There is an idea getting popular in AI circles. The atomic unit of work, the argument goes, is shifting from the person to the process. Companies will dissolve into orchestrated workflows. Org charts will become irrelevant. Eventually, the people themselves become optional.
Parts of this are right. Most of it is wrong in a specific way that matters.
The right part is that work which used to require people no longer does. Customer support. Content production. Routine analysis. Code execution. These were once roles. They are becoming processes. That shift is real and it is not reversing.
The wrong part is what comes next. The argument assumes that if a layer can be removed, removing it is correct. That logic ends at zero. A company with no people. Pure agents executing pure processes.
That endpoint does not exist. Not because the technology cannot get there. Because something has to decide what the company is for.
What gets removed
Walk down the path of unbundling. A traditional fifty-person company has marketing, sales, engineering, support, finance, ops. Full org chart. Each function staffed by humans.
With early automation, that company runs leaner. Maybe thirty people. With better tooling, fifteen. With current agents, one founder plus a stack can handle most of what the fifteen did.
The next step looks obvious if you only count layers. Why not zero?
Because removing the last person is not the same as removing the others. The others were doing work. The last one is making decisions. Those are different categories.
What does not get removed
Work is something a process can do. Deciding what work to do is not.
A company has to take a position in the world. It has to choose what to build and what to ignore. Which customers to serve and which to refuse. What it stands for and what it never will. These are not workflows. They are claims. Someone has to make them.
An agent can be asked to make these claims. It will produce an answer. The answer will be defensible. It will also be average. The agent does not have a position. It has a probability distribution over what reasonable answers look like, given the inputs.
A company that runs on agent-shaped claims becomes an agent-shaped company. Technically functional. Strategically indistinguishable from every other company built the same way. The thing that made it a company in the first place is gone.
Why the founder is the unit
The atomic unit of productivity may be a process. The atomic unit of a company is a person who decides which processes are worth assembling.
This is not romantic. It is structural. Below one person, the company stops being itself. It becomes a portfolio of automated decisions held together by inertia.
Some categories will tolerate this. Pure execution businesses. Commodity services. Anything where the right answer is generic. There the unbundling can go further than one.
For everything else, the floor is one. One head making the calls that compound into identity. Everything around that head can be automated. The head itself cannot.
What this means in practice
If you are building a solo company on AI, the leverage is not in adding more agents. It is in getting sharper about the decisions only you can make.
Most founders running this configuration spend too much time on the stack and not enough on the position. The agents are not competing with you. They are competing with the employees you never hired.
Once that boundary is right, the remaining optimization happens at the founder layer. Not the tooling layer. Better tools help. They do not make the company itself better.
The atomic unit is not the agent. It is the person who decides what the agent should do.
The picture that emerges is not a world without companies. It is a world with more companies, each smaller and more specific than what came before. The one-person company is not a stop on the way to something bigger. It is a stable shape. A new floor.
The unbundling goes far. It does not go to zero. It stops at one.
.svg)
