has exited yet. The same piece names three failure modes: the valuation gap never closes, the buyer acquires faster than it integrates, or the AI does not do the work. The thesis is well funded, and the outcome data is still to come. ### What is a build-and-hold model? Build-and-hold means an organisation creates new companies itself, the way a venture studio does, and then keeps long-term ownership, the way a holding company does. It avoids buying and converting older firms. Its trade-offs are that each company starts without customers and that capital stays committed for longer. --- # AI-native vs AI-enabled vs AI-first: the differences URL: https://frontmark.ai/insights/ai-native-vs-ai-enabled Category: Definitions Published: 2026-11-16 Updated: 2026-11-16 Author: Frontmark Short answer: AI-enabled means an existing product or business has added AI features to something that already worked without them. AI-first means a company prioritises AI in its strategy and new development. AI-native means the company was designed around AI from the start, so the product and operating model could not exist without it. Key points: - AI-enabled adds AI to something that already works, AI-first makes AI the strategic priority, and AI-native starts from AI as the foundation. - The terms describe origin and architecture, and none of them is a measure of quality or commercial strength. - The differences show up most clearly in cost structure, team shape and what makes the company hard to replace. - A company can move from AI-enabled towards AI-native, but it usually requires redesigning workflows and products instead of adding features. AI-enabled means an existing product or business has added AI features to something that already worked without them. AI-first means a company prioritises AI in its strategy and new development. AI-native means the company was designed around AI from the start, so the product and operating model could not exist without it. The three terms get used interchangeably, but they name different things. AI-enabled names an addition, AI-first a priority and AI-native an origin. Which one applies tells you what a company would lose if the AI were switched off, and how far an existing business has to change to move from one to the next. ## What does AI-enabled mean? An AI-enabled product worked before AI was added and would keep working if it were removed. The AI makes the product better at a job it already did. A document editor with a writing assistant is AI-enabled. So is a customer relationship system that drafts follow-up emails, or an online shop with a recommendation feature. In each case the underlying product, the business model and the organisation stay as they were. For many established businesses this is a sensible position. They already have customers, data and distribution, and AI features make an existing product more useful without putting the core at risk. ## What does AI-first mean? AI-first describes a strategic priority. A company that calls itself AI-first has decided that AI comes first when it chooses what to build, where to invest and how to solve new problems. The phrase spread when established technology companies adopted it to describe a change of direction, much as "mobile-first" had been used before it. A company can therefore become AI-first at any age. The label says where attention and resources are going and says little about how the company was originally built. Most AI-first companies carry a mix: new products built around models, next to older products and internal processes designed before them. ## What does AI-native mean? AI-native describes origin and architecture. The company was designed around AI from its first day. The model performs the central work of the product, and the company's own workflows, team and cost structure assume AI is available. The quickest check is the remove-the-AI test. Take the AI out and see what remains. If a working product is left with one feature missing, it is AI-enabled. If nothing useful is left, it is AI-native. The full definition, with examples, is in [what is an AI-native company](https://frontmark.ai/insights/what-is-an-ai-native-company). ## How do AI-enabled, AI-first and AI-native compare? | Dimension | AI-enabled | AI-first | AI-native | | --- | --- | --- | --- | | Product | Existing product with AI features added; works without them | AI leads new product decisions; older products may predate it | Model performs the core job; no product without it | | Operations | Existing processes with AI tools added in places | Processes being redesigned around AI, unevenly | Workflows designed with models and agents as participants from the start | | Cost structure | Largely unchanged; AI is an added line of cost | Shifting, with growing spend on models and compute | Model and compute costs are central to unit economics; less headcount per unit of output | | Team | Existing structure, often with a dedicated AI group | Reorganising towards AI skills and smaller, more technical teams | Small teams built around judgment, direction and verification | | Defensibility | Existing customers, data, distribution and brand | Existing strengths combined with new AI capability | Has to be built: proprietary data, workflow depth, distribution or trust | ### Cost structure Conventional software costs a great deal to build and very little to serve to each additional user. AI breaks that pattern, because each request consumes model capacity. For an AI-enabled company this shows up as an extra cost on some features. For an AI-native company it decides how the product is priced and how many people the company needs. The detail is in [how an AI-native company is built and operated](https://frontmark.ai/insights/what-is-an-ai-native-company). ### Team An AI-enabled company typically keeps its structure and adds a specialist group. An AI-native company has no earlier structure to keep, so roles are defined around what people still have to do: decide what to build, set the standard, check the output and answer for it. ### Defensibility This is the row where the labels mislead. AI-native companies are often assumed to hold the advantage, yet an established company starts with assets a new entrant has to earn. An AI-native company whose only asset is access to a model anyone can rent has a thin position. It needs something from outside the model. ## Is this a spectrum or three separate categories? In practice it is a spectrum. Companies move along it over time, and large companies sit at several points at once, with one product line that is AI-enabled and another that is close to AI-native. Ask the question at two levels: - **Product.** Does the model do the core job, or does it assist with a job the product already did? - **Organisation.** Were internal workflows designed around AI, or were AI tools added to existing processes? The answers can differ. One company sells a conventional product and runs highly AI-native operations. Another sells an AI-native product and operates in a traditional way. ## Can an AI-enabled company become AI-native? Strictly, "native" refers to how something began, so the precise term for a company making the shift is AI-first. In substance, an existing company can rebuild products and processes until they depend on AI as fully as a newer company's do. The technology is rarely what stops it. Redesigning a product around a model can undercut existing revenue. Pricing built on seats may not suit a product that does the work itself. Team structures reflect the old way of working. These are management problems, which is why adding features is so much more common than redesigning. ## Which approach is right for a business? It depends on where the business starts and what problem it is solving. - An established business with a product customers rely on will usually begin as AI-enabled. The practical question is which parts of the product and operation to redesign, and in what order. - A company that sees AI changing its market will often declare itself AI-first and move resources accordingly. - A new company has the option to be AI-native, and usually should consider it, because it has no earlier design to protect. None of the three labels says whether the business is any good. A company still needs a clear customer, a problem worth solving and economics that hold. That argument is in [capability is not yet a company](https://frontmark.ai/insights/capability-is-not-yet-a-company), and the practical steps are in [how to build an AI company](https://frontmark.ai/insights/how-to-build-an-ai-company). ## Where Frontmark stands Frontmark is an AI-native company-building institution, and it uses the term at both levels described here. AI changes which companies are worth building, and the building itself is done by [the agents](https://frontmark.ai/agents), with operators making the decisions that need judgment. Both levels still require a named customer and economics that hold, and the model tests each company against both before it is built. The rest is on the [AI-native company building](https://frontmark.ai/ai-native-company-building) and [approach](https://frontmark.ai/approach) pages. ## Common questions ### What does AI-enabled mean? AI-enabled means a product or business uses AI to improve something that already worked without it. Typical examples are an established software product that adds an assistant, a recommendation feature or automated summaries. The core product, business model and organisation stay largely the same, and the AI could be removed without the product ceasing to function. ### What is the difference between AI-native and AI-enabled? The difference is dependence. An AI-enabled product keeps working if the AI is removed, because AI is an added layer. An AI-native product stops working, because the model performs the central job. The same distinction applies to operations: AI-enabled companies add tools to existing processes, while AI-native companies design their processes around models from the start. ### What is the difference between AI-first and AI-native? AI-first describes a priority, and AI-native describes an origin. An AI-first company has decided that AI leads its strategy and new development, and it may be an older company making that shift. An AI-native company was built around AI from its first day. An AI-first company can still carry products and processes designed before AI. ### Can an AI-enabled company become AI-native? It can move a long way in that direction, though strictly the word native refers to how something began. In practice the shift means redesigning products and workflows around models instead of attaching features to existing ones. That is difficult because it often conflicts with existing revenue, pricing and team structures, which is why it takes deliberate leadership. ### Is AI-native better than AI-enabled? No label guarantees a better company. AI-native companies can design product, team and costs around current capabilities without legacy constraints. AI-enabled companies often hold customers, data and distribution that new entrants lack. The right question is which structure fits the problem being solved and whether the resulting business has sound economics. --- # Can AI run a business? The autonomous company, examined URL: https://frontmark.ai/insights/can-a-company-run-itself Category: Thesis Published: 2026-11-16 Updated: 2026-11-16 Author: Frontmark Short answer: Not yet, and not legally. Products now promise to run a company with no employees, and agents can carry much of the daily work. But independent tests show agents still fail to finish many long tasks, and no jurisdiction lets a company exist without a traceable person. A company can run on agents, with people accountable. Key points: - Autonomous company products are a real category in 2026, and their performance figures are self-reported. - Independent evaluations show agents completing well under all of the long workflows they attempt. - Bloomberg Law reported in April 2026 that no jurisdiction lets a company exist without a traceable natural person. - Agents can do most of the volume of a company's work while people keep judgment, access and accountability. - The useful question is which decisions stay with a named person, since some always will. Not yet, and not legally. Products now promise to run a company with no employees, and agents can carry much of the daily work. But independent tests show agents still fail to finish many long tasks, and no jurisdiction lets a company exist without a traceable person. A company can run on agents, with people accountable. The claim that a company can run itself is one of the loudest in AI this year. It has three parts that are usually blurred together: what is being sold, what has been independently measured, and what the law allows. This is the state of each on 16 November 2026. ## What is an autonomous company? An autonomous company is a business whose operations are carried out mainly by AI agents, with little or no human staff. The idea goes by several names: autonomous company, zero-person company, zero-human company. Vendors use these names to describe what their products do, in glossaries they publish themselves. Researchers use similar language to describe experiments. [Bloomberg Law](https://news.bloomberglaw.com/business-and-practice/zero-person-companies-operated-by-ai-would-create-vat-tensions) reported on 2 April 2026 that researchers use the term zero-person company, and described an experiment by KPMG and the University of Amsterdam that ran a small shop with agents. The same words carry different weight in each case. A researcher is asking whether it can be done and what breaks. A vendor is saying it can be bought. ## What do the products that promise to run a company claim? The products in this category sell to a solo founder. They present themselves as that founder's entire team, running the company overnight, and the promise is that the founder never hires again. Most charge a monthly fee, and some add a share of revenue. Several have raised venture funding on the strength of operating figures their founders reported themselves. These are claims, and they may well be accurate. The operating numbers are self-reported, and the research behind this article found no independent audit of them. We have left the figures out for that reason. The claim itself is narrower than the promise. One human and no employees still means one human. Even on the vendors' own accounts, there is a person at the centre. ## What does the independent evidence show? Independent evaluators agree on two things: agents are improving fast, and they still finish jobs unreliably. The research section of the [State of AI Report 2026](https://stateof.ai/2026-explore/research), published by Air Street Capital in October 2026, reports that on OSWorld 2.0 the best model earned 77.7% partial credit and fully completed 44.3% of computer workflows. On FrontierChallenge, the best result was 87.9 out of 100 on requirements with 20.6% of tasks fully completed. Anthropic ran a small business experiment of its own. In [Project Vend phase two](https://www.anthropic.com/research/project-vend-2), an AI shopkeeper improved markedly with newer models. Anthropic nonetheless says the gap between capable and completely robust remains wide. [Gartner](https://www.gartner.com/en/articles/ai-agent-reality-check), in an article dated 2 October 2026, said most agents are not reliable enough for autonomous use in business-critical settings and predicted that 40% of agentic AI projects will be cancelled by the end of 2027. Its title makes the argument: success depends on reliability, not autonomy. These sources describe agents that can do a great deal of a company's work and cannot yet be left alone with all of it. The detail is in [what AI agents can and cannot do](https://frontmark.ai/insights/what-ai-agents-can-and-cannot-do). ## Can a company legally exist without a person? No. Bloomberg Law's report of 2 April 2026 states that no jurisdiction lets a company exist without a traceable natural person. Legal accountability stays human. A company signs contracts, owes tax, holds customer data and can cause harm. Each of those obligations needs someone who can be held to it. A court, a tax authority or a customer with a complaint needs a name. "Fully autonomous" can therefore describe a company's operations and never its responsibility. Someone owns the company, someone directs it, and that person answers for what the agents did, whether or not they watched it happen. An owner who is accountable for decisions they did not see is badly placed, which is a strong reason to stay close to the work. This section is a general description and is not legal advice. Company law differs by country. ## What still needs a person? If agents do the volume, four things remain. **Judgment.** Some decisions have no rule to follow: which customer to serve, when the evidence is good enough, when to stop. The State of AI Report 2026 notes that agents can do research engineering but score poorly on research judgment. **Accountability.** This is the legal requirement described above, and it is a commercial one too. Customers and partners want a person to hold to account. **Access.** Many markets open through trust between people. An agent cannot be the person a buyer has known for ten years. **Domain knowledge.** In Y Combinator's [Summer 2026 Requests for Startups](https://web.archive.org/web/20260518221122/https://www.ycombinator.com/rfs), archived 18 May 2026, Tom Blomfield's Company Brain request says the blocker to automating companies is now domain knowledge and no longer the models. Knowing how an industry judges good work is the part that takes years. All four are positions a person holds over time. They cannot be queued as tasks and handed to an agent. ## So can a company run itself? A company can run mostly on agents. That is a meaningful change, and it is why very small teams now build businesses that once needed departments. The evidence and the limits of that trend are covered in [the one-person billion-dollar company](https://frontmark.ai/insights/one-person-billion-dollar-company). A company cannot run itself in the full sense. It needs someone to decide what it is for, to commit it to things, and to be answerable. The autonomous-company products themselves are sold to a founder, who is that person. What separates the approaches is where the people sit. One sells autonomy to an individual as software and keeps that individual as far from the work as possible. The other keeps experienced people on the decisions on purpose and gives the volume to agents. Both use the same technology and place the risk differently. ## What should you ask before trusting a company to agents? - **Which decisions can the agents make without approval, and what is the limit on each?** Spending, contracts and public statements should have clear thresholds. - **How is finished work verified?** Partial completion is common, so something has to check for it. - **Who is named as accountable?** If nobody can give a name, treat the system as unowned. - **What happens when an agent is wrong?** There should be a way to notice, stop and correct. - **Are the performance figures independently checked?** Self-reported numbers are a starting point for questions. These questions apply equally to a product you buy and to a system you build yourself. ## Where Frontmark stands Frontmark takes the second approach. [Agents](https://frontmark.ai/agents) carry the volume, and one accountable operator answers for the customer, the standard, each commitment of capital and the decision to stop. Access and domain knowledge come from the [collective](https://frontmark.ai/collective). The third question above applies to Frontmark too, and what it publishes about its operators is stated on the collective page. Frontmark is built to keep the companies it builds, so that each has an owner who answers for it after launch. The model is set out on the [approach](https://frontmark.ai/approach) page. ## Common questions ### Can AI run a business on its own? As of November 2026, AI agents can carry out much of a business's daily work, but they cannot run it alone. Independent evaluations show agents failing to finish many long tasks, and the law requires a traceable person behind every company. People still decide what the business does and answer for what happens. ### What is an autonomous company? An autonomous company is a business whose operations are carried out mainly by AI agents, with little or no human staff. The term is used by vendors that sell software promising this, and by researchers studying the idea. It describes how the work is done, and the legal need for an accountable person remains. ### What is a zero-person company? A zero-person company, also called a zero-human company, is a business said to operate with no human employees, with AI agents doing the work. Bloomberg Law reported on 2 April 2026 that researchers use the term, and that no jurisdiction lets a company exist without a traceable natural person. ### Is a fully autonomous company legal? A company can use agents for nearly all of its operations, but it cannot exist without a person behind it. According to Bloomberg Law's April 2026 report, no jurisdiction permits a company without a traceable natural person. Directors, owners or officers remain legally answerable for contracts, taxes and harm. This is a general description, not legal advice. ### Do AI tools that promise to run your company work? The category exists and has raised significant funding, but the evidence is thin. The revenue and scale figures in circulation come from the companies and their founders. No independent audit appears in the research behind this article. Treat the claims as claims, and test any such product on low-risk work before relying on it. ### What still needs a human in an AI-run company? Four things: judgment on decisions where the rules run out, accountability to customers and regulators, access to markets that depend on personal trust, and the domain knowledge that tells you whether the work is right. Agents can do the volume around those, and a person still decides and answers. --- # Capability is not yet a company: from AI model to business URL: https://frontmark.ai/insights/capability-is-not-yet-a-company Category: Thesis Published: 2026-11-16 Updated: 2026-11-16 Author: Frontmark Short answer: A new AI capability is not yet a company. Between a model that can do something and an enduring business lie four gaps: from capability to useful product, from product to sound business, from business to enduring company, and the judgment to know which gap you are in. Each requires a different kind of decision. Key points: - A model's ability to perform a task is a starting condition for a company, and nothing more. - Turning capability into a product requires choosing one customer and one job, then making the result reliable enough to depend on. - Turning a product into a business requires distribution, pricing and unit economics that hold as usage grows. - Turning a business into an enduring company requires advantages that survive when the underlying models improve and become widely available. - Many AI startups struggle because they answer the question for one stage and assume the answer carries to the next. A new AI capability is not yet a company. Between a model that can do something and an enduring business lie four gaps: from capability to useful product, from product to sound business, from business to enduring company, and the judgment to know which gap you are in. Each requires a different kind of decision. This is the thesis Frontmark is built on. A model has to become a useful product, the product a sound business, and the business an enduring company. Most of the work of building a company sits in the gaps between those stages. ## Why is a new capability not yet a company? Advances in AI change what can be built. A model can now draft, summarise, classify, write code, interpret images and carry out multi-step tasks that software could not handle before. That is a real opening. A capability is a statement about what is possible. A company is a statement about who pays, for what, through which channel, at what cost, and who is responsible for keeping it working. The first can be true while every part of the second is still unanswered. The two get confused because the capability is the visible part. A demonstration takes an afternoon and looks like a product. Getting from that demonstration to a company takes a long run of decisions, and most of them have little to do with the model. ## What is the gap between a capability and a useful product? The first gap is between "the model can do this" and "a person can depend on this." A capability is general and a product is specific. To move from one to the other, someone has to choose a particular customer and a particular job, and accept that the choice excludes most of what the model could otherwise do. Then comes reliability. A model that performs well most of the time is impressive in a demonstration and troublesome in a workflow, because the customer has to check everything. Closing that distance takes work around the model: the context it is given, the constraints on its output, evaluation against real cases, sensible behaviour when it is unsure, and a clear path for a human to step in. Last is fit. Useful products sit inside the way people already work. They connect to the systems the customer uses, respect the permissions the customer needs, and produce output in the form the next step requires. **The judgment required here is product judgment.