Can AI run a business? The autonomous company, examined

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 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, 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, an AI shopkeeper improved markedly with newer models. Anthropic nonetheless says the gap between capable and completely robust remains wide.
Gartner, 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.
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, 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.
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 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.
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 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.