You built it with agents. Now who is your first human?

Most commentary aimed at these founders is scolding. Hire people, build a culture, stop trying to do everything alone. It lands badly, because the founder is not being stubborn. They are making an accurate calculation.

Headcount carries management time, onboarding, payroll risk, notice periods and the considerable cost of being wrong about a person. Agents carry none of that. They arrive competent, they do not need a one-to-one, and they can be switched off on a Tuesday afternoon with no consequence. Against that comparison, "you have too much to do" has stopped being a reason to hire anybody.

So the honest version of this argument concedes the point. Agents genuinely did remove the throughput justification for early headcount. What they did not remove, and cannot remove, is the ceiling.

Agents extend the floor. They take whatever the founder already knows how to specify and they do more of it, faster, at any hour. The ceiling is different. The ceiling is the limit of what the company is capable of attempting, and in a company of one, the founder is the ceiling. Every agent added raises the floor closer to it. None of them lift it.

Before naming any role, apply one test. If an agent can hold it, an agent should hold it. There are exactly four things an agent cannot, and every defensible early hire traces back to one of them.

Three roles, and no others, survive the test at this stage. Each one is a person the founder is currently being, badly, in the gaps between building.

In plain terms, before the titles get in the way: operations, HR, and marketing and sales. One person for each. Not a department, not a function to be staffed out, and not the beginning of an org chart. Three people, chosen because between them they cover everything the founder cannot hold and nothing that an agent already handles.

Notice what is absent. No general manager, no first engineer, no assistant in the pre-agent sense. Section 06 explains why each of those feels obvious and is not.

Any article claiming a single universal first hire is guessing. The order genuinely changes, and it turns on one observable fact rather than on stage, funding or team size.

Do you have customers yet? That is the whole branch. Both routes arrive at the same three people within a year, and starting on the wrong one costs six months.

In both sequences HR and Agent Operator are two of the first three. The argument does not depend on which route a founder takes.

Founders have started describing this as their company AI brain. The term will date, so call it what it is: institutional memory. The accumulated context of how this company operates, what it decided and why, what is true about the product today, and how it sounds when it speaks.

It deserves a section here because it reorders the hiring question entirely. Every founder at that event has access to the same models. Agents are rented and interchangeable. Institutional memory is the only part that is proprietary, which makes it the actual asset being built, and it changes who the first humans should be. They are not the people performing tasks. They are the people who build and maintain the asset.

Three uncomfortable properties follow. Memory does not populate itself: it is assembled from what humans notice, decide and record, so a company of one has a memory made of exactly one person's context, and it starves because that person is occupied building. Memory left alone gets actively worse: stale pricing, superseded decisions, product claims no longer true, all of it acted on confidently and at volume, which makes a neglected memory more dangerous than none. And memory is what makes new people productive quickly, because onboarding is almost entirely context transfer.

That last property is the one that closes the loop. The asset that makes agents effective is the same asset that makes humans effective. Which is why the two roles below both feed it as a byproduct of doing their own jobs rather than as a documentation project nobody ever gets to.

Each of these was the correct early hire five years ago. Each fails the test now, for a different reason, and the reasons are worth separating because two of them are not about AI at all.

Notice that two of the four were removed by ordinary self-serve tooling rather than by AI. Founders reasoning only about what agents absorbed will still hire one job that quietly ended years ago.

Here is the part that resolves the objection rather than arguing with it. Everything above can be tested through Volunteers, which is not headcount. No payroll, no notice period, no permanent commitment made on the strength of a hypothesis about a role that barely has a title yet.

That matters most for the two roles a founder is least certain about. Bringing in an Agent Operator means discovering, within weeks, whether handing over the fleet actually returns the founder to the product. Bringing in an HR Volunteer means finding out whether a human layer is genuinely a throughput problem or whether it was imagined. Both questions are answered by trying, and neither answer requires a contract.

The exchange is honest in both directions. The founder receives judgement they could not otherwise justify at this stage. The Volunteer receives something a course cannot manufacture: a real founder who needs the outcome, real customers on the other side, and consequences attached to being wrong. Essay 04 argued that this is where the competences are actually acquired. The same arrangement is what lets a founder discover their first three humans without betting the company on the guess.

And the sequence is self-reinforcing. The HR Volunteer learns how this specific company operates, deposits it into institutional memory, and then sources the Agent Operators and the demand talent that follow. The second human is the one who finds the third, the fourth and the tenth.

Which is the real reason to start with three rather than with a team. Nothing here argues that a company of four is the destination. It argues about order. A team assembled after these three arrive is quicker to bring in, clearer about the standard, and considerably less likely to contain a role that no longer needs to exist.