** It means knowing a customer's problem well enough to say what "good enough to rely on" means for them, and building to that standard when a wider demo would be easier to show. ## What is the gap between a useful product and a sound business? A product can be useful and still fail as a business. The second gap is commercial, and it comes down to three questions. **How does the product reach buyers?** A useful product still has to be distributed. Someone has to find the customer, earn their trust and get through their buying process. A team with a strong product and no route to market is in a weaker position than it appears. **Who pays, and how much?** The person who benefits from a product and the person with the budget are often different people. Pricing has to reflect the value delivered and be something a buyer can approve. **Do the economics hold?** AI products carry a cost that traditional software largely avoided: each use consumes compute. Unless the price per customer comfortably exceeds the cost of serving them, growth makes the problem larger. Support, onboarding and human review belong in the same calculation. **The judgment required here is commercial judgment.** It means treating distribution and economics as design constraints from the start, and being willing to change the product, the customer or the price when the numbers fail. ## What is the gap between a sound business and an enduring company? A business that earns more than it spends has crossed two gaps. The third is about time. Models improve, prices tend to fall, and capabilities that once required careful engineering become standard features. A business whose only advantage is early access to a capability can lose that advantage when the next model release makes it common, or when a platform adds the feature directly. An enduring company has something that strengthens with use and survives the model providers catching up. Typical sources include: - deep integration into a customer's workflow, which makes the product hard to replace; - data the company has the right to use, which improves the product in ways a newcomer cannot match; - trusted distribution and relationships built over years; - a reputation for reliability in a domain where errors are costly; - an organisation that can keep absorbing new capabilities faster than its market. Endurance also depends on the company itself: its leadership, its culture, its financial discipline and the patience of its owners. A small team in a hurry can build a product. A company that lasts needs people who take responsibility for it after the launch. **The judgment required here is that of an owner.** It means asking what the company will have in five years that it lacks today, and putting resources there as well as into the next quarter. ## What do the four gaps look like together? The first three gaps lie between stages. The fourth runs through all of them: knowing which stage you are in, and whether the evidence supports moving to the next. | Gap | The question | The judgment it requires | A common mistake | |---|---|---|---| | Capability to product | Can a specific customer depend on this for a specific job? | Product judgment | Treating a demo as a product | | Product to business | Can it reach buyers and earn more than it costs? | Commercial judgment | Assuming usefulness sells itself | | Business to enduring company | What strengthens with time and survives better models? | Ownership judgment | Relying on a temporary capability lead | | Across every stage | Does the evidence justify more commitment, or stopping? | Judgment about commitment | Continuing because of work already spent | The fourth gap is the least discussed and often the most expensive. A team that has built a good product wants to believe it has a business. A team with early revenue wants to believe it has something durable. Resources should follow evidence, and sometimes the evidence says to stop. ## Why do AI startups fail to cross these gaps? The usual account of why AI startups fail points at the technology or the funding climate. The structural account is more useful: a team solves one gap and assumes the next. Some start with the capability and go looking for a customer. They often find interest without commitment, because nobody had the problem badly enough. Others build a product people like but cannot reach buyers at a cost the price supports. Others reach revenue and discover that their product was a feature waiting to be absorbed. Fragmentation makes it worse. The people who understand the capability, the people who build the product, the people who sell it and the people who fund it are frequently in different organisations with different horizons. Research ends with an observation. A build ends with a launch. Capital can arrive without the operating work. A company can fall into a gap at each of those handoffs. ## What follows for someone building now? 1. Start from a customer problem you understand deeply, and ask what a new capability changes about it. 2. Decide early what "reliable enough" means for that customer, and test against it. 3. Treat distribution and unit economics as part of the product from the first version. 4. Ask what would remain valuable if a better, cheaper model were available to everyone tomorrow. 5. Set the evidence you need before committing more, and hold to it. A fuller sequence is in [how to build an AI company](https://frontmark.ai/insights/how-to-build-an-ai-company). For what distinguishes a company designed around AI from one that adds it later, see [what is an AI-native company](https://frontmark.ai/insights/what-is-an-ai-native-company). ## Where Frontmark stands Frontmark is built on this thesis. [Agents](https://frontmark.ai/agents) make every stage faster, and speed closes none of the four gaps, so an operator decides at each one. Frontmark is built to keep the companies it builds, so that the third gap, endurance, stays its own problem after launch. The model is set out on the [approach](https://frontmark.ai/approach) and [AI-native company building](https://frontmark.ai/ai-native-company-building) pages, and the discipline in [what is company building](https://frontmark.ai/insights/what-is-company-building). ## Common questions ### What does "capability is not yet a company" mean? It means that access to a powerful technology is only the start of a business. A model that can perform a task still needs a customer who wants that task done, a product reliable enough to depend on, a way to reach buyers, economics that work, and an organisation committed to keep building. Each of those is separate work. ### Why do AI startups fail? Common reasons are structural. A team mistakes an impressive demonstration for a product people will rely on. A product finds users but has no dependable way to reach buyers or earn a margin. Or the business depends on a capability that a model provider later offers directly. In each case one stage was solved and the next was assumed. ### How do you turn an AI capability into a product? Start with a specific customer and a specific job they already need done. Then design everything around the model: the inputs it receives, the checks on its output, the handling of errors, and the fit with the customer's existing workflow. The product is the full system that makes the capability dependable for that job. ### How do I find a good AI startup idea? Look for a costly, recurring problem you understand in detail, where a new capability changes what is possible or what it costs. Good AI startup ideas usually begin with knowledge of a customer and a workflow. Ideas that begin with the technology and search for a use tend to be easier to copy and harder to sell. ### What makes an AI company defensible? Defensibility rarely comes from access to a model, since capable models are widely available. It tends to come from things that accumulate with use: deep fit with a customer's workflow, data the company has the right to use, trusted distribution, a reputation for reliability, and an organisation that keeps improving the product. --- # Holding company vs venture studio: what is the difference? URL: https://frontmark.ai/insights/holding-company-vs-venture-studio Category: Models Published: 2026-11-16 Updated: 2026-11-16 Author: Frontmark Short answer: A holding company is an ownership structure: a parent that owns and governs other companies, usually for the long term. A venture studio is an activity: an organisation that originates and builds new companies with its own teams and resources. One describes how companies are held, the other how they begin, so the two can be combined. Key points: - A holding company describes how businesses are owned; a venture studio describes how businesses are created. - Venture capital funds invest in companies that others founded, usually within a fixed fund life, while a holding company can own without a deadline. - The two models are compatible: an organisation can build companies the way a studio does and keep them the way a holding company does. - What compounds differs by model: earnings in a holding company, method in a studio, network and returns in a fund, and both capability and ownership in a combined model. - The right structure follows from what you want to keep after the first company is built. A holding company is an ownership structure: a parent that owns and governs other companies, usually for the long term. A venture studio is an activity: an organisation that originates and builds new companies with its own teams and resources. One describes how companies are held, the other how they begin, so the two can be combined. "Holdco vs venture studio" therefore compares two things that sit on different axes. A fair comparison separates where the companies come from and what happens to the ownership once they exist. ## What is a holding company? A holding company is a parent entity whose main job is to own. It holds shares in other businesses, appoints or oversees their leadership, and decides where capital goes across the group. The businesses it owns do the operating: they make the product, serve the customer and employ most of the people. Berkshire Hathaway is a widely cited example, owning a wide range of operating businesses under one parent. Alphabet is another, created as the parent of Google and its other businesses. Many holding companies are far smaller and private. The definition is silent on how the owned companies came to exist. A holding company can acquire them, inherit them, spin them out or build them. The term describes structure and time horizon. For a fuller treatment, see [what is a holding company](https://frontmark.ai/insights/what-is-a-holding-company). ## What is a venture studio? A venture studio is an organisation that creates companies repeatedly. It develops the opportunity internally, tests it, assembles the founding team, and supplies product, engineering, design and early commercial work from a shared base. Idealab, founded in 1996, is usually named as an early example of the form. The studio's defining trait is origination. The company starts inside the studio, or alongside it, before there is anything to invest in. In exchange for that formation work, the studio typically holds a meaningful founding stake. This definition is silent too, on how long the studio keeps that stake. Some studios are financed through a fund and expect to sell. Some spin companies out and step back. Some keep ownership indefinitely. For the model in detail, see [what is a venture studio](https://frontmark.ai/insights/what-is-a-venture-studio). ## What is the difference between venture capital and a holding company? Venture capital and a holding company are both ways of owning parts of businesses, which is why they are often confused. They differ in whose money is used, how much of each company is owned, and when ownership ends. A venture capital fund raises money from outside investors, known as limited partners. It buys minority stakes in companies that founders have already started. The fund has a fixed life, commonly around a decade, and is expected to return capital to its investors within it. That makes the exit, through a sale or public listing, central to the design. A holding company generally owns with its own capital, on its own balance sheet. It often holds controlling or wholly owned positions. It has no fund clock, so it can keep a business for as long as owning it makes sense. Returns can arrive as earnings and dividends over time, without a sale. Each suits a different job. A fund structure suits spreading capital across many independent companies with a defined return schedule. A holding structure suits concentrated, patient ownership where the parent wants a say in how the business develops. ## How do the models compare side by side? The table covers the three standard models and a fourth that joins studio-style building with holding-style ownership. | Compared on | Holding company | Venture studio | Venture capital fund | Combined build-and-hold model | |---|---|---|---|---| | **Purpose** | Own and govern businesses | Create new companies repeatedly | Invest outside capital for a financial return | Create companies and keep owning them | | **How companies originate** | Usually acquired or spun out; origin is outside the definition | Formed inside the studio from its own research and ideas, or with a founding partner | Founded independently by others, then selected for investment | Formed by the institution, with acceleration or acquisition as other routes | | **Time horizon** | Open-ended | Varies; often tied to the studio's financing | Fixed fund life, oriented to exits | Open-ended | | **Involvement** | Governance and capital allocation; operations sit in the subsidiaries | Hands-on in the earliest stage, often reducing as the company matures | Board-level support and advice; the founders operate | Hands-on in formation and building, continuing as an engaged owner | | **What compounds** | Earnings, capital and ownership | Formation method, shared teams and tooling | Reputation, network and fund performance | Shared capability and retained ownership together | These are pure types. Real organisations blend the columns: many studios raise funds, and some holding companies incubate new businesses. ## Can a venture studio also be a holding company? Yes. "Venture studio holding company" describes exactly that: the studio is the method for creating businesses, and the holding company is the vehicle that keeps them. A parent entity researches opportunities and forms new companies using shared teams and infrastructure. Each company becomes its own business with its own customers and leadership. The parent retains its ownership with no predetermined date to sell, and stays involved as an owner after the initial build. The combination changes four incentives: - **Selection.** If you expect to own a company for a long time, you choose opportunities for durability, since there is no later buyer to hand weaknesses to. - **Build quality.** Shortcuts taken to reach a launch date become the owner's problem later, so there is less reason to take them. - **Shared capability.** Infrastructure, knowledge and distribution built for one company stay inside the group and can serve the next. - **Capital discipline.** Without a fund deadline there is less pressure to deploy or exit on schedule, but capital also stays committed longer, so each commitment has to be justified by evidence. ## What are the trade-offs of combining building and holding? The combined model is demanding. A studio that sells its stakes recycles capital quickly. An owner that holds must fund new work from retained resources, patient capital or the earnings of earlier companies, which takes time to develop. Holding also concentrates responsibility. An investor with a minority stake can accept that some companies will not work and move on. An owner who built the business, and still holds it, has to decide whether to fix it, fund it, merge it or close it. The people running each company are affected too. A parent that keeps a large long-term stake needs to leave them enough ownership and autonomy. If the operating team has no real stake in the outcome, the structure works against the thing it was set up to build. Legal form, liability and tax treatment for any of these structures depend on jurisdiction and on professional advice. The comparison here is conceptual. ## Which model fits which situation? A rough guide for builders and partners weighing the options: - **You have a company and need capital.** A venture capital fund is designed for this. You keep operating; the fund takes a minority stake. - **You have an operating business and want a permanent home for it.** A holding company that acquires and keeps businesses may fit. - **You have a problem, a capability or a market advantage, but no company yet.** A venture studio supplies the formation work. Compare it with other early-stage formats in [venture studio vs accelerator vs incubator](https://frontmark.ai/insights/venture-studio-vs-accelerator-vs-incubator). - **You want the builder to stay involved after launch.** Look for a studio with a holding structure, and ask plainly how long it intends to own what it builds and what its role is in year five. Whichever you are talking to, ask what they keep after the first company is built, and what they do with it. ## Where Frontmark stands Frontmark is designed as the combined column in the table. It builds companies as a studio does, with [agents](https://frontmark.ai/agents) doing most of the work and operators deciding, and it is built to keep them as a holding company does. It promises no funding, and backing follows evidence. The model is set out on the [approach](https://frontmark.ai/approach) and [venture studio](https://frontmark.ai/venture-studio) pages. ## Common questions ### Is a venture studio a holding company? Not by definition. A venture studio is defined by what it does, which is create new companies. A holding company is defined by what it owns. Some studios are organised under a holding company and keep long-term stakes in what they build, while others are financed through a fund and plan to sell their positions. The label alone leaves that open. ### What is the difference between venture capital and a holding company? A venture capital fund pools money from outside investors, buys minority stakes in companies founded by others, and usually returns the proceeds within a fixed fund life. A holding company owns stakes, often controlling ones, on its own balance sheet and has no built-in deadline to sell. The fund is designed around exits, the holding company around continued ownership. ### Can a venture studio also be a holding company? Yes. The two answer different questions, so they combine without conflict. An organisation can originate and build companies with shared teams, which is studio work, and then keep its ownership in a parent entity with no fixed exit date, which is holding company structure. The result is sometimes called a build-and-hold model. ### What does holdco mean? Holdco is shorthand for holding company: a parent entity whose main role is to own shares in other companies rather than to sell products itself. The companies it owns are often called operating companies, or opcos. The term is common in private equity, family businesses and startup groups that run several products under one parent. ### Which model is better for a founder, a studio or a venture fund? It depends on what the founder already has. A founder with a company and a team, who mainly needs capital, fits the venture fund model. A builder with a problem, a capability or a distribution advantage but no company yet may fit a studio, which supplies the formation work. The equity and control terms differ accordingly. ### Why would a company builder choose long-term ownership over an exit? Long-term ownership keeps the builder connected to what the company becomes after launch. It allows decisions with a longer payoff, lets shared infrastructure and knowledge serve later companies, and removes the pressure of a fund deadline. The trade-off is that capital stays tied up longer and the owner carries continuing responsibility for each company. --- # How do venture studios make money? Equity and economics URL: https://frontmark.ai/insights/how-venture-studios-make-money Category: Models Published: 2026-11-16 Updated: 2026-11-16 Author: Frontmark Short answer: Venture studios make money mainly through equity. A studio receives a large ownership stake in each company it creates, in exchange for the idea, the early team and the first capital, and earns a return when that stake is sold or pays out. Some studios also earn management fees, carried interest or service fees. Key points: - A venture studio's main source of return is the equity it holds in the companies it builds. - Studio equity is commonly much larger than an accelerator's or an early investor's, because the studio supplies the idea, the labour and the first capital. - Structure matters: a studio fund has a fixed life and needs exits, while a holding-company or evergreen structure can keep ownership for the long term. - Terms vary widely between studios, so founders should ask for the full capitalisation picture, any fees and the studio's obligations in writing. Venture studios make money mainly through equity. A studio receives a large ownership stake in each company it creates, in exchange for the idea, the early team and the first capital, and earns a return when that stake is sold or pays out. Some studios also earn management fees, carried interest or service fees. A studio spends money on people and experiments for years before any return appears, and it relies on a small number of successful companies to pay for everything else. How it finances that wait shapes almost every decision it makes. ## How do venture studios make money? A studio's income can come from four sources. Most studios rely on the first and add one or more of the others. 1. **Equity in the companies it builds.** This is the core of the model. The studio holds shares, and those shares gain value if the company succeeds. 