The AI Firm

Why do companies exist?

It’s a surprisingly difficult question.

In 1937, economist Ronald Coase tried to answer it in a paper called The Nature of the Firm.

His insight was simple: markets aren’t free.

If I need something done, I can hire an employee to do it. Or I can find someone outside the company.

But using the market requires work.

I have to find someone, explain what I want, negotiate a contract, monitor the work, resolve problems, and make sure everyone has the information they need.

Sometimes it is easier to just hire the person.

That is one reason firms exist, but there is a limit.

As a company gets bigger, coordinating everything internally becomes more difficult. More employees means more communication, more management, more bureaucracy, and more decisions that have to move through the organization.

Eventually, it can become cheaper to stop doing something internally and buy it from the market instead.

Coase’s basic argument was that firms grow until the cost of organizing another transaction internally becomes greater than the cost of carrying out that transaction through the market.

That gives us a useful way to think about the size of a company.

It also raises an interesting question about AI.

What happens when the cost of coordination changes?

Technology Changes the Boundary

This isn’t the first time technology has changed the economics of coordination.

The railroad created an entirely new management problem.

A company could suddenly have thousands of employees spread across hundreds of miles. Someone had to figure out who was responsible for what, how information moved through the organization, and how managers could keep track of everything.

The answer was hierarchy. Companies developed departments, reporting structures, standardized processes, and layers of management.

Communication technology then started attacking some of those coordination costs.

The telegraph and telephone made it easier to communicate across distance. Computers made information easier to process. The internet made communication nearly instantaneous.

Each technology changed what was economical to coordinate.

I wrote about a version of this in an earlier 25th Hour article about the economics of game development. The interesting question was whether technology could change which activities needed to happen inside a company at all.

AI brings that question back in a much bigger way.

AI Is Different

A computer makes a person more productive.

AI can make an organization more productive.

An employee using AI might research a problem faster. An AI agent can potentially research the problem itself.

It can search for information, synthesize it, communicate with other systems, execute tasks, monitor results, and continue working after the employee has gone home.

Microsoft’s current vision for AI at work increasingly reflects this shift. Humans set direction and own outcomes while agents take on more of the execution. Microsoft reported that the number of active agents in its Microsoft 365 ecosystem had grown 15x year over year as of early 2026.

This changes the amount of coordination one person can perform.

Consider a company with ten employees.

Today, those ten people might need several managers simply because someone has to coordinate their work.

Now imagine each employee has a collection of AI agents.

One researches. One analyzes. One handles routine communication. One monitors a process. Another checks the work.

The human still matters, but the amount of economic activity that person can oversee has increased dramatically.

The basic organizational unit starts to change.

Instead of an employee it’s now a human-agent team.

So a company could have fewer employees while being capable of doing much more.

The company of the future may be smaller in headcount but larger in economic scope.

The Coasean Paradox

This is where Coase becomes especially useful.

AI can make it cheaper to coordinate things inside a company, but it can also make it cheaper to coordinate things outside the company.

Imagine that a company needs a specialized piece of analysis.

Today, it could hire an employee, or it could hire a consulting firm, or it could find a freelancer.

Each option comes with transaction costs. Someone has to find the right person. Explain the problem. Transfer information. Negotiate the contract. Monitor the work. Review the result.

Now imagine an AI agent can handle much of that process. It finds the appropriate provider. It supplies the relevant context. It monitors the work. It checks the output.

The market becomes easier to use.

That leaves us with an interesting problem.

AI could make companies want to become both larger and smaller.

If AI makes internal coordination much cheaper, there is less reason to outsource. Why hire another company if an AI system makes it inexpensive to run the function yourself?

At the same time, if AI makes it much easier to find, contract with, monitor, and work with outside parties, there is less reason to bring those activities inside the company.

Why hire an employee if the outside market becomes dramatically easier to coordinate?

This is essentially the Coase problem all over again.

The difference is that the technology is attacking both sides of the equation.

We don’t yet know which side will fall faster.

The New Firm

This is why I don’t think the most interesting question about AI is whether companies will become smaller.

They might.

Or they might become much larger.

The more interesting question is:

What does a company need to own?

Historically, owning something inside the company was often useful because coordination was difficult.

If you couldn’