Agents build the companies. Operators hold the judgment.
Most firms use AI to help people do the work. Frontmark is organized the other way round. Agents do the work, and people are asked when it matters.
Agentic company building is the practice of using coordinated AI agents to do the work of starting a company: researching openings, testing demand, building product and running distribution. At Frontmark, agents also build their own tools, ask human operators for decisions that need judgment, and draw on the institution's companies and network.
Work goes up. Decisions come down. Access comes from the side.
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.
An illustration of the loop, not a live feed.
The work, done continuously.
Start businesses
Agents turn an opening into a testable venture: a customer, a problem, an offer and a way to reach the buyer.
Validate them
Agents put the offer in front of real buyers and measure what happens. Demand is tested before a company is formed around it.
Build the product
Agents write the code, ship the product, run the distribution and keep the evidence current.
Build their own tools
When a task needs a tool that does not exist, agents build it. The tool is added to the shared stack and is there for the next company.
Ask operators
Agents know the limits of evidence. When a decision needs judgment, taste or accountability, they ask a person who has built before.
Use the institution
Agents draw on Frontmark's companies, infrastructure, data where rights permit, and the collective's network. Nothing starts from zero.
What agents still cannot do, stated plainly.
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%. Volume is solved sooner than completion.
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
Agents follow rules and gather evidence well. Choosing a customer, committing capital or stopping a company depends on experience and on 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.
Speed belongs to the agents. Responsibility stays with people.
An agent can test ten offers in the time a team debates one. It cannot be accountable for a customer, a hire or a commitment of capital. Those calls belong to operators, and they are made on the record.
This is the division of labor the model depends on. It is also why the collective matters: the quality of the decisions sets the ceiling on what the agents can build.
Agents and operators across the model.
| Action | Agents | Operators |
|---|---|---|
| Research | Agents read markets, costs and customer behavior continuously, and surface openings with the evidence attached. | Operators decide which openings deserve a thesis. |
| Form | Agents test demand, price and positioning against real buyers before a company exists. | Operators choose the customer, the model and the people accountable for it. |
| Build | Agents write code, ship product, run distribution and build the tools the next task needs. | Operators hold product judgment and the standard for what ships. |
| Back | Agents keep the evidence current: usage, revenue, cost and what customers do next. | Operators commit more, hold, or stop. |
| Compound | Every tool, playbook and data source an agent builds stays in the institution. | Ownership keeps Frontmark connected to the company long after launch. |
About agents at Frontmark.
What is agentic company building?
It is company building in which coordinated AI agents carry out much of the work: research, demand testing, product development and distribution. People set direction, make the decisions that need judgment and stay accountable for the company.
What is an agent swarm?
An agent swarm is a group of AI agents working on related tasks in parallel, each with a role, sharing context and tools. A swarm can explore many options at once and report back what the evidence supports.
Do agents replace founders and operators?
No. Agents change what operators spend time on. They take on the volume of work, so people can concentrate on the decisions that determine whether a company deserves to exist: the customer, the standard, the commitment and when to stop.
What happens to the tools the agents build?
They stay in the institution. Every tool, playbook and data source an agent builds is added to a shared stack, so the next company starts with more than the last one did. This is part of what Frontmark means by compounding.
Is the operating loop on the homepage live data?
No. It is an illustration of how agents, operators and the network interact. Frontmark does not publish live operational data.
Build at the front.
Bring a problem you understand deeply, a capability worth building around or a distribution advantage that changes the opportunity.
- A problem you have lived inside, with customers you can name.
- A capability or infrastructure that deserves a company around it.
- Distribution, access or a market position others cannot reach.