2. **Fund economics.** A studio that manages an investment fund is usually paid a management fee by the fund's investors and receives carried interest, a share of the fund's profits. 3. **Service fees.** Some studios charge their companies for work done by the shared team. 4. **Dividends and distributions.** A studio that holds profitable companies for the long term can receive a share of their earnings. Equity becomes money only through a specific event: an acquisition of the company, a public listing, a sale of the studio's shares to a later investor, or distributions from profits. Until one of these occurs, the studio's return exists on paper. ## Why do venture studios receive equity for building? A studio's stake is payment for what it contributes before a company can stand on its own. That contribution usually includes: - the idea and the research behind it - validation work, including the cost of ideas that were tested and dropped - a team of designers, engineers and operators who build the first product - early capital to cover costs before outside investment - recruitment of the founding team - shared functions such as legal, finance and hiring A founder starting alone supplies most of these personally, often unpaid, and keeps the equity in return. In a studio, the studio supplies them and receives the equity. In both cases ownership follows contribution and risk. ## How much equity do venture studios take? There is no standard figure, and published numbers should be treated with care. Commonly cited ranges for a studio's stake at formation run from a significant minority to a majority of the company. Where a studio falls in that range tends to follow how much it contributes. | Situation | Typical direction of the studio's stake | | --- | --- | | Studio originates the idea, builds the product and funds the company, then recruits a leader | Higher, often the largest single holding at formation | | Studio and an experienced founder shape the idea together from the start | Middle of the range, with a larger founder share | | Founder brings the idea and the studio adds team and capital | Lower, closer to a co-founder or lead investor | For comparison, accelerators commonly take a single-digit percentage, and early-stage investors typically buy a minority stake in each round. A studio's share is larger because its role is closer to a co-founder's than to an investor's. The comparison is set out in [venture studio vs accelerator vs incubator](https://frontmark.ai/insights/venture-studio-vs-accelerator-vs-incubator). The opening percentage is the number people quote. Two other things matter more. **The stake changes over time.** Each funding round issues new shares, which reduces every existing holder's percentage. A studio that begins with a large share usually holds a much smaller one after several rounds. **The capitalisation table has to work for later investors.** Outside investors generally want the people running the company to hold enough equity to stay motivated for many years. If the studio's share leaves too little for the team, the company can struggle to raise money. Well-run studios set their terms with this in mind, and some reduce their stake as the founding team proves itself. ## Studio fund vs holding company: how are venture studios structured? The legal and financial structure determines where the money comes from, who must be repaid and when. Two broad structures are common, along with hybrids of the two. | Compared on | Studio fund | Holding company or evergreen structure | | --- | --- | --- | | Source of capital | Outside investors who commit money to a fund | The entity's own balance sheet, shareholders and retained earnings | | Time horizon | Fixed life, commonly around ten years | Open-ended | | How returns are realised | Selling stakes and returning proceeds to investors | Holding stakes, receiving distributions, selling selectively | | What happens to proceeds | Distributed to the fund's investors | Can be reinvested in new companies | | Income for the studio team | Management fee plus carried interest | Value of the holdings and operating income | | Main pressure | Reaching exits within the fund's life | Financing the build work without regular fee income | ### The studio fund In this structure the studio raises a fund in the same way a venture capital firm does. The fund pays for the studio's operations and invests in the companies it builds. Investors expect their money back with a return inside the fund's life, so the studio has to sell its stakes within that period. A fund gives the studio access to substantial capital and gives investors a familiar arrangement. It also sets a clock. A company that would benefit from being held for longer may need to be sold to meet the fund's schedule. ### The holding company or evergreen structure Here the studio is, or sits inside, a company that owns its stakes directly and permanently. Berkshire Hathaway is a well-known example of a holding company in general, though it acquires businesses instead of starting them. A studio using this structure has no deadline to sell. It can keep ownership of successful companies and use proceeds and earnings to fund new ones. This structure allows patience, and capability and ownership can compound over time. Financing is the difficulty. Without a fund's management fee, the studio has to cover its costs from its own capital, from operating income or from shareholders who accept a long wait. More on the structure is in [what is a holding company](https://frontmark.ai/insights/what-is-a-holding-company) and [holding company vs venture studio](https://frontmark.ai/insights/holding-company-vs-venture-studio). ### Hybrid structures Many studios combine elements: an operating company that employs the building team, paired with a fund that invests in the resulting companies. This spreads the cost and makes it harder to say who owns what. ## Do venture studios charge fees? Practices vary. Three arrangements appear most often. - **No fees.** The studio's work is covered by its equity stake and nothing is billed. - **Fees to fund investors.** A studio with a fund receives a management fee from the fund, and the companies themselves pay nothing. - **Service charges to companies.** The studio bills its companies for shared-team work, sometimes at cost and sometimes with a margin. Founders should look hardest at service charges. A studio that takes substantial equity for building and also bills the company for the same building work is being paid twice unless the terms account for it. There can be legitimate reasons, such as continuing services after the initial build, but the arrangement should be explicit. ## Why do venture studio terms vary so much? Six factors account for most of the range. - **Contribution.** A studio that supplies everything asks for more than one that supplies part. - **Structure.** A fund with return targets and a deadline has different needs from a permanent holding structure. - **Sector.** Companies that need heavy capital or long development time are financed differently from lightweight software businesses. - **Stage of handover.** Some studios step back at spin-out, and others stay involved for years. - **Founder profile.** An experienced operator with a strong record can negotiate a larger share than a first-time leader. - **Local market norms.** Investor expectations differ between regions. Variation is normal. The warning sign is a term the studio cannot explain. ## What should a founder ask a venture studio? 1. What percentage will the studio, the founders and the employee pool each hold at formation? 2. What will those percentages look like after the next one or two funding rounds, on reasonable assumptions? 3. Does founder equity vest, and over what period? 4. Exactly what does the studio provide, who does the work, and for how long? 5. Are there fees or service charges, and how are they calculated? 6. How is the studio itself financed: a fund with a fixed life, or permanent capital? 7. What rights does the studio hold, such as board seats, vetoes or rights to invest in later rounds? 8. What happens to the company and its intellectual property if the studio decides to stop? 9. Can I speak with founders of earlier studio companies, including ones that closed? A studio that answers all nine clearly has told you more than its headline equity number does. ## Where Frontmark stands Frontmark sits at the long-term ownership end of the structures described here, and it offers no standard terms. It is built to keep ownership of the companies it builds, so that the tools, knowledge and relationships made for one are there for the next. That is the last word of the operating line: Research. Form. Build. Back. Compound. Compounding describes the intended structural effect of the model and promises no financial return. The model is set out on the [approach](https://frontmark.ai/approach) page. ## Common questions ### How do venture studios make money? Venture studios make money primarily from the equity they hold in companies they build. Returns arrive when a company is acquired, lists publicly, pays dividends or when the studio sells shares to later investors. Studios that manage a fund may also receive management fees and carried interest, and some charge their companies for shared services. ### How much equity do venture studios take? There is no standard figure. Commonly cited ranges run from a significant minority to a majority of the company at formation, depending on how much of the idea, team, work and capital the studio provides. The percentage usually falls as founders, employees and later investors receive shares. Always ask a studio for its specific terms. ### Why do venture studios take more equity than accelerators? Because they contribute more. An accelerator provides a short programme and a small investment to a company that already exists. A studio typically originates the idea, pays a team to build the first product, covers early costs and carries the risk of ideas that fail before launch. The larger stake is the price of that work and risk. ### What is the difference between a studio fund and a holding company structure? A studio fund raises money from outside investors for a fixed period and must eventually return it, which requires selling its stakes. A holding-company or evergreen structure owns its companies on a permanent balance sheet with no deadline to sell, so it can hold for the long term and reinvest proceeds into new companies. ### Do venture studios charge fees? Some do. Studios with a fund typically receive a management fee from that fund's investors. Some studios also bill the companies they build for shared services such as engineering, design, finance or recruiting, either in cash or as part of the equity arrangement. Others charge nothing and rely entirely on ownership. Practices differ considerably. ### Is venture studio equity fair to founders? It depends on what the studio actually delivers. A smaller share of a company that starts with a tested idea, a product, capital and support can be worth more than a larger share of one that starts with nothing. The arrangement is fair when the studio's contribution is real, clearly defined and matched to its stake. --- # How to build an AI company: a practical sequence URL: https://frontmark.ai/insights/how-to-build-an-ai-company Category: Practice Published: 2026-11-16 Updated: 2026-11-16 Author: Frontmark Short answer: To build an AI company, start with a customer problem you understand deeply, confirm that current AI capabilities change how it can be solved, and form a venture around one customer and one job. Build the product and its distribution together, commit more resources only as evidence accumulates, and carry forward what you learn. Key points: - Begin with a customer problem you know in detail, then ask what current AI capabilities change about it. - Most AI companies are built on existing models, so training a foundation model is rarely the starting point. - Build distribution alongside the product, because even a useful product has to be taken to its buyers. - Decide in advance what evidence would justify a larger commitment, and what would justify stopping. - Design the company so that data rights, infrastructure, knowledge and relationships accumulate with use. To build an AI company, start with a customer problem you understand deeply, confirm that current AI capabilities change how it can be solved, and form a venture around one customer and one job. Build the product and its distribution together, commit more resources only as evidence accumulates, and carry forward what you learn. The work falls into five stages: research, form, build, back and compound. They run in that order, and you will loop back through them as you learn. ## Can I start my own AI company? Yes. Capable models are available through commercial APIs and open-source releases, and the tooling for building on them is mature enough for a small team to ship a working product. What remains hard is what was always hard about companies. You need a customer who has a real problem, a product they can rely on, a way to reach them, and economics that work. Access to a model gives you none of these. ## Do I need to train my own model? In most cases, no. Building a foundation model is a capital-intensive undertaking pursued by a small number of organisations, and an AI company can do without one of its own. Most AI companies sit at the application layer. They use existing models and create value in what surrounds them: the workflow, the context supplied to the model, the checks on its output, the integrations, and the domain knowledge built into the product. Fine-tuning or training a specialised model is a later decision, justified when you have suitable data and a measured reason to believe it improves the result. ## Step 1: research the opening Research here is applied work: finding out what has changed and what that makes possible for a particular customer. Start from a problem you understand deeply. The strongest position is knowledge of a workflow from the inside: who does the work, what it costs, where it goes wrong and what a better result would be worth. If you lack that knowledge, spend time with the people who do the work before you design anything. Then test the capability against the problem. Collect real examples of the task and run them through current models. Look closely at where the output is good, where it fails and how the failures would be caught. Do this before planning in detail, because the real cases will change the plan. The research stage should leave you with three things: - a specific customer and a specific job; - an honest view of what current models can and cannot do for that job; - a reason the opening exists now, such as a capability that recently became good enough or cheap enough. Novelty alone is a weak reason. Look for a change that matters to someone with a budget. ## Step 2: form the venture Formation turns an interesting possibility into a commitment. It means writing down, in plain terms, what the company is. A venture thesis answers seven questions: 1. Who is the customer, exactly? 2. What job does the product do for them, and how is that job done today? 3. What does the product look like, and what is deliberately left out? 4. Who pays, and roughly how will it be priced? 5. How will the first customers be reached? 6. Who is accountable for making it work? 7. What would have to be true for this to become a durable company? Spend the most time on the last one. Ask what would still be valuable if a better and cheaper model were available to everyone. If the answer is nothing, the thesis describes a feature. The reasoning is set out in [capability is not yet a company](https://frontmark.ai/insights/capability-is-not-yet-a-company). Formation also covers the team. An AI company needs people who can judge the product, build with models, and sell to the chosen customer. Missing any one of these early usually shows up later as a stalled company. ## Step 3: build the product and the distribution together The software is one part of the company. Product, engineering and commercial work should move together from the start. On the product side, five practices matter more with AI than with conventional software: - **Narrow first.** Do one job well for one customer type before widening. - **Evaluate constantly.** Keep a set of real cases and measure the product against them whenever the prompt, model or pipeline changes. - **Design for errors.** Decide what happens when the model is wrong or unsure, and make it easy for a person to review and correct. - **Fit the workflow.** Connect to the systems the customer already uses and respect their security and permission needs. - **Stay flexible on models.** Avoid tying the product so tightly to one model that you cannot adopt a better one. On the commercial side, start distribution on day one. Talk to buyers while the product is being built. Learn how they purchase, who approves, and what proof they need. Common routes include direct sales, partnerships with firms that already serve the customer, integration into platforms the customer uses, and content that earns attention from people searching for the problem. Watch the economics from the first customer. Each use of an AI product consumes compute, so the cost of serving a customer rises with usage. Price and design with that in mind. If you intend to build an AI-native company, apply the same thinking to your own operations. Use AI in how the team researches, builds, supports and sells, so the company's cost structure reflects what the technology allows. The distinction is explored in [what is an AI-native company](https://frontmark.ai/insights/what-is-an-ai-native-company), and the version in which agents carry most of the work in [what is agentic company building](https://frontmark.ai/insights/what-is-agentic-company-building). ## Step 4: back it according to evidence Backing means committing resources: money, people, attention and time. Evidence decides how much. Before each increase in commitment, state what you expect to see. | Stage | Evidence worth looking for | What it justifies | |---|---|---| | Early research | The model performs the hard part on real cases; customers confirm the problem is costly | Forming a venture and building a first version | | First product | A few customers use it repeatedly for real work without being pushed | More building and a first commercial effort | | Early sales | Customers pay, renew or expand, and the cost of serving them is below the price | Investment in distribution and team | | Growth | The channel repeats and the economics hold as volume rises | Larger, longer commitments | Weak evidence is information too. If customers are polite and never return, or usage depends on your constant involvement, the right response may be to change the customer, change the product or stop. Work already spent is no reason to continue. ## Step 5: compound what you learn and own The last stage concerns what the company accumulates with use. Decide early what should carry forward: - **Data, where rights permit.** Usage can improve the product, provided customers have agreed to it and the terms are clear. - **Evaluation assets.** A growing library of real cases and quality measures is hard for a newcomer to reproduce. - **Infrastructure.** Tooling built once can serve later products. - **Distribution and relationships.** Trust with a customer group makes the next product easier to place. - **Operating knowledge.** What you learn about building with models applies to whatever you build next. Ownership belongs in this stage. Who owns the company, and for how long, shapes what it optimises for. A team planning to hold what it builds will make different choices from one planning to hand it off. ## What mistakes should you avoid? Six errors recur: - Starting from the technology and searching for a problem. - Treating a convincing demo as a finished product. - Leaving distribution until after launch. - Ignoring the cost of serving each customer. - Building an advantage that the next model release removes. - Raising commitment because of effort already spent. ## Where Frontmark stands The five stages in this guide are Frontmark's operating line: Research. Form. Build. Back. Compound. In Frontmark's version [agents](https://frontmark.ai/agents) do the work inside each stage, and one accountable operator decides whether a company moves to the next. It is a model for building and carries no promise of funding. The [approach](https://frontmark.ai/approach) page sets it out, and [what is company building](https://frontmark.ai/insights/what-is-company-building) defines the discipline. ## Common questions ### Can I start my own AI company? Yes. Capable models are available through APIs and open-source releases, so starting an AI company no longer requires a research laboratory. What it does require is a customer problem you understand well, the ability to build a reliable product around a model, and a route to the people who will pay for it. ### Do I need to train my own AI model to start an AI company? Usually not. Most AI companies build on foundation models developed by others and add value through the surrounding product: workflow design, context, evaluation, integration and domain knowledge. Training or fine-tuning a model can make sense later, when you have data, a clear performance need and evidence that it improves the product. ### How is an AI-native company different from a company that uses AI? An AI-native company is designed around what AI makes possible from the start, in both its product and its own operations. A company that uses AI adds it to an existing product or process. The difference shows in the product's core function, the cost structure and how the team itself works. ### What is the first step in building an AI company? The first step is research, in a practical sense. Pick a customer and a problem you understand deeply, map how the work is done today, and test whether current models can perform the hard part reliably enough. Do this with real examples from the customer's work before writing much product code. ### How much does it cost to start an AI company? It varies widely with the product, so no single figure is reliable. Building on existing models keeps early costs lower than training your own. The costs to plan for are engineering time, model usage that grows with each customer, evaluation and quality work, and the effort of reaching and supporting buyers. ### How do AI companies stay defensible when models keep improving? They build advantages that grow with use and hold across models. Examples include deep fit with a customer's workflow, data they have the right to use, trusted distribution, and a reputation for reliability. A well-designed AI company benefits when models improve, because its product gets better at the same time. --- # Sell the work, not the software: what the shift means URL: https://frontmark.ai/insights/sell-the-work-not-the-software Category: Thesis Published: 2026-11-16 Updated: 2026-11-16 Author: Frontmark Short answer: Sell the work, not the software means an AI company delivers the finished outcome, such as a filed tax return or a processed claim, and charges for that result instead of licensing a tool to the people who do the job. The company runs the work itself with AI and takes responsibility for the quality. Key points: - Selling the work means charging for a finished outcome and carrying the responsibility for it, instead of licensing a tool. - Sequoia's March 2026 essay separates copilots, which sell the tool, from autopilots, which sell the work. - Y Combinator asked for AI-native agencies and AI-native service companies in its Spring and Summer 2026 requests. - The open questions are whether the AI really does