Humans are the worst thing that can happen to a startup

“Let's start something.”

For a moment, the sentence makes everyone look in the same direction. There is a problem worth solving, an idea worth testing, perhaps even a business worth building. People talk about what they could make together. The language is collective: we, our, us.

Then the work begins, and the pronouns stop describing the behavior.

A decision that should move the product forward becomes a negotiation over visibility. A difficult week becomes someone else's responsibility. A promise survives in conversation long after it has disappeared from the calendar. Nobody formally withdraws from the project. They simply begin serving a different one.

That is the false pact. “Let's start something” sounds like a shared commitment, but it often means several private intentions temporarily using the same business address.

The intentions behind the agreement

Not everyone arrives wanting to build a system that survives contact with reality.

• Some want the game: the adventure of a startup, not the repetition required to operate one.
• Some want fame: the business as content, with execution somewhere behind the camera.
• Some want their ego protected: being right matters more than getting the result.
• Some want family, social visibility, or a version of success abroad. Those are legitimate goals. They are not automatically compatible with the sacrifice a particular project demands.
• Some wait for luck. Others wait for someone else to absorb the cost of making the business work.

The problem is not that people have lives outside a company. They should. The problem is agreeing to one level of responsibility while privately intending to carry another.

Few are willing to lose their humanity, and almost everyone expects someone else to lose it for them.

I do not mean that losing your humanity is a virtue. A business that requires someone to disappear into it has its own problem. I mean that the costs people refuse to name do not vanish. They move. Someone else receives the unfinished work, the uncertainty, the exhaustion, and the obligation to keep the promises alive.

What was presented as a team becomes a mechanism for transferring those costs.

What the pattern taught me

For years, I chose badly the people who accompanied me. I met very few people with whom the connection was worth the investment. There were exceptions, and they matter. But the recurring pattern was not one unusually bad person. It was self-interest dressed up as a team.

Almost all of them looked after an interest that was not the project, while continuing to speak as if the project were the reason they were there.

I could turn that history into an inventory of disappointments. It would explain the wound, but it would not explain what I am building now. The useful evidence is shorter: the same mismatch appeared often enough that I stopped treating it as an accident.

Where the money should have gone

Disappointment used to send me back to the same question: where do I find the right person?

Eventually, I asked a different one: which parts of this work actually require a person?

My development agents now surpass me in parts of the work, and they surpass work I previously paid people to do. That comparison is not philosophical. It is a comparison between what was delivered then and what I can get built now.

If I could go back, much of that money would go to inference.

Not to another meeting about commitment. Not to another role whose existence required more explanation than its output. To the capacity to turn a specification into something I can inspect, run, and improve.

AI does not represent my intentions because it is a good person. It is not a person. It does not bring a personal biography that the business must quietly accommodate. It does not need the project to validate its status, advance its social life, or give it a stage.

That does not make it infallible. An agent can misunderstand, fabricate, fail, or follow a bad instruction with remarkable efficiency. Its output still needs constraints and verification. But those are engineering problems, not a hidden negotiation over what the collaborator really came to obtain.

AI is not a kinder partner. It is a different kind of collaborator: one without a parallel personal agenda.

Presence sold as capability

The economic lesson is uncomfortable because it is more precise than “humans are the worst.”

Many of the jobs I hired for were not justified by the result. They were justified by the fact that someone could talk, walk, attend a meeting, and occupy a recognizable role.

I call those simulated jobs: presence sold as capability.

This is not a claim that every human job is a simulation. Judgment, care, responsibility, and real expertise are not interchangeable with attendance. The distinction is between work that produces something and the performance that substitutes for it.

For years, I accepted the human form as evidence that execution existed. Someone had a title. Someone could explain the process. Someone was available to discuss the difficulty. Those signals became proxies for a result that remained missing.

Inference reverses the criterion. The question is no longer whether a role has been filled. It is whether the work has been done, whether the output passes inspection, and whether the system can do it again.

A convincing conversation does not satisfy that test. Neither does a confident model response. The result has to exist.

The business without the theater

That is why I am building an autonomous orchestrator.

I am not interested in a system that merely accompanies a business, offering suggestions while the old machinery remains mandatory. I want a system that assembles and operates businesses: turning intent into work, coordinating execution, and checking what comes back.

The promise is not that every decision disappears or that software becomes exempt from accountability. It is that building and operating a business no longer needs the layer of human theater that used to stand between an idea and its execution.

The scarce resource should be a clear objective and the ability to evaluate results—not access to people willing to perform commitment convincingly enough to be hired.

My past explains why this matters to me. It does not get to become the design specification. A system built to punish would remain attached to the people it was supposed to leave behind. A system built to remove an inefficient layer can move on.

What happens when the room gets quiet

Entrepreneurship does not fail for lack of talent alone. It fails when the business is assembled from people whose real goals lie somewhere else, and that difference is hidden until the costs come due.

For a long time, the status quo held because hiring humans was the only practical way to execute much of the work. You could dislike the theater and still have to buy a ticket.

That is ending.

The people who lived off the simulation are going to notice the silence.