Contact

Most of Frontmark is not people.

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.

In short

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.

Agents report, operators decide, and the collective opens the market.

Agents report what they did and what the evidence shows. Operators answer the questions that need a person. The collective opens what a new company could not reach alone.

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.

  1. 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.

  2. 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.

  3. 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.

  4. 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.

Build at the front.

Bring a problem you know from the inside, a market you can open, or capital with patience. We answer what fits.