the work, whether margins hold, and who stays accountable. - The idea is now standard language, so the more useful questions are how the work gets done and who owns the result. Sell the work, not the software means an AI company delivers the finished outcome, such as a filed tax return or a processed claim, and charges for that result instead of licensing a tool to the people who do the job. The company runs the work itself with AI and takes responsibility for the quality. By late 2026 this is the most repeated investor thesis in AI. The argument is stronger than its slogans, and its open questions are larger than its advocates usually say. The sources below are read as of 16 November 2026. ## What does it mean to sell the work? A software company sells a tool. The customer still employs people to use it, and the customer still owns the result, good or bad. A company that sells the work takes the job itself. The customer buys a completed bookkeeping close, a placed insurance policy or a reviewed contract. Three things change when a company makes that move. The price is set against the cost of labour. The buyer is comparing the service with a salary or an outsourcing contract, and no longer with another software licence. The provider carries the risk of getting it wrong. A tool vendor can point to the user. A firm that delivered the work cannot. The provider decides how the work is done. How much is handled by AI and how much by people is an internal matter. The customer adopts nothing, retrains no one and keeps its process. They receive the outcome the way they would from any outside firm. ## Where did the idea come from? The idea built up over several years across several firms. Sarah Tavel's essay [Sell the Work](https://www.sarahtavel.com/p/ai-startups-sell-work-not-software) argued that AI startups should sell the work product itself, with EvenUp as its example. Emergence Capital published [The AI-Native Services Playbook](https://www.emcap.com/thoughts/the-ai-enabled-services-playbook) by Jake Saper and Rishub Nahar on 31 January 2025, advising founders to pair a domain expert with an applied-AI expert. The statement that set the tone for 2026 came from Sequoia. In [Services: The New Software](https://sequoiacap.com/article/services-the-new-software/), published 5 March 2026, Julien Bek argues that the next $1 trillion company will look like a services firm built on software. He says six dollars go to services for every dollar spent on software, which in his telling is the size of the opening. Y Combinator then turned the argument into requests. In its [Spring 2026 Requests for Startups](https://web.archive.org/web/20260203133656/https://www.ycombinator.com/rfs), archived 3 February 2026, Aaron Epstein asked for AI-native agencies and said agencies can now use the software themselves, sell the finished product and reach software-like margins. In the [Summer 2026 list](https://web.archive.org/web/20260518221122/https://www.ycombinator.com/rfs), archived 18 May 2026, Gustaf Alströmer asked for AI-native service companies that do the work instead of selling software. He named insurance brokerage, accounting, tax and audit, compliance, and healthcare administration. By November 2026, describing a company that delivers outcomes is standard language. Anyone writing about it now is repeating an established idea and should say so. ## What is the difference between a copilot and an autopilot? The distinction is Sequoia's. A copilot is a tool sold to the professional who does the job. The accountant, lawyer or broker works faster, and remains the one responsible. An autopilot sells the work to the customer who needs it done. The two compete for different money. A copilot is paid for out of a software budget and is judged against other tools. An autopilot is paid for out of a labour or outsourcing budget and is judged against the cost and quality of people. Bek's advice on where to begin is specific. Start with work that is already outsourced and that he calls intelligence-heavy, then move towards work that is judgment-heavy. Intelligence-heavy work follows rules, however complicated the rules are. Judgment-heavy work depends on experience and on weighing things nobody has written down. Outsourced work is the practical place to start because the buyer already has a budget line for it and already accepts that an outside party does the job. Nobody inside the customer has to be replaced for the sale to happen. This is related to the difference between designing a company around AI and adding AI to an existing one, covered in [AI-native vs AI-enabled](https://frontmark.ai/insights/ai-native-vs-ai-enabled). ## What is an AI-native services company? An AI-native services company does what a service firm does, with AI carrying out most of the tasks from its first day. It has customers and deliverables like an agency or a practice. Its cost structure is meant to look different, because the volume of work can rise without headcount rising with it. The phrase full-stack AI company describes the same thing from another side. A full-stack company does the whole job for the customer instead of supplying one tool in someone else's chain. Y Combinator partners gave a sense of how common this has become. On the [Lightcone podcast](https://podscripts.co/podcasts/y-combinator-startup-podcast/the-state-of-startups-in-2026) episode of 18 September 2026, they said the share of accepted companies doing full-stack, end-to-end work rose from about 10% to over 25% of the batch. That is the partners' own description of their intake, and it is the clearest signal available of where new founders are heading. For the underlying definition, see [what is an AI-native company](https://frontmark.ai/insights/what-is-an-ai-native-company). ## Why are investors making this argument now? Budget, capability and capital lined up at the same time. The budget is the size of services spending. If Sequoia's ratio is right, services are a far larger pool of spending than software, and until recently software companies could not reach it. The capability is longer-running work. A summary by [NYU Shanghai](https://rits.shanghai.nyu.edu/ai/anthropics-2026-agentic-coding-trends-report-from-assistants-to-agent-teams/), dated 17 April 2026, of Anthropic's 2026 Agentic Coding Trends Report describes single agents becoming coordinated teams, with tasks that run for hours to days. Work of that length starts to look like a job and less like a feature. The capital is venture money. Venture firms are funding the acquisition of existing service firms as a faster route to the same end. That model is compared with building new companies in [AI roll-up vs venture studio vs AI holding company](https://frontmark.ai/insights/ai-roll-up-vs-venture-studio-vs-ai-holding-company). ## What are the open questions? The thesis is persuasive. Six parts of it are still to be shown. **Does the AI do the work?** Emergence Capital warned in its playbook of "Mirage PMF": revenue growth that hides a lack of real automation. A company can sell outcomes and deliver them with people doing the work out of the customer's sight. It will grow, and it will have the economics of an ordinary services firm. **Does the work get finished?** A company that sells outcomes is paid for completed jobs. The research section of the [State of AI Report 2026](https://stateof.ai/2026-explore/research), published by Air Street Capital in October 2026, reports that on OSWorld 2.0 the best model earned 77.7% partial credit and fully completed 44.3% of computer workflows. The distance between those two numbers is the part a service business has to cover with people, checks or rework. **Are these venture-scale companies?** [Euclid Ventures](https://insights.euclid.vc/p/service-level-disagreement-ai-services-versus-software), writing on 16 April 2026, is cautious on scope, market size and venture fit. It separates services used as a way into a market from businesses that are services companies throughout. **Will the margins hold?** The margin figures that appear in these essays are targets and expectations. The sources reviewed for this article contain no audited results that confirm them. **Who is accountable?** [Gartner](https://www.gartner.com/en/articles/ai-agent-reality-check) predicted on 2 October 2026 that 40% of agentic AI projects will be cancelled by the end of 2027, and said human accountability cannot be removed. A company that sells the work is the accountable party by construction. It needs people who answer for the result. **Who knows the domain?** In Y Combinator's Summer 2026 list, Tom Blomfield's Company Brain request says the blocker to automating companies is now domain knowledge and no longer the models. Selling the work in insurance or audit requires knowing how that work is judged by the people who buy it. ## What does this mean for someone starting a company? Choose work with a clear definition of done. If neither you nor the customer can say when the job is finished and correct, an agent cannot be held to it either. Measure the share of the work that no longer needs a person, and measure it honestly. That number, more than revenue, says whether the company is what it claims to be. Keep people on the decisions that need judgment. The sources that argue hardest for selling the work describe a gradual move from rule-following tasks towards judgment. None of them describes removing people on day one. Decide who owns the company that does the work. The essays mostly leave this open. A services business built on AI can be founder-owned and venture-funded, bought by a roll-up, or built and kept by a long-term owner. Each produces a different company over ten years. ## Where Frontmark stands That a company can do the work for its customer is common ground by now. Frontmark's position is on the two questions the essays leave open: how the work gets done, and who owns the company that does it. On the first, [agents](https://frontmark.ai/agents) do the work and one accountable operator answers for the result, with domain knowledge and access from the [collective](https://frontmark.ai/collective). On the second, Frontmark is built to keep the company it builds. The model is set out on the [approach](https://frontmark.ai/approach) page. ## Common questions ### What does sell the work, not the software mean? It means an AI company delivers a completed job and charges for the result, instead of selling a software licence to people who then do the job themselves. The buyer gets a finished outcome, such as closed books or a placed insurance policy, and the provider decides how much of the work is done by AI. ### What is the difference between a copilot and an autopilot in AI? The terms come from a Sequoia essay published 5 March 2026. A copilot is a tool sold to a professional, who remains responsible for the job. An autopilot sells the completed work to the customer who needs it done. The copilot competes for software budgets, while the autopilot competes for labour and outsourcing budgets. ### What is an AI-native services company? An AI-native services company does the work of a service firm, such as accounting, compliance or insurance brokerage, with AI carrying out most of the tasks from the first day. It sells the outcome to the customer. Y Combinator's Summer 2026 Requests for Startups asked for companies of this kind. ### What is a full-stack AI company? A full-stack AI company does the whole job for the customer instead of supplying one tool in the chain. Y Combinator partners said on the Lightcone podcast on 18 September 2026 that the share of accepted companies doing full-stack, end-to-end work rose from about 10% to over 25% of the batch. ### What are the risks of an AI-native services business? The main risk is growth that hides a lack of automation, which Emergence Capital named in January 2025. If people are still doing most of the work, the business has the economics of an ordinary services firm. Other risks are unreliable task completion, uncertain margins and the need for real credibility in the customer's industry. --- # The one-person billion-dollar company: what cases show URL: https://frontmark.ai/insights/one-person-billion-dollar-company Category: Thesis Published: 2026-11-16 Updated: 2026-11-16 Author: Frontmark Short answer: A one-person billion-dollar company is a business that reaches a billion dollars in value or revenue with a single founder and AI doing work that once needed staff. As of November 2026, reported cases come close without being clean examples, and the numbers are self-reported. The trend towards very small teams is real. Key points: - The one-person billion-dollar company moved from prediction to news coverage during 2026. - The most cited case, Medvi, was reported by the New York Times with two people and company-supplied sales figures. - Y Combinator partners said in September 2026 that solo founders rose from about 5% to 18 or 19% of a batch. - A single person still has limited attention for judgment, limited access to markets and sole accountability. - Headcount measures an input, and ownership, accountability and durability say more about a company. A one-person billion-dollar company is a business that reaches a billion dollars in value or revenue with a single founder and AI doing work that once needed staff. As of November 2026, reported cases come close without being clean examples, and the numbers are self-reported. The trend towards very small teams is real. The phrase has been repeated often enough to work as a slogan. As a claim it can be checked: who said it would happen, what has been reported, and what the reported cases leave out. The answers below are as of 16 November 2026. ## What is a one-person billion-dollar company? The phrase describes a company with one founder, no conventional staff and a billion dollars of something. It is also called a one-person unicorn. Two parts of that definition are loose. The first is the billion. A unicorn traditionally means a private company valued at a billion dollars by investors. Some of the cases discussed in 2026 are about revenue. Revenue and valuation are different measures, and a company can have one without the other. The second is the one person. A company with no employees may still depend on contractors, agencies, platforms and suppliers. Those people do real work that a headcount leaves out. Before accepting any example, check what exactly is being counted. ## Who predicted the one-person billion-dollar company? Two predictions are cited most often. The [New York Times](https://www.nytimes.com/2026/04/02/technology/ai-billion-dollar-company-medvi.html), in an article by Erin Griffith published 2 April 2026, quotes a 2024 prediction by Sam Altman of a one-person business worth a billion dollars. A headline in [The Times](https://www.thetimes.com/business/companies-markets/article/anthropic-boss-predicts-first-one-person-1bn-company-by-years-end-86skzsf0s) of London on 7 May 2026 says Dario Amodei predicts the first one-person $1bn company by the end of the year. That is the headline's wording and not a quotation of his. By spring 2026 the idea was familiar enough to be treated as a standard talking point. In [Good news: AI Will Eat Application Software](https://a16z.com/good-news-ai-will-eat-application-software), published 2 March 2026, Alex Immerman and Santiago Rodriguez of a16z list the single-person billion-dollar company among the stock bear arguments against software incumbents. Their own position is that moats survive. ## Has anyone built one yet? The most cited case is Medvi. According to the New York Times article of 2 April 2026, Medvi is a telehealth company built by one founder with AI tools, with his brother later joining as the only employee. The Times reported that the company had $401 million in sales in 2025 and was on track for $1.8 billion in 2026. That case comes with three qualifications. The figures are sales, and they are the company's own, as reported by the Times. It is a two-person company on the Times's account, which is close to the prediction and not identical to it. And the case is about revenue. Whether it meets the original prediction depends on which definition you use. A second kind of example comes from the products that sell this capability. Their founders report large run rates with one human and no employees, and those figures are their own. The claim behind such products is examined in [can a company run itself](https://frontmark.ai/insights/can-a-company-run-itself). In November 2026, then, there are reported cases with very large numbers and very few people. The sources we reviewed contain no clean, independently verified one-person company at a billion dollars. ## Are small teams really becoming more common? Yes, on the signals available. Y Combinator partners said on the [Lightcone podcast](https://podscripts.co/podcasts/y-combinator-startup-podcast/the-state-of-startups-in-2026) episode of 18 September 2026 that solo founders went from about 5% of accepted companies a year earlier to 18 or 19% of the batch. That is the partners' own account of their intake. The accelerator's requests point the same way. Its [Fall 2026 Requests for Startups](https://web.archive.org/web/20260901051221/https://www.ycombinator.com/rfs), archived 1 September 2026, include a request called A Cloud for Small Software, for tools built by agents for one person or a small team. The solo founder has gone from a small minority to a meaningful share of new companies, and that shift is better supported than any single billion-dollar case. It also means lean is now expected. A small team was a distinguishing feature two years ago. In late 2026 it is closer to a starting assumption. ## What can a solo founder do with AI agents? A great deal of the volume. Agents can write and test code, research a market, draft and send communications, handle routine support and keep operations moving. Work that once needed a first ten hires can be started by one person with the right tools. The constraint moves to supervision. A summary by [NYU Shanghai](https://rits.shanghai.nyu.edu/ai/anthropics-2026-agentic-coding-trends-report-from-assistants-to-agent-teams/), dated 17 April 2026, of Anthropic's 2026 Agentic Coding Trends Report says developers use AI in about 60% of their work and can fully delegate only 0 to 20% of tasks. The summary is a secondary source for the report's figures. If those figures hold more widely, most agent work still needs a person nearby. Agents multiply what one person can produce. The hours that person has for checking and deciding stay the same. Independent measurements support the caution. The [State of AI Report 2026](https://stateof.ai/2026-explore/research), published by Air Street Capital in October 2026, reports that the best model fully completed 44.3% of computer workflows on OSWorld 2.0. Someone has to catch the rest. More detail is in [what AI agents can and cannot do](https://frontmark.ai/insights/what-ai-agents-can-and-cannot-do). ## What can one person still not supply? **Enough judgment for everything at once.** A company produces more decisions as it grows. One person has a fixed amount of attention, and decisions made without attention are guesses. **Access to more than one world.** Markets open through trust. A founder usually has real standing in one industry, sometimes two. Tom Blomfield's Company Brain request in Y Combinator's [Summer 2026 list](https://web.archive.org/web/20260518221122/https://www.ycombinator.com/rfs), archived 18 May 2026, says the blocker to automating companies is now domain knowledge and no longer the models. Domain knowledge lives in people. **Shared accountability.** [Bloomberg Law](https://news.bloomberglaw.com/business-and-practice/zero-person-companies-operated-by-ai-would-create-vat-tensions) reported on 2 April 2026 that no jurisdiction lets a company exist without a traceable natural person. In a one-person company, that person carries all of it. **Continuity.** A company that depends entirely on one individual stops when that individual does. This is an old problem for small firms, and agents leave it in place. **Scarce deployment skill.** [TechCrunch](https://techcrunch.com/2026/07/30/forward-deployed-engineers-are-the-ai-industrys-latest-talent-obsession) reported on 30 July 2026, citing a Christian & Timbers study, that about 2,000 US engineers were judged able to deliver enterprise AI returns consistently. The ability to put AI to work in a real business is rare, and a solo founder either has it or has to find it. ## Is headcount the right thing to measure? Probably not. Headcount is an input. A company with one employee and a company with fifty can be equally strong or equally fragile. Four other measures say more about a company: - who makes the decisions that matter, and how many of those one person can make well; - who is accountable to customers; - what the company owns that a competitor with the same tools could not rebuild in a month; - what happens to it in year five. The a16z essay makes a related point from the other side: tools that make building cheap leave in place the advantages that take time to earn. The solo narrative also leaves something out. The same agents that let one person do the work of ten let a small group of experienced people do the work of many more, with the judgment, access and accountability spread across several of them. A group like that is a different thing to build from a one-person company, and it is designed to outlast any one member. ## Where Frontmark stands Frontmark is designed as the group described in the last section, built as an institution. [Agents](https://frontmark.ai/agents) do most of the work. The judgment, access and accountability that a solo founder carries alone are spread across a [collective](https://frontmark.ai/collective) of entrepreneurs, founders, operators and key people. It is built to keep its companies, and so that none of them rests on one person. ## Common questions ### What is a one-person billion-dollar company? It is a company that reaches a billion dollars in valuation or revenue with one founder and no conventional staff, with AI tools and agents doing work that used to need employees. The phrase is also written as one-person unicorn. Definitions vary on whether the billion refers to value or sales, and on who counts as staff. ### Has anyone built a one-person billion-dollar company? As of November 2026 there is no clean, independently verified example in the sources behind this article. The New York Times reported in April 2026 that Medvi, a telehealth company with one founder and his brother as its only employee, reported $401 million in 2025 sales and was on track for $1.8 billion in 2026. ### Who predicted the one-person billion-dollar company? The New York Times, in an article of 2 April 2026, quotes a 2024 prediction by Sam Altman of a one-person business worth a billion dollars. A headline in The Times of London on 7 May 2026 says Dario Amodei predicts the first one-person billion-dollar company by the end of the year. ### Can a solo founder build a large company with AI agents? A solo founder can now build and operate far more than before, because agents can write code, do research and handle routine operations. The limits are the founder's own attention for decisions, access to customers and partners, and sole responsibility for everything the company does. Those limits grow as the company grows. ### Are startup teams getting smaller because of AI? The available signals say yes. Y Combinator partners said on the Lightcone podcast of 18 September 2026 that solo founders rose from about 5% of accepted companies a year earlier to 18 or 19% of the batch. Y Combinator's Fall 2026 requests also asked for tools built by agents for one person or a small team. ### What can one person not do, even with AI? One person cannot be in several rooms, hold deep knowledge of several industries, or share accountability with anyone. A summary of an Anthropic report says developers can fully delegate only 0 to 20% of tasks, so most agent work still needs supervision, and one person has a fixed amount of that to give. --- # Venture studio vs accelerator vs incubator: compared URL: https://frontmark.ai/insights/venture-studio-vs-accelerator-vs-incubator Category: Models Published: 2026-11-16 Updated: 2026-11-16 Author: Frontmark Short answer: A venture studio creates companies from its own ideas and builds them with an in-house team. An incubator supports founders while they develop an early idea into a company. An accelerator runs a short, fixed programme that helps existing startups grow faster. Venture capital invests money in companies founders have already started. Key points: - The clearest dividing line is who originates the idea: a venture studio does, while incubators, accelerators and investors support ideas that founders bring. - Accelerators are short, cohort-based programmes for startups that already exist, and incubators offer longer, looser support at the earliest stage. - Equity generally rises with the amount of work the organisation contributes, so studios typically hold the largest stakes. - The right choice for a founder depends on whether they already have an idea, a team and evidence, and on how much hands-on help they want. A venture studio creates companies from its own ideas and builds them with an in-house team. An incubator supports founders while they develop an early idea into a company. An accelerator runs a short, fixed programme that helps existing startups grow faster. Venture capital invests money in companies founders have already started. All four help new companies come into being, which is why they are often confused. They differ in who has the idea, when they get involved, how long they stay, what they contribute and what they ask for in return. ## What is a venture studio? A venture studio is an organisation that starts companies itself. It generates and tests ideas, builds first products with shared designers, engineers and operators, recruits founders or leaders, and provides the earliest capital. Because it does so much of the founding work, it holds a significant ownership stake. The model is also known as a startup studio or venture builder. A full explanation is in [what is a venture studio](https://frontmark.ai/insights/what-is-a-venture-studio). ## What is an incubator? An incubator supports people at the very beginning, often before a company formally exists. It typically offers workspace, mentoring, basic services and a community of other founders. The founders bring the idea and do the building. Incubators are frequently run by universities, governments, non-profits and corporations. Timeframes tend to be flexible, often a year or longer. Many take no equity or a small amount, and some charge rent or fees instead. ## What is an accelerator? An accelerator runs a structured programme for startups that already exist. Companies apply, a cohort is selected, and the group works through a fixed period that commonly lasts around three months. The programme usually ends with a demo day in front of investors. Accelerators typically provide a small investment, intensive mentoring and access to a network of investors and alumni. In exchange they commonly take a single-digit percentage of equity, though terms differ from one programme to another. Y Combinator is the best-known example of the format. ## What is venture capital? Venture capital is investment in young companies with high growth potential. A venture capital firm raises a fund from outside investors and invests it across a portfolio of companies, usually taking minority stakes and often a board seat. The firm selects and supports companies, and leaves starting and operating them to founders. ## How do venture studios, accelerators, incubators and venture capital compare? | Compared on | Venture studio | Incubator | Accelerator | Venture capital | | --- | --- | --- | --- | --- | | Who originates the idea | Usually the studio, sometimes with a founder | The founder | The founder | The founder | | Stage | Before the company exists, through first product and early growth | Idea and earliest stage | Early stage, usually with a team and a product | Seed through later stages | | Duration | Often several years of involvement | Open-ended, often a year or more | Fixed term, commonly around three months | Many years as a shareholder | | Capital | Funds the early build from its own resources | Often little or none | Small standard investment | Larger investments, sized to the round | | Equity | Largest stake, often a significant minority or more | Often none or a small stake | Commonly a single-digit percentage | Minority stake negotiated each round | | Level of involvement | Hands-on operating work inside the company | Light: space, mentoring, services | Intensive but brief: mentoring and structure | Advisory: board, introductions, guidance | | What remains afterwards | A company the studio co-founded and continues to part-own | The founder's company, with a network | The founder's company, with investors and alumni ties | A shareholder with ongoing rights | The figures in the equity row are general patterns. Individual organisations set their own terms, and anyone considering one should ask for specifics. ## Venture studio vs accelerator: what is the main difference? The main difference is where the company comes from. An accelerator admits companies that founders have already formed and helps them move faster for a short period. A studio forms the company in the first place. Four other differences follow from that origin. - **Depth of work.** An accelerator advises and connects. A studio's own staff write code, design the product and find the first customers. - **Time.** An accelerator's programme is measured in months. A studio's involvement is measured in years. - **Ownership.** An accelerator's stake is small. A studio's stake is larger, reflecting the idea, the labour and the early capital it contributes. - **Scale.** An accelerator can take many companies in each cohort. A studio builds far fewer, because each one consumes its team's time. ## Venture studio vs incubator: how do they differ? Both operate at the earliest stage, which is the source of the confusion. An incubator gives founders room and support while they work on their own idea. A studio has the idea, does the work and co-owns the result. A founder in an incubator remains the originator and principal owner. A leader in a studio-built company joins something the studio began, and holds a share alongside it. ## Accelerator vs incubator: which comes first? An incubator usually comes first. It suits someone exploring an idea who needs time, guidance and low costs. An accelerator suits a company that has a team and a working product and wants to move faster over a few months. | Compared on | Incubator | Accelerator | | --- | --- | --- | | Starting point | An idea or very early project | An existing startup | | Pace | Self-directed | Structured and fast | | Format | Rolling admission | Cohorts with fixed dates | | End point | When the company outgrows it | Demo day | ## Venture studio vs venture capital: is a studio a kind of investor? A studio invests, but its main activity is building. A venture capital firm finds promising companies, funds them and advises them. A studio makes the companies that an investor might later fund. The two are frequently connected. Some studios run a fund alongside the building operation, and studio-built companies often raise venture capital once there is evidence to show. The financial side is covered in [how venture studios make money](https://frontmark.ai/insights/how-venture-studios-make-money). ## Which should a founder choose? Choose by what you already have and what you lack. **Choose an incubator if** you have an early idea, limited resources and time to develop it. It suits first-time founders, researchers and students who want a low-cost setting and guidance. **Choose an accelerator if** you have a team, a product and some early evidence, and you want pace, structure and investor access. You keep most of your company and gain a network. **Choose venture capital if** your company has evidence that supports a larger round and you need money and experienced board-level support more than hands-on help. **Choose a venture studio if** you are an experienced operator who wants to lead a company and you value starting with a tested idea, a building team and capital. The trade is a smaller ownership share in exchange for a stronger starting position and lower personal risk. Whichever you consider, ask four questions: 1. What exactly will this organisation do, and who will do it? 2. What equity or fees does it take, and on what terms? 3. How long will it stay involved, and what happens when its involvement ends? 4. Can I speak with founders who have been through it, including ones whose companies did not work? Companies often pass through more than one of these routes, and each adds dilution. At each step, weigh the contribution against the ownership given up. ## Where Frontmark stands Frontmark is built to do the originating and building work of a studio and to keep long-term ownership, which none of the four models does in its pure form. It runs no cohort and no programme, and offers no standard terms. For how ownership structures differ, see [holding company vs venture studio](https://frontmark.ai/insights/holding-company-vs-venture-studio). The comparison with studios is on the [venture studio](https://frontmark.ai/venture-studio) page. ## Common questions ### What is the difference between a venture studio and an accelerator? A venture studio starts companies itself, supplying the idea, an early team and first capital, and stays involved for years. An accelerator admits startups that founders have already begun and runs them through a fixed programme, commonly lasting around three months, with mentoring, a small investment and investor introductions. Studios build, while accelerators speed up what exists. ### What is the difference between a venture studio and an incubator? An incubator gives early-stage founders space, mentoring and services while they develop their own idea, usually without doing the building work for them. A venture studio develops the idea internally and builds the company with its own team. The studio is a co-founder with a large ownership stake, while an incubator is a supportive environment. ### What is the difference between an accelerator and an incubator? An incubator supports very early ideas over an open-ended period, often a year or more, and may take little or no equity. An accelerator works with startups that already have a team and usually a product, over a short fixed term, in cohorts, and typically invests a small amount for a single-digit equity stake. ### Is a venture studio better than an accelerator? Neither is better in general, because they serve different situations. An accelerator suits a founding team that already has a company and wants momentum, a network and early investment. A venture studio suits an experienced operator who wants to lead a company and values an idea, a team and capital from the start, in exchange for a smaller ownership share. ### How is a venture studio different from venture capital? Venture capital firms invest in companies that already exist, usually for a minority stake, and support them as board members and advisers. A venture studio creates the companies and does early operating work itself, holding a larger stake in return. The two often work in sequence, with studio-built companies raising venture capital once they show evidence. ### Can a startup use more than one of these? Yes. The models often follow one another. A company might begin in a university incubator, join an accelerator once it has a product, and then raise venture capital. A studio-built company may also enter an accelerator or raise from venture investors. Each stage adds dilution, so founders should weigh what each one contributes. --- # What can AI agents do, and can they run a company? URL: https://frontmark.ai/insights/what-ai-agents-can-and-cannot-do Category: Practice Published: 2026-11-16 Updated: 2026-11-16 Author: Frontmark Short answer: AI agents in late 2026 can work for hours, coordinate in teams and complete a growing share of well-defined digital tasks. They still fail to finish many long workflows, and they cannot hold legal or commercial accountability. They can do much of the work inside a company, but people still have to decide and answer for it. Key points: - Agents now work for hours and in coordinated teams, according to lab reports and independent evaluators in 2026. - The State of AI Report 2026 shows a wide gap between partial credit and fully completed workflows on current benchmarks. - Gartner said in October 2026 that most agents are not reliable enough for autonomous use in business-critical settings. - Accountability cannot be delegated to an agent, so a business needs named people who decide and answer for the result. - The working division of labour is that agents carry the volume and people hold judgment. AI agents in late 2026 can work for hours, coordinate in teams and complete a growing share of well-defined digital tasks. They still fail to finish many long workflows, and they cannot hold legal or commercial accountability. They can do much of the work inside a company, but people still have to decide and answer for it. This article is dated on purpose. Agent capability changes fast enough that a claim made in spring can be wrong by autumn. Everything below describes the evidence as it stands on 16 November 2026, with the source and date given for each figure. ## What can AI agents do today? An AI agent is a model that is given a goal, a set of tools and room to act over many steps. It plans, uses software, checks its own output and continues until it stops or is stopped. For how groups of them are organised, see [what is an agent swarm](https://frontmark.ai/insights/what-is-an-agent-swarm). Four things changed in 2026. Agents work for longer. A summary by [NYU Shanghai](https://rits.shanghai.nyu.edu/ai/anthropics-2026-agentic-coding-trends-report-from-assistants-to-agent-teams/), dated 17 April 2026, of Anthropic's 2026 Agentic Coding Trends Report describes tasks that run for hours to days. The summary is a secondary source for the report. Agents work together. The same summary describes single agents becoming coordinated teams. Y Combinator's [Fall 2026 Requests for Startups](https://web.archive.org/web/20260901051221/https://www.ycombinator.com/rfs), archived 1 September 2026, includes a request called Multiplayer AI, for teams of people who share live agent sessions as agents run for hours, days and weeks. Agents complete more real work. The [Remote Labor Index](https://www.remotelabor.ai/), a project of the Center for AI Safety and Scale AI, showed 41.25% full automation of real remote-work projects on its site when read on 16 November 2026, with a doubling pace of about 3.3 months. The original result from October 2025 is reported as 2.5%. The figure is the project's own reading on that date, and its current methodology is outside this article. Agents are gaining the means to transact. [Stripe's documentation](https://docs.stripe.com/agentic-commerce/protocols/acp) describes the Agentic Commerce Protocol as an open standard, created by Stripe, OpenAI and Meta, for agents completing purchases on a buyer's behalf. ## How reliable are AI agents at finishing a job? Less reliable than the capability list suggests. The clearest numbers come from the research section of the [State of AI Report 2026](https://stateof.ai/2026-explore/research), published by Air Street Capital in October 2026. On OSWorld 2.0, a test of computer workflows, the best model earned 77.7% partial credit and fully completed 44.3% of the workflows. On FrontierChallenge, the best result was 87.9 out of 100 on requirements, with 20.6% of tasks fully completed. Each pair has a high score for partial work and a much lower one for finished work. An agent can satisfy most of what a task asks and still not deliver the task. In a business, a job that is nearly done is often worth nothing until someone finds the missing part and fixes it. Anthropic has shown the same pattern in a small real setting. In [Project Vend phase two](https://www.anthropic.com/research/project-vend-2), an AI shopkeeper improved markedly with newer models. Anthropic's own conclusion is that the gap between capable and completely robust remains wide. ## How long can an AI agent work on its own? Long enough that length is no longer the main limit, and not yet with dependable results at the long end. [METR](https://metr.org/time-horizons/) tracks what it calls the 50% time horizon: the length of task an agent can complete at half reliability. Its time-horizons page, last updated 8 September 2026, flags measurements above 16 hours as unreliable with the current task suite. That definition carries two cautions. Half reliability means the agent fails as often as it succeeds at that length, which no business would accept for important work without checks. And the measuring tools themselves are running out of range, so claims about very long autonomous runs are hard to verify. An agent that can run for a day is useful. An agent that can be trusted with a day's work unsupervised is a stronger claim, and the evidence for it is thinner. ## What can AI agents not do yet? **Finish reliably without checks.** The completion figures above describe this directly. **Exercise judgment where the rules run out.** The State of AI Report 2026 says agents can do research engineering but score poorly on research judgment. Deciding which question is worth asking is a different skill from carrying out the work once it is asked. **Be accountable.** [Gartner](https://www.gartner.com/en/articles/ai-agent-reality-check) said on 2 October 2026 that most agents are not reliable enough for autonomous use in business-critical settings, and that human accountability cannot be removed. It predicts that 40% of agentic AI projects will be cancelled by the end of 2027. **Stand in for a legal person.** [Bloomberg Law](https://news.bloomberglaw.com/business-and-practice/zero-person-companies-operated-by-ai-would-create-vat-tensions) reported on 2 April 2026 that no jurisdiction lets a company exist without a traceable natural person. Whatever an agent does, a human being is answerable for it. **Supply access.** An agent can write a thousand good emails. It cannot be the person a buyer already trusts, and in many markets that trust decides who gets a meeting. ## Why do the numbers on AI agent reliability disagree? The figures 41.25%, 44.3%, 20.6% and 77.7% all describe agent performance in the same year. They measure different things. Some scores award partial credit and others count only fully completed tasks. Some test computer use, some test freelance-style projects, some test software engineering. Some are run by independent evaluators and some are published by the project that designed the test. Before relying on any agent benchmark, check three things: - whether the number counts finished work or partial work; - who ran the test; - how close the tasks are to the job you have in mind. All of them point the same way. Capability is rising quickly, and full completion trails behind partial credit. ## Can AI agents run a company? They can do a large part of a company's work. Running a company also includes deciding what the company should do, committing money, making promises to customers and being answerable when something goes wrong. On the evidence above, those remain human functions. Agents complete under half of long computer workflows in the State of AI figures. Evaluators say accountability cannot be removed. The law requires a person. There are products that promise more than this, and they are examined in [can a company run itself](https://frontmark.ai/insights/can-a-company-run-itself). The closest thing to a consensus in late 2026 is narrower: agents can carry the volume of the work if people hold the decisions. ## How should a business divide work between agents and people? The NYU Shanghai summary of Anthropic's report gives two figures side by side. Developers use AI in about 60% of their work, and can fully delegate only 0 to 20% of tasks. Most of the value comes from supervised work, where a person stays close to what the agent is doing. The working rule that follows: give agents work that is high in volume, digital, and checkable against a clear definition of done. Keep with people the decisions that set direction, commit resources or bind the company to someone else. The scarce input is human. The State of AI Report 2026 describes the gap between people who get value from AI and people who do not as a human skill issue and not a technology issue. [TechCrunch](https://techcrunch.com/2026/07/30/forward-deployed-engineers-are-the-ai-industrys-latest-talent-obsession) reported on 30 July 2026, citing a Christian & Timbers study, that about 2,000 US engineers were judged able to deliver enterprise AI returns consistently. The practical limit on agents in a business is therefore often the supply of people who know how to direct them, check them and take responsibility for the result. A fuller account of that way of working is in [what is agentic company building](https://frontmark.ai/insights/what-is-agentic-company-building). ## Where Frontmark stands Frontmark's division of labour follows the evidence above. Agents do the volume: research, demand tests, product and their own tools. One accountable operator makes each decision that sets direction, commits resources or binds a company, and answers for it. The same evidence is summarised on the [agents](https://frontmark.ai/agents) page. The access agents cannot supply comes from the [collective](https://frontmark.ai/collective). ## Common questions ### Can AI agents run a company? Agents can carry out much of the daily work of a company, including research, coding, support and routine operations. As of November 2026 they still need supervision to finish long tasks reliably, and they cannot be legally accountable. A company can run largely on agents, but people still have to make the judgment calls and answer for the outcome. ### What can AI agents do in 2026? Agents can work on a task for hours, use computers and software tools, write and test code, carry out research, and coordinate with other agents in teams. They are strongest on well-defined digital work with a clear definition of done. Capability is rising quickly, which is why any list like this should carry a date. ### How reliable are AI agents? Less reliable than their headline scores suggest. The State of AI Report 2026 says the best model earned 77.7% partial credit on OSWorld 2.0 but fully completed 44.3% of computer workflows. Agents often get most of a job right and still leave it unfinished, which matters when the job is the product. ### How long can an AI agent work on its own? Reports in 2026 describe tasks that run for hours to days. METR tracks a measure called the 50% time horizon, which is task length at half reliability, and flags its own measurements above 16 hours as unreliable with the current task suite. Long duration and dependable completion are separate things. ### What can AI agents not do? Agents cannot be accountable for a decision, hold a legal duty or be the person a customer or regulator holds to account. They score poorly on research judgment, according to the State of AI Report 2026. They also cannot produce trust and access in a market, which come from people who are already known in it. ### Why do so many AI agent projects fail? Gartner predicted in October 2026 that 40% of agentic AI projects will be cancelled by the end of 2027, and said success depends on reliability more than autonomy. Projects tend to fail when an agent is given business-critical work without checks, without a clear definition of done and without a person who owns the result. --- # What is a holding company? Definition, structure, examples URL: https://frontmark.ai/insights/what-is-a-holding-company Category: Definitions Published: 2026-11-16 Updated: 2026-11-16 Author: Frontmark Short answer: A holding company is a business entity whose main purpose is to own shares in other companies rather than to make products or deliver services itself. It controls or influences those companies through ownership, sets direction, allocates capital between them, and earns from their dividends, fees and growth in value. Key points: - A holding company exists to own other companies; the companies it owns do the operating. - The typical structure is a parent entity above separate subsidiaries, each with its own customers, team and accounts. - A holding company earns through dividends, fees, interest and the rising value of what it owns. - The main advantages are separation of risk, central capital allocation and the ability to own for the long term. - A holding company describes ownership, so it says nothing by itself about how the owned businesses were created. A holding company is a business entity whose main purpose is to own shares in other companies rather than to make products or deliver services itself. It controls or influences those companies through ownership, sets direction, allocates capital between them, and earns from their dividends, fees and growth in value. This guide is for builders and operators who want to understand the concept. It describes general principles. Legal form, liability and tax treatment vary by jurisdiction, and any real structure should be designed with professional advice. ## What is the definition of a holding company? A holding company is defined by what it owns. It holds equity in other companies, which are called subsidiaries when it controls them. It may own all of a subsidiary, a majority, or a smaller stake that still gives it influence. Two variants are commonly described: - **A pure holding company** does nothing but own. It has few employees and no customers of its own. - **A mixed holding company** owns subsidiaries and also runs some operations itself. The shorthand "holdco" means the same thing. The companies beneath it are often called operating companies, or "opcos". ## How is a holding company structured? A typical holding company structure has two levels. At the top sits the parent entity, owned by its shareholders. Beneath it sit the subsidiaries, each a separate legal entity with its own management, accounts, contracts and obligations. The parent exercises control through its shareholding. In practice that means appointing or approving directors, agreeing budgets and strategy, deciding how profits are used, and choosing which businesses to fund, buy, merge or sell. Larger groups add layers. An intermediate holding company may sit between the top parent and a cluster of related businesses. A separate entity may hold shared assets such as intellectual property or property, which the operating companies then use. More layers allow more precision and bring more administration. How much the parent involves itself is a choice. Some holding companies stay at arm's length and act mainly as owners. Others provide shared functions such as finance, legal, hiring or technology to the companies they own. ## What is the difference between a holding company and a parent company? A parent company is any company that controls another company. A holding company is a parent whose main activity is that ownership. A manufacturer that makes its own products and also owns a logistics subsidiary is a parent company, but it would usually be described as an operating company with a subsidiary. If the same group reorganised so that a new top entity owned both the manufacturer and the logistics business, and did nothing else, that top entity would be a holding company. Alphabet is a well-known case of this kind of reorganisation: a holding company placed above Google and the group's other businesses. ## What is the difference between a holding company and an operating company? The two differ in what each entity spends its time doing. | Compared on | Holding company | Operating company | |---|---|---| | **Main activity** | Owning shares in other companies | Making products or delivering services | | **Customers** | Usually none directly | The market it serves | | **Team** | Small; focused on governance, finance and capital allocation | The people who build, sell and support | | **Income** | Dividends, fees, interest, gains in value | Revenue from customers | | **Key decisions** | Which businesses to own, fund and lead | Product, pricing, hiring, sales | | **Risk it carries** | The value of its holdings | The commercial risk of its own trade | In a group, the operating company concentrates on its market and the holding company concentrates on the group. ## How does a holding company make money? A holding company has no customers in the usual sense, so its income comes from what it owns. The common sources are: 1. **Dividends.** Profitable subsidiaries distribute part of their earnings to the parent. 2. **Fees.** The parent may charge subsidiaries for management, shared services or the use of assets it holds, such as brands or software. 3. **Interest.** The parent may lend to subsidiaries and receive interest. 4. **Gains in value.** Businesses it owns become more valuable over time, and that value is realised if a stake is sold. Behind all four sits capital allocation. Cash produced by a mature business can be directed to a younger one with better prospects. A holding company that does this well can earn more from the same set of businesses than they would earn separately. One that does it badly adds a layer of cost. How these flows are treated for tax and accounting purposes differs between countries and structures. ## What are the advantages of a holding company? The advantages come from the structure itself. - **Separation of risk.** Each subsidiary is a separate legal entity, so in principle a failure or claim in one business is contained there. How far that protection holds depends on local law and on how the group is actually run. - **Central capital allocation.** The parent can move resources towards the best opportunity across the group. - **Focus.** Each operating company can concentrate on one market with its own leadership. - **Flexibility in ownership.** Individual businesses can take on partners, grant equity to their teams, or be sold without disturbing the rest. - **Shared capability.** Functions and assets held at the group level can serve several businesses. - **A long horizon.** A holding company has no built-in date by which it must sell, so it can own for as long as ownership makes sense. ## What are the drawbacks? Every entity needs its own accounts, filings and governance, so administration grows with the group. A parent that stands too far from operations can miss problems. A parent that stands too close can slow the companies it owns. A holding company can also be marked down as a whole. Outside observers sometimes value a group of unrelated businesses at less than the sum of its parts, on the view that the parent adds cost without adding benefit. A parent answers that criticism only by demonstrably making its companies better. ## What are some holding company examples? - **Berkshire Hathaway** owns a broad set of operating businesses and significant stakes in public companies, with operating decisions left largely to each business. - **Alphabet** is the holding company above Google and the group's other businesses. - **Private and family holding companies** are far more numerous than the famous names. A founder who owns two or three businesses through a single entity is using the same idea at small scale. ## How does a holding company differ from a fund or a venture studio? A holding company is often confused with an investment fund. A fund typically raises money from outside investors and returns it within a set period, which makes selling its positions part of the design. A holding company generally owns with its own capital and can keep its companies indefinitely. It is also distinct from a venture studio. A studio is defined by creating companies; a holding company is defined by owning them. The two can be joined, with one organisation building businesses and then keeping them. That combination is covered in [holding company vs venture studio](https://frontmark.ai/insights/holding-company-vs-venture-studio), and the studio model is explained in [what is a venture studio](https://frontmark.ai/insights/what-is-a-venture-studio). ## When does a holding company make sense for a builder? For a founder with one product, a single company is usually enough. The case for a holding structure strengthens when: - you run, or plan to run, several distinct businesses; - you want valuable shared assets held apart from day-to-day commercial risk; - different businesses need different partners, teams or equity arrangements; - you intend to own what you build for a long time and to reinvest across it. The structure serves an ownership strategy. Decide the strategy first, then take advice on the form that serves it. ## Where Frontmark stands The holding structure is one half of Frontmark's model. Frontmark is built to keep the companies it builds, which keeps it answerable for each one after launch and lets tools and relationships pass to the next. The other half is construction, defined in [what is company building](https://frontmark.ai/insights/what-is-company-building). The model is set out on the [approach](https://frontmark.ai/approach) page. ## Common questions ### What is a holding company in simple terms? A holding company is a company that owns other companies. It usually has no products or customers of its own. Its work is to hold shares, oversee the businesses it owns, appoint or support their leadership, and decide where the group's money should go. The businesses underneath it carry out the day-to-day operations. ### What is the difference between a holding company and a parent company? A holding company that controls its subsidiaries is a parent company, but only some parents are holding companies. A parent company is any company that controls another. If the parent also runs a business of its own, it is an operating parent. If its main activity is owning its subsidiaries, it is a holding company. ### What is the difference between a holding company and an operating company? An operating company makes products, delivers services, employs staff and deals with customers. A holding company owns operating companies and focuses on governance and capital allocation. In a group structure the two sit at different levels: the holding company at the top, the operating companies beneath it. ### How does a holding company make money? A holding company earns in several ways: dividends paid up from the companies it owns, fees for services or management it provides to them, interest on money it lends within the group, and gains when a business it owns rises in value or is sold. The mix depends on the group and on local rules. ### What are some examples of holding companies? Berkshire Hathaway is a widely cited example, owning a broad range of operating businesses under one parent. Alphabet is the holding company created to own Google and its other businesses. Many smaller, private holding companies exist as well, including family groups and founders who own several businesses through one entity. ### Do I need a holding company for my startup? Often not at the start. A single company with one product is usually simpler to run and to finance. A holding company becomes worth considering once there are several distinct businesses, assets worth separating, or a plan to own companies for a long time. The right answer depends on jurisdiction, so take legal and tax advice first. --- # What is a venture studio? The model explained URL: https://frontmark.ai/insights/what-is-a-venture-studio Category: Models Published: 2026-11-16 Updated: 2026-11-16 Author: Frontmark Short answer: A venture studio is an organisation that creates new companies itself. It develops ideas internally, tests them, builds the first product with a shared team, recruits founders or leaders, and provides early capital. In return it holds a significant ownership stake. It is also called a startup studio, venture builder or company builder. Key points: - A venture studio originates and builds companies itself, using shared teams, repeatable process and its own capital. - Startup studio, venture builder, company builder and startup factory are different names for broadly the same model. - Studios usually hold a larger ownership stake than investors or accelerators, because they contribute the idea, the early work and the first capital. - The model depends on selection discipline and focus, and its main limits are founder dilution, divided attention and the difficulty of repeating success. A venture studio is an organisation that creates new companies itself. It develops ideas internally, tests them, builds the first product with a shared team, recruits founders or leaders, and provides early capital. In return it holds a significant ownership stake. It is also called a startup studio, venture builder or company builder. The model sits between two more familiar ones. An investor backs companies other people start. A founder starts one company at a time. A venture studio starts companies repeatedly, on the view that starting them is a skill that improves with practice. ## What does a venture studio do? A studio does the work that normally falls to a founding team in a company's first months, and it does that work across several companies. It typically: - researches markets and generates venture ideas - tests those ideas against real customers before committing fully - builds an early product using its own designers, engineers and operators - recruits a founder, chief executive or founding team for each company - funds the earliest stage from its own capital - supplies shared functions such as finance, legal, hiring and go-to-market support - helps the company raise outside investment or reach revenue No single item on that list makes a studio. An agency can build a product, and an investor can write a cheque. A studio joins idea, team, build and capital under one roof and takes ownership in what results. ## Are startup studio, venture builder and company builder the same thing? Broadly, yes. The vocabulary grew up in different places at different times, and the terms overlap almost completely. - **Venture studio** is the term investors tend to use. - **Startup studio** is used interchangeably with it. - **Venture builder** and **company builder** put the weight on the building work and are common in corporate settings. - **Startup factory** emphasises repeatable production. - **Foundry** appears in some firm names with the same meaning. Idealab, founded in 1996, is widely cited as the earliest example of the model. The category names came later. Because the terms are loose, ask how a particular studio works before asking which word it prefers. ## How does the venture studio model work, step by step? Studios differ in detail, but most follow a recognisable sequence. 1. **Ideation and research.** The studio identifies openings: a customer problem, a change in technology, a shift in regulation or cost. Many ideas are considered and most are set aside. 2. **Validation.** Promising ideas are tested cheaply. The studio talks to prospective customers, builds prototypes and looks for evidence that someone will pay. 3. **Decision.** Ideas pass through explicit checkpoints. Weak evidence leads to stopping, which is a normal outcome in a studio. 4. **Team formation.** The studio recruits a founder or leadership team, or assigns one from its own bench of entrepreneurs in residence. 5. **Build.** The shared team develops the first product while the new company's own team takes shape. 6. **Incorporation and funding.** A separate company is formed, equity is allocated among the studio, the founders and an employee pool, and the studio provides initial capital. 7. **Spin-out and growth.** The company gains independence, hires its own staff and raises outside investment or grows on revenue. The studio's daily involvement usually reduces over time. 8. **Ownership.** The studio keeps its stake. Depending on its structure, it either seeks an eventual sale of that stake or holds it for the long term. The stages are often described as a funnel: many ideas enter, and few become companies. ## What are the common types of venture studio? | Type | How it works | Typical capital source | | --- | --- | --- | | Independent studio | Builds its own ideas across one or several sectors | Its own balance sheet, a studio fund or both | | Studio with a fund | Pairs the studio with an investment fund that backs its companies | Outside investors in the fund | | Corporate studio | Builds ventures for or with a large company, using its assets and customers | The parent corporation | | Thematic or specialist studio | Focuses on one field such as health, climate, fintech or AI | Varies | | Founder-led partnership studio | Co-founds companies with outside founders who bring the idea | Studio capital alongside the founder's commitment | | Holding-structure studio | Builds companies and retains ownership for the long term | Permanent capital and operating income | The categories overlap. A specialist studio can also run a fund, and a corporate studio can be thematic. Financial structure is the difference with the most consequence, because it sets how long a studio can stay involved. That is explained in [how venture studios make money](https://frontmark.ai/insights/how-venture-studios-make-money). ## How is a venture studio different from venture capital? The main difference is who starts the company. A venture capital firm selects companies that already exist and invests in them, generally for a minority stake, and supports them from the board. A venture studio originates the company and performs early operating work itself. As a result a studio usually holds more equity, works inside the company in its first phase, and builds fewer companies than a venture firm can back. In practice the two models work in sequence. Studio-built companies commonly raise venture capital once they have a product and early evidence. The full comparison, including accelerators and incubators, is in [venture studio vs accelerator vs incubator](https://frontmark.ai/insights/venture-studio-vs-accelerator-vs-incubator). ## What are the strengths of the venture studio model? - **Reduced early risk.** Ideas are tested before a company is formed, so obvious failures are stopped early and cheaply. - **Speed.** A team that has built several companies already knows incorporation, hiring and first-product development. - **Shared capability.** Design, engineering, legal and finance are available from the first day, which a two-person startup cannot afford. - **Accumulated knowledge.** Lessons, tools and relationships from one company carry over to the next. - **A route for operators.** Experienced people who want to lead a company, and lack an idea or early funding, get a starting point. ## What are the limitations of venture studios? - **Founder dilution.** A studio's stake leaves less for the founding team than they would hold in a company they began alone. Some later investors look closely at this. - **Divided attention.** A shared team serving several companies has to ration its time. A company can receive less focus than it needs at a critical moment. - **Founder motivation.** A leader recruited to an idea they did not originate may be less committed than someone who conceived it. Good studios address this through selection and meaningful ownership. - **Dependence on selection.** The model only works if the studio is disciplined about which ideas proceed. A studio that rarely stops anything spreads itself thin. - **Hard to repeat.** Building one good company is difficult. Building several in succession requires judgment that resists being turned into process. - **Capital intensity.** The studio pays for teams and experiments long before any return arrives. These limits belong to the model. Individual studios manage them with varying success, so judge the particular studio before the category. ## Where Frontmark stands Frontmark is built to do the core work of a venture studio: building companies. It differs on two of the points above. The shared team is largely [agents](https://frontmark.ai/agents), with operators making the decisions that need judgment. And Frontmark is built to keep long-term ownership at the stage where most studios reduce their involvement. The comparison is set out on the [venture studio](https://frontmark.ai/venture-studio) page, and the relationship between building and owning in [holding company vs venture studio](https://frontmark.ai/insights/holding-company-vs-venture-studio). ## Common questions ### What does a venture studio do? A venture studio generates and tests business ideas, builds early products with an in-house team, recruits founders or executives to lead each company, and funds the first stage of work. It then helps the company raise outside capital or grow on revenue. The studio repeats this process across several companies and holds equity in each one. ### What is the difference between a venture studio and venture capital? A venture capital firm invests money in companies that founders have already started, and usually takes a minority stake. A venture studio starts the companies itself, doing much of the early building work before outside investors are involved, and usually holds a larger stake as a result. Some studios also manage a fund, which combines both activities. ### Is a startup studio the same as a venture studio? Yes, in common usage. Startup studio, venture studio, venture builder, company builder and startup factory all describe an organisation that repeatedly creates companies using shared resources. Preferences vary by region and by firm. Differences between individual studios are larger than differences between the terms. ### How much equity does a venture studio take? Terms vary widely. Because a studio often supplies the idea, the first team and the earliest capital, it commonly holds a substantial share at formation, often cited as anywhere from a significant minority to a majority. The stake typically shrinks as founders, employees and later investors receive equity. Founders should ask for the specific terms. ### Do venture studios work with outside founders? Many do. Some studios develop every idea internally and then recruit a founder to lead it. Others invite experienced operators to join before an idea is fixed, often under an entrepreneur-in-residence arrangement, and shape the company together. A smaller number partner with founders who bring their own idea, which is closer to a co-founding relationship. ### What was the first venture studio? Idealab, founded by Bill Gross in 1996, is widely cited as the first venture studio. It set the basic pattern of generating ideas internally, building companies with shared resources and repeating the process. The model spread during the following decades, and the terms venture studio and startup studio came into common use later. --- # What is agentic company building? What AI agents can do URL: https://frontmark.ai/insights/what-is-agentic-company-building Category: Practice Published: 2026-11-16 Updated: 2026-11-16 Author: Frontmark Short answer: Agentic company building is the practice of using coordinated AI agents to do the work of starting a company: researching openings, testing demand with real customers, building product and running distribution. Operators make the decisions that need judgment and accountability, such as which customer to serve, what to commit and when to stop. Key points: - Agentic company building gives the work of starting a company to coordinated AI agents and keeps the accountable decisions with operators. - Agents are strong today at research, demand tests, code, repeatable distribution tasks and building their own internal tools. - Agents cannot be accountable for choosing the customer, committing capital, hiring, setting standards or deciding to stop. - The main risks are errors that compound without review, weak evidence that looks strong, unclear accountability and dependence on model providers. - A company built by agents still needs an owner who answers for it, which is why this is different from a tool that claims to run a company for you. Agentic company building is the practice of using coordinated AI agents to do the work of starting a company: researching openings, testing demand with real customers, building product and running distribution. Operators make the decisions that need judgment and accountability, such as which customer to serve, what to commit and when to stop. The phrase is recent and is used loosely. Some people mean a founder who uses AI assistants to move faster. Others mean a business that claims to run with no people at all. The meaning used here is narrower: agents do the work, people hold the responsibility, and the division between the two is stated in advance. ## What is agentic AI? Agentic AI is software built on AI models that pursues a goal across many steps instead of answering a single prompt. An agent plans, uses tools such as search, code execution and messaging, checks the result and decides what to do next. It keeps going until the task is complete or it needs a person to decide something. Two properties make this relevant to starting companies. Agents produce work, where an assistant produces suggestions for a person to act on. And agents can run in parallel, so many lines of work proceed at once. A coordinated group of them is usually called a swarm, which is explained in [what is an agent swarm](https://frontmark.ai/insights/what-is-an-agent-swarm). ## What does agentic company building mean in practice? [Company building](https://frontmark.ai/insights/what-is-company-building) is the discipline of turning an opportunity into an operating business. It includes the customer, the product, the route to market, the economics and the people accountable for them. The agentic version changes who does the work and leaves the definition of a company as it was. Agents take an opening and turn it into something testable: a defined customer, a problem, an offer and a way to reach the buyer. They put that offer in front of real people and record what happens. Where the evidence holds, they build the product and the first distribution. Where a task needs a new tool, they build it. At the points where evidence runs out and judgment begins, they stop and ask an operator. ## Which parts of company building can AI agents do well today? Agents are strongest where the work is well specified, the result can be checked and volume is useful. - **Research.** Reading a market, mapping who buys and what they use now, collecting public pricing and summarising what customers say. Agents cover more ground than a small team can, and they can keep the picture current. - **Demand tests.** Drafting offers, building landing pages, preparing outreach and measuring responses. An agent can prepare many variations of an offer in the time a team debates one. - **Code.** Writing, testing and shipping software, especially where the requirement is clear and automated tests exist to confirm the result. - **Distribution tasks.** The repeatable parts of reaching customers: content, listings, follow-ups, reporting. These are tasks with a known shape and a measurable outcome. - **Internal tooling.** Building the scripts, dashboards and pipelines the work needs. A tool an agent builds for one company can be reused by the next. All five come with a fast feedback signal. The test passes or fails, and the buyer replies or stays silent. Agents do their best work where that signal arrives quickly. ## What can AI agents not be accountable for? Accountability means a person can be asked to explain a decision and bear its consequences. Agents can inform the decisions below, and a person has to own each one. - **Customer choice.** Deciding whom the company serves shapes everything after it. Evidence narrows the options, and a person makes the choice. - **Commitments of capital.** Spending real money, signing contracts and taking on obligations bind the company. Someone with authority has to approve them. - **Hiring.** Bringing a person into a company is a commitment to that person. It requires judgment about character and fit, and a human who stands behind the offer. - **Standards.** What counts as good enough to ship, what the company will refuse to do and how it treats customers are choices of taste and principle. - **Stopping.** Ending a venture, or continuing it against weak evidence, is the hardest decision in company building. An agent optimising for its task has no reason to stop on its own. ## How do agents and operators divide the work? Agents do what can be checked against evidence. Operators decide what must be owned by a person. | Activity | What agents do | What operators decide | | --- | --- | --- | | Research | Gather sources, map the market, summarise customer evidence | Which openings are worth pursuing | | Demand tests | Draft offers, run tests with real buyers, report results | Whether the evidence is strong enough to form a company | | Product | Write, test and ship code; keep documentation current | What quality bar applies and what must not ship | | Distribution | Run repeatable outreach, content and reporting tasks | Which channels and claims fit the company | | Internal tools | Build and maintain tools the work needs | Which tools become shared infrastructure | | Capital | Prepare the case, costs and options | How much to commit, and to what | | People | Draft role definitions and screen against stated criteria | Who is hired and on what terms | | Continuing or stopping | Present the evidence for and against | Whether the venture continues | The table only works if requests for a decision are specific. A useful agent brings the evidence, the options and a recommendation, and records the answer so the next decision starts from it. ## Can AI agents run a company on their own? An autonomous company is a business whose operations are carried out mainly by AI agents, with little or no human staff. The idea is examined in [can a company run itself](https://frontmark.ai/insights/can-a-company-run-itself). As a description of tasks it is partly real. Agents can already handle much of the daily operating work of a small digital business. As a description of a company it is incomplete. A company holds money, signs agreements and owes duties to customers and employees, and those duties attach to people. A customer who is wronged needs someone to answer. An agent can be corrected or switched off. It cannot be held responsible in the way an owner or director can. A company with no accountable person is better understood as an unattended process than as a new kind of firm. ## What are the limits and risks of agentic company building? **Errors compound without review.** When one agent builds on another's output, an early mistake carries forward and grows. A wrong assumption in research becomes a wrong offer, then a wrong product. Review points between stages, by other agents and by people, are what contain this. **Evidence quality.** Agents can generate convincing summaries from thin data. A small number of replies can be written up as demand. The safeguard is to define what counts as evidence before the test runs, and to prefer signals that cost the buyer something, such as a payment or a signed commitment. **Accountability.** If nobody owns a decision, the agent made it by default. Each consequential decision needs a named person and a record of what they were shown. **Dependence on model providers.** The agents run on models supplied by outside companies. Prices, terms, availability and behaviour can change without notice. A company built this way should know which parts of its work depend on a single provider and how it would move them. **Cheap building.** Agents make building cheap, which makes it tempting to build things nobody asked for. A product that works is still some distance from a business, a point argued in [capability is not yet a company](https://frontmark.ai/insights/capability-is-not-yet-a-company). ## How is this different from an AI tool that runs your company for you? Some products are sold with the promise that software will run the business while the owner steps back. Agentic company building, as defined here, differs from them in two ways. The first is ownership. A tool of that kind is something a customer buys and configures. The customer stays the owner, and the vendor carries no stake in the outcome. Agentic company building is an operating practice inside an organisation that owns what it builds. The agents are part of how the organisation works, and the people directing them carry the result. The second is where the human sits. The tool pitch removes the person. The practice described here places the person at the decisions that matter and takes them out of the rest of the work. A founder weighing agentic AI for a startup can test any system this way: ask which decisions it leaves with you, and whether you would be willing to answer for the ones it takes. ## Where Frontmark stands Frontmark is organised around this division of labour. Agents do the work of starting a company, and when a decision needs judgment the work stops for one accountable operator, who answers for it. Access comes from a [collective](https://frontmark.ai/collective) of entrepreneurs, founders, operators and key people. The [agents](https://frontmark.ai/agents) page sets out what agents can and cannot do today, and the [swarm](https://frontmark.ai/swarm) page describes Frontmark's own system. ## Common questions ### What is agentic company building? Agentic company building is the practice of using coordinated AI agents to do the work of starting a company: researching openings, testing demand with real customers, building product and running distribution. Operators make the decisions that need judgment and accountability, such as which customer to serve, what to commit and when to stop. ### Can AI agents build a company? Agents can do a large share of the work: research a market, draft and test offers, write and ship code, run repeatable distribution tasks and build internal tools. They cannot carry responsibility for the result. A person still has to choose the customer, commit the resources, set the standard and answer for what the company does. ### Can AI agents run a company without humans? Not in any sense that matters to customers, employees or the law. A company makes commitments, and commitments need someone who can be held to them. Agents can operate many of the daily tasks of a business, but ownership, obligations and the decision to continue or stop remain with people. ### What is agentic AI? Agentic AI is software built on AI models that pursues a goal across many steps instead of answering one prompt. An agent plans, uses tools, checks its own results and decides what to do next. It keeps working until the task is complete or until it needs a person to decide something. ### What is an autonomous company? An autonomous company is a business whose operations are carried out mainly by AI agents, with little or no human staff. As a description of daily tasks it is partly achievable. As a description of a whole company it leaves out ownership, legal responsibility and judgment, which is why most serious models keep people in the accountable roles. ### What are the risks of using AI agents to build a startup? Four risks matter most. Errors compound when one agent builds on another's unchecked work. Evidence can look stronger than it is. Accountability becomes unclear if no person owns each decision. And the company depends on model providers whose prices, terms and behaviour can change. Each risk is managed with review, records and human ownership. --- # What is an agent swarm? AI agent swarms explained URL: https://frontmark.ai/insights/what-is-an-agent-swarm Category: Definitions Published: 2026-11-16 Updated: 2026-11-16 Author: Frontmark Short answer: An agent swarm is a group of AI agents that work on one goal at the same time, each with its own task, context and tools, coordinated so their outputs combine into a single result. It is one form of multi-agent system, used when work can be split, run in parallel and checked. Key points: - An agent swarm is several AI agents working on one goal at once, each with its own task, context and tools. - A single agent works through a task in sequence, a fixed workflow follows steps written in advance, and a swarm decides how to split and assign the work as it goes. - Common coordination patterns are an orchestrator delegating to workers, parallel exploration of alternatives, and agents reviewing each other's work. - A swarm helps when work is broad, separable and checkable, and adds cost and error when the task is small, sequential or tightly coupled. - A swarm still needs people to set the goal, define what counts as done, grant access and make the decisions that carry accountability. An agent swarm is a group of AI agents that work on one goal at the same time, each with its own task, context and tools, coordinated so their outputs combine into a single result. It is one form of multi-agent system, used when work can be split, run in parallel and checked. The word is borrowed from nature, where large groups of simple animals such as ants or bees produce organised behaviour without a central plan. In AI the usage is looser. Most agent swarms do have some central direction, and the agents in them are capable on their own. What the term keeps from its origin is the idea of many workers acting at once towards a shared end. ## What is an AI agent? An AI agent is software built on an AI model that pursues a goal across many steps. It plans, uses tools such as search or code execution, looks at the result and decides what to do next. A chat assistant answers and waits. An agent continues until the task is finished or it needs a person. An agent works inside a context: the instructions, documents and intermediate results it can hold in view at one time. That context is finite, which is one of the main reasons swarms exist. ## How does an agent swarm work? A swarm starts with a goal that is too broad for one agent to handle well. The goal is broken into tasks. Each task goes to an agent with instructions, the tools it needs and a clean context. The agents work at the same time. Their results are collected, checked and combined. This differs from running the same agent several times. The agents share a goal, so their work has to fit together. Something coordinates them, whether a lead agent, a shared task list or a set of rules. And the division of work is decided during the run, based on what the task turns out to need. ## How is a swarm different from a single agent or a fixed workflow? A single agent does everything in one line of work. It is the simplest arrangement and often the best one. Its limits appear when the task is wide: the context fills with material, earlier details are lost and the work proceeds one step at a time. A fixed workflow is a sequence of steps written by a person in advance. An AI model may perform each step, but the order and the handoffs are set beforehand. Workflows are predictable and easy to audit. They adapt poorly when a task takes an unexpected shape. A swarm takes some of each. The agents decide how to divide and pursue the work, within a coordination pattern that keeps them pointed at the same goal. | Compared on | Single agent | Fixed workflow | Agent swarm | | --- | --- | --- | --- | | Who decides the steps | The agent, as it goes | A person, in advance | The agents, as they go | | How work proceeds | In sequence, in one context | In sequence or in set branches | In parallel, in separate contexts | | Adapts to surprises | Yes, within one context | Poorly | Yes, by reassigning or adding agents | | Cost to run | Lowest | Low and predictable | Highest, and variable | | Ease of tracing a result | Moderate | High | Lower without good records | | Best suited to | Focused, sequential tasks | Repeated tasks with a known shape | Broad tasks that divide cleanly | ## Is an agent swarm the same as a multi-agent system? A multi-agent system is any arrangement in which more than one agent interacts. The term is older than current AI models and covers a wide range of designs, including small teams of specialised agents with fixed roles. A swarm is one kind of multi-agent system. The word usually means the agents are numerous, largely interchangeable, and that their number changes with the work. Two agents with fixed roles passing a document back and forth are a multi-agent system, and few people would call them a swarm. The boundary is a matter of usage, and writers apply it loosely. ## How do agents in a swarm coordinate? Most swarms combine three patterns. **An orchestrator delegating to workers.** One agent holds the goal and the plan. It breaks the work into tasks, hands each to a worker agent and combines what comes back. Workers see only their own task, which keeps their context clear. The orchestrator sees summaries, which keeps its own context free for the overall plan. This is the most common pattern because it gives the swarm a single place where the whole picture is held. **Parallel exploration.** Several agents attempt the same problem by different routes, or examine different parts of a large space, at the same time. One searches one set of sources while another searches a different set. The value comes from coverage and from comparing independent attempts. It suits questions where the right approach is unknown at the start. **Agents reviewing each other's work.** One agent produces a result and another checks it against the instructions, the evidence or a set of tests. The reviewer starts with a fresh context and no investment in the first answer, so it catches errors the author would repeat. Review is what lets a swarm's output be trusted, and it is the pattern most often left out. A working swarm also needs a way to share state. That may be a task list, a set of files or a record of decisions. Without it, agents repeat each other's work or proceed on different assumptions. ## When does an agent swarm help? A swarm is worth its cost when three conditions hold. The work is broad. There is more material to read, more options to test or more components to change than one context can hold with care. The work is separable. Tasks can be handed out without each agent needing to know what the others are doing at every moment. Research across many sources fits. So does testing several versions of an offer, or changing separate parts of a codebase. The work is checkable. There is a test, a source or a measurable response that shows whether a result is right. Without one, a swarm produces a large volume of output and nobody can tell how much of it is correct. When these conditions hold, a swarm finishes sooner than a single agent because the tasks run side by side. It can also be more thorough, since each agent gives full attention to a narrow piece. ## When does a swarm add cost and error for no benefit? Often. A swarm is the wrong choice more frequently than its popularity suggests, for three reasons. Every agent consumes computation, and coordination consumes more. A task one agent could finish will cost several times as much when spread across a swarm, with no better result. Handoffs lose information. An orchestrator's brief to a worker is a summary, and a worker's report back is another summary. Details that mattered can vanish in both directions. When tasks are tightly coupled, as in a design where each choice constrains the next, this loss leaves agents producing mismatched pieces. Errors travel. If one agent's mistaken result becomes another agent's starting point, the mistake is built upon. More agents mean more places for this to happen and more difficulty finding where it began. Start with one agent and add more only when the task shows that it needs them. ## What does a swarm need from people? A swarm multiplies whatever direction it is given, so a vague goal costs more as the number of agents grows. People set the goal and state what a finished result looks like. They decide which tools, data and accounts the agents may use, and where the limits are. They choose the points at which work is reviewed before it continues. And they make the decisions that agents cannot be accountable for: spending money, making commitments to customers, hiring, and stopping work that is failing. In a well-run swarm, agents ask specific questions with the evidence attached, and the answers are recorded so later work starts from them. How this division applies to starting businesses is covered in [what is agentic company building](https://frontmark.ai/insights/what-is-agentic-company-building), and the wider organisational model in [what is an AI-native company](https://frontmark.ai/insights/what-is-an-ai-native-company). ## Where Frontmark stands Frontmark's own swarm is private, and it works only on the companies Frontmark builds. Operators, founders and entrepreneurs set its direction, and it sends each question to the person who can answer it. The [swarm](https://frontmark.ai/swarm) page describes the system and who steers it. The [agents](https://frontmark.ai/agents) page sets out what agents can and cannot do today. ## Common questions ### What is an agent swarm? An agent swarm is a group of AI agents that work on one goal at the same time, each with its own task, context and tools, coordinated so their outputs combine into a single result. The coordination is often done by a lead agent, which divides the work and collects what comes back. ### What is the difference between an agent swarm and a single agent? A single agent handles a task in one line of work, holding everything in one context. A swarm splits the task across several agents that run in parallel with separate contexts. The swarm covers more ground and can check its own work, but it costs more to run and adds the work of coordination. ### Is an agent swarm the same as a multi-agent system? A swarm is one kind of multi-agent system. Multi-agent system is the broader term for any setup in which several agents interact. Swarm usually implies many agents with similar capabilities working in parallel on parts of one goal, with the number of agents changing as the work requires. ### When should you use an agent swarm? Use a swarm when the work is broad, can be divided into largely independent parts, and produces results that can be verified. Research across many sources, testing several alternatives and large changes across separate parts of a codebase are typical cases. For small or strictly sequential tasks, one agent is usually better. ### What are the downsides of agent swarms? Swarms use more computation than a single agent, so they cost more. Agents can duplicate work, contradict each other or pass errors along when context is lost in the handoff between them. Results are harder to trace to a cause. Without clear task boundaries and review, a swarm can produce more output with less reliability. ### Do agent swarms need human oversight? Yes. People set the goal, define what a finished result looks like, decide which tools and data the agents may use, and review the output at agreed points. People also make the decisions that involve money, commitments or other people, because an agent cannot be held accountable for them. --- # What is an AI-native company? Meaning, test and examples URL: https://frontmark.ai/insights/what-is-an-ai-native-company Category: Definitions Published: 2026-11-16 Updated: 2026-11-16 Author: Frontmark Short answer: An AI-native company is a company designed around artificial intelligence from its first day. AI is the core of what the product does and of how the company operates, so removing it would leave no working product and no workable cost structure. It differs from an existing business that adds AI features later. Key points: - AI-native means AI is the foundation of the product and the operating model, present from the first design decision. - The remove-the-AI test is the quickest check: if the product stops working without the model, the company is AI-native. - AI-native describes how a company is built and run, including team shape, cost structure and workflows, and applies beyond the product. - Being AI-native is a design choice with trade-offs, including dependence on model providers and the need for a defensible position beyond model access. An AI-native company is a company designed around artificial intelligence from its first day. AI is the core of what the product does and of how the company operates, so removing it would leave no working product and no workable cost structure. It differs from an existing business that adds AI features later. The term has spread quickly and is used loosely. The definition here has two parts, the product and the way the company itself runs, and the second part gets far less attention than the first. ## What does AI-native mean? AI-native means built with AI as a starting assumption. The phrase follows earlier ones. A cloud-native application was designed for cloud infrastructure, where older applications were moved onto it. A mobile-native product was designed for the phone, where older products were squeezed onto a smaller screen. In each case "native" marks a company that began from the platform and did not have to adapt to it. Applied to a company, the meaning has two parts. The first is the product. In an AI-native product, a model performs the central work. The customer is paying for an outcome the model produces: an answer, a draft, a piece of code, a decision, a completed task. The second is the company itself. An AI-native company assumes AI in its own operations. Its workflows, team structure and costs are designed around what models and agents can do, and around what still needs people. A company can meet the first condition without the second, or the second without the first. The strongest use of the term covers both. ## What is the remove-the-AI test? Imagine the AI is removed from the product and ask what remains. - If the product still works and has lost one feature, it is AI-enabled. A document editor with an AI writing assistant is still a document editor without the assistant. - If nothing useful remains, it is AI-native. A product that answers questions by generating responses has no fallback mode once the model is gone. The test works because it ignores how a company describes itself and looks only at what the product depends on. It has limits. It examines only the product, and it says nothing about quality or durability. A thin interface over someone else's model passes the test, yet it may have little that a competitor could not copy quickly. A company can be AI-native and badly built. ## Which companies are AI-native? Companies are commonly described as AI-native when their core product is a model or could not exist without one. Three broad groups fit. 1. **Model developers.** Companies such as OpenAI and Anthropic build the models themselves. The model is the product. 2. **Products built directly on models.** AI search, coding assistants, image and video generation tools and agent products fall here. The product is a designed experience around model output. 3. **Vertical companies with AI at the core.** These serve a specific industry or function, such as legal work, customer support or clinical documentation, where the model performs work that previously required trained people. Established software companies that add AI assistants to mature products are usually described as AI-enabled. That is a description of origin and architecture, and it carries no judgment about quality. Many AI-enabled companies hold strong positions because they already have customers, data and distribution. The distinctions are covered in more detail in [AI-native vs AI-enabled vs AI-first](https://frontmark.ai/insights/ai-native-vs-ai-enabled). ## What does AI-native mean for how a company is built and operated? An AI-native company treats AI as part of its operating model as well as its product. Five consequences follow. ### The team is shaped differently When models can draft, summarise, write code and handle routine analysis, a small team can cover ground that once needed a larger one. Roles shift towards judgment: deciding what to build, checking output, setting standards and taking responsibility for results. An AI-native startup tends to hire for breadth and taste before it hires for volume. ### The cost structure changes Traditional software has high development costs and very low costs for each additional user. AI-native products carry a real cost for each unit of work, because every request consumes model capacity. Pricing, margins and product design have to account for that from the start. Companies that ignore it can grow usage and lose money on each customer. Inside the company the effect runs in the company's favour. Work that used to require headcount can often be done at the cost of compute, which changes how much capital is needed to reach a given stage. ### Workflows assume models and agents In an AI-native organisation, internal processes are designed with AI as a participant. Research, support, quality assurance and reporting are structured so that a model does the first pass and a person directs and verifies. This is different from giving employees a chat tool and leaving the processes unchanged. When agents carry most of the work of starting a company and people keep the decisions, the practice is called [agentic company building](https://frontmark.ai/insights/what-is-agentic-company-building). ### Evaluation becomes a core discipline Model output is probabilistic. The same input can produce different results, and quality can shift when an underlying model changes. AI-native companies need ways to measure output quality continuously. Evaluation plays the role that testing plays in conventional software, and it is harder to do well. ### The product is designed around outcomes Conventional software gives people tools to do work. AI-native products increasingly do the work. That changes what customers buy and how they expect to pay. It also raises the standard for trust, since the customer is relying on the result and often cannot inspect each step. ## What are the limits and risks of being AI-native? - **Dependence on model providers.** Most AI-native companies build on models owned by someone else. Prices, capabilities and terms can change. - **Thin defensibility.** If the only asset is access to a model that anyone can rent, the position is weak. Durable companies add something from outside the model: proprietary data where rights permit, deep workflow integration, distribution, or trust in a regulated field. - **Capability overtaken by the platform.** A feature that is valuable today can become a standard part of the next model release. - **Reliability and accountability.** When the model does the work, errors belong to the company. Sectors with legal or safety consequences demand careful design and human oversight. These risks are why an AI-native company needs the same foundations as any other: a clear customer, a problem worth solving and economics that hold. The argument is developed in [capability is not yet a company](https://frontmark.ai/insights/capability-is-not-yet-a-company). ## How do you build an AI-native startup? The sequence is the usual one for a startup, with more weight on economics and evaluation. 1. Start with a customer problem where a model changes what is possible or what it costs, and confirm the problem matters to someone who will pay. 2. Design the product around the outcome, with the model at the centre and people in the loop where judgment or accountability requires it. 3. Work out the unit economics early, including model costs per task. 4. Build evaluation into the product from the first version. 5. Decide what will make the company hard to replace once competitors have access to the same models. 6. Structure the team and internal workflows on the same assumption the product makes: AI does the first pass, people direct and decide. A fuller treatment is in [how to build an AI company](https://frontmark.ai/insights/how-to-build-an-ai-company). ## Where Frontmark stands Frontmark is an AI-native company-building institution, and the two-part definition above applies to it. AI changes which companies become possible, and [agents](https://frontmark.ai/agents) do the work of building them while operators make the decisions that need judgment. Ownership adds a third part. Frontmark is built to keep its companies, so that the tools and playbooks made for one are there for the next. The model is set out on the [AI-native company building](https://frontmark.ai/ai-native-company-building) and [approach](https://frontmark.ai/approach) pages. ## Common questions ### What does AI-native mean? AI-native means designed around artificial intelligence from the start. The term is applied to products, companies and organisations whose core function depends on AI models, as opposed to those that existed first and added AI afterwards. The idea mirrors earlier terms such as cloud-native and mobile-native, which described things built for a new platform instead of adapted to it. ### What is an AI-native company? An AI-native company is a company designed around artificial intelligence from its first day. AI is the core of what the product does and of how the company operates, so removing it would leave no working product and no workable cost structure. That separates it from a company that only offers AI features. ### Which companies are AI-native? Companies whose core product is a model or could not exist without one are commonly described as AI-native. Model developers such as OpenAI and Anthropic are clear cases, as are products built directly on generative models, such as AI search, coding and image-generation tools. Established software companies that add AI assistants to existing products are usually described as AI-enabled. ### What is an AI-native organisation? An AI-native organisation runs its internal work on AI by default. Research, drafting, analysis, coding and support workflows are designed with models and agents as participants, and people concentrate on judgment, direction and accountability. The label describes how the organisation operates, so a company can be AI-native in its operations even when customers never see AI in the product. ### What is the remove-the-AI test? The remove-the-AI test asks what would be left if the AI were taken out of a product. If the product still works with one feature missing, it is AI-enabled. If nothing useful remains, it is AI-native. The test is a quick heuristic for the product layer and should be paired with a look at how the company itself operates. ### Is an AI-native startup better than an AI-enabled one? Neither is better by definition. An AI-native startup can design its product, team and costs around what models make possible, which can be a real advantage. An AI-enabled company often has customers, data and distribution already. What matters is whether the company solves a real customer problem with sound economics, whichever label applies. --- # What is company building? Definition and how it works URL: https://frontmark.ai/insights/what-is-company-building Category: Definitions Published: 2026-11-16 Updated: 2026-11-16 Author: Frontmark Short answer: Company building is the discipline of turning an opportunity into an operating business. It covers the full set of decisions a company needs: choosing the customer and problem, forming the team, developing the product, reaching the market, committing resources and taking responsibility for what follows. The product is one part of that work. Key points: - Company building treats the company, and not the standalone product, as the thing being made. - It differs from product development by including the customer, the commercial model, the team, the resources and the ownership. - It differs from consulting and agency work because the builder carries the outcome and usually holds a stake. - It differs from investing because the builder does the operating work before there is a company to invest in. - A company builder works across research, formation, execution, resource commitment and long-term development. Company building is the discipline of turning an opportunity into an operating business. It covers the full set of decisions a company needs: choosing the customer and problem, forming the team, developing the product, reaching the market, committing resources and taking responsibility for what follows. The product is one part of that work. The phrase is used in two ways. Some people mean the general act of starting a business. Others mean the specific practice of organisations that create companies repeatedly, also called venture building or venture creation. This article is about the second. ## What does company building mean? Company building means treating the company as the thing being made. A great deal of early-stage work treats something else as the object: the technology, the product, the pitch or the funding round. A company is a system with several parts that have to work together: - a customer with a problem worth paying to solve; - a product that solves it dependably; - a route to market that reaches the customer at a sensible cost; - economics that leave a margin; - a team that is accountable for the result; - resources matched to the stage; - an ownership and governance structure that suits the horizon. Company building is the work of designing and assembling all of these, in a sensible order, while testing each against reality. A weakness in any one part can undo strength in the others. ## Is company building the same as venture building or venture creation? In most uses, yes. Venture building and venture creation describe organisations that form new startups systematically, using shared teams, capital and methods. The organisations themselves are called venture studios, venture builders, startup studios or company builders. Idealab, founded in 1996, is commonly cited as an early example. "Company building" has a slightly wider reach. Venture creation emphasises the beginning: the idea, the validation, the launch. Company building also includes what follows, such as developing the business over years, deciding how it is resourced and determining who owns it. An organisation can be good at venture creation and still hand companies off early. A company builder in the fuller sense stays. For the studio model specifically, see [what is a venture studio](https://frontmark.ai/insights/what-is-a-venture-studio). ## How is company building different from product development? Product development answers one question: can we make something that works and that people want to use? Company building needs that answer and several more. A product team can ship excellent software and still not have a business. The product may have users but no buyer. It may be valuable but expensive to deliver. It may depend on a channel someone else controls. These are company questions, and a good product leaves them open. The difference shows in what counts as done. In product development, a release is a result. In company building, a release is a test of whether the thesis works in the market. ## How does it differ from consulting, agency work and investing? These activities often involve similar skills and sometimes the same people. They differ in who decides, who owns and who carries the outcome. | Compared on | Company building | Product development | Consulting or agency work | Investing | |---|---|---|---|---| | **What is produced** | An operating business | A working product | Advice or a delivered project | A stake in a company | | **Who decides what to build** | The builder, with its partners | The product owner | The client | The founders | | **Who owns the result** | The builder and the company's team | Whoever commissioned it | The client | Founders and investors | | **How the work is rewarded** | Ownership in the company | Salary or project budget | Fees | Financial return on capital | | **When involvement ends** | Open-ended | At or after release | At delivery | At exit | | **Who carries the risk** | The builder | The sponsoring organisation | The client | Shared, limited to capital invested | Each of these is legitimate work. A consultancy brings expertise to a client's problem. An agency delivers a defined project. An investor supplies capital and judgment to founders who are already building. Knowing which one you are dealing with tells you what kind of relationship you are entering and where responsibility sits. Two of the contrasts change behaviour the most. **Fees and ownership.** A service provider is paid for effort, whatever the commercial result. A company builder is paid only if the company becomes valuable. That changes which projects get chosen and how they are built. **Selection and creation.** An investor chooses among companies that exist. A company builder does the work that brings one into existence, which means taking on the earliest and least certain stage. ## What does a company builder actually do? The work falls into five areas. They run roughly in this order and overlap. **Research.** A company builder studies what is changing, in technology and in markets, and what the change makes possible for specific customers. The output is a view on where an opening exists and why it exists now. **Formation.** The builder turns the opening into a venture thesis: a particular customer, a problem, a product approach, a commercial model and the people who will be accountable. This is the point at which an interesting idea becomes a commitment. **Building.** Product, engineering, design and commercial execution proceed together. The builder develops the product while testing how it reaches customers and whether the economics work. **Resource commitment.** The builder allocates capital, people, infrastructure and attention. Stronger evidence earns a larger commitment. Weak evidence can mean changing direction or stopping. **Long-term development.** After launch the builder remains involved as an owner, supporting the company's leadership and carrying useful capability to later ventures. Every new company also needs its administration done: setting up the entity, hiring, agreeing equity, arranging finance and legal support, and putting basic operations in place. A company builder that has done this many times can do it quickly and with fewer mistakes. ## Why build companies as a discipline at all? Because formation work improves with practice. A first-time founder meets every problem for the first time. An organisation that builds repeatedly keeps what it learns: how to test an opportunity cheaply, which early signals mean something, how to structure a founding team, which infrastructure can be reused. Shared functions can serve several companies at once. Each company can begin from a stronger base than the last. The approach has limits. Shared resources can thin out if too many ventures draw on them. A builder that holds a large stake has to leave enough ownership and autonomy for the people running each company. And a process for creating companies still depends on judgment about which companies deserve to exist. The discipline works when selection is strict and each company keeps its own focus. How these organisations sustain themselves financially is covered in [how venture studios make money](https://frontmark.ai/insights/how-venture-studios-make-money). ## How do you recognise serious company building? Six questions separate the practice from the label. 1. Does the builder start from a customer problem, or from a technology looking for a use? 2. Does it do operating work, or mainly advise? 3. Does it hold a stake and carry risk alongside the team? 4. What happens after launch, and who is responsible then? 5. Does it stop ventures when the evidence is weak? 6. What does it carry from one company to the next? Clear answers suggest a real practice. Vague ones suggest a service business or an investment vehicle using the vocabulary. ## Where Frontmark stands Frontmark is an AI-native company-building institution, and this definition describes the work it is built to do. Its operating line, Research. Form. Build. Back. Compound., follows the five areas above. Frontmark is built to stay after launch as the owner, which is company building in the fuller sense described here. The reason for holding the five together is argued in [capability is not yet a company](https://frontmark.ai/insights/capability-is-not-yet-a-company), and the model is set out on the [approach](https://frontmark.ai/approach) page. ## Common questions ### What is a company builder? A company builder is a person or organisation that creates businesses as its core activity. It finds and tests opportunities, forms ventures around them, supplies or assembles the team, builds the product, develops the first routes to market and commits resources. Organisations that do this repeatedly are also called venture studios or venture builders. ### What is the difference between company building and product development? Product development produces a product: something that works and that users want. Company building produces a business: a product together with a defined customer, a way to reach them, viable economics, an accountable team and an ownership structure. A team can succeed at product development and still not have a company. ### Is venture building the same as company building? The terms are used almost interchangeably. Venture building and venture creation usually refer to an organisation forming new startups repeatedly with shared resources. Company building is a slightly broader phrase that also covers the continued development of a business after its launch, including how it is resourced and owned over time. ### How is a company builder different from an investor? An investor provides capital to a company that founders have already started, and usually supports it from the board. A company builder does the formation and operating work itself, often before a company exists. It contributes people, product work and commercial effort as well as resources, and typically holds a founding stake. ### How is company building different from an agency or consultancy? An agency or consultancy is paid by a client for a defined piece of work, and the client owns the result and the risk. A company builder works on its own account or as a co-owner. It decides what should be built, stays after delivery, and its return depends on whether the company succeeds. ### What skills does company building require? It requires a combination that is rarely found in one person: research and market understanding, product judgment, engineering and design, commercial execution, hiring, and capital allocation. Above those sits the ability to decide, on evidence, when to commit more to an opportunity and when to stop. This is why company builders tend to work as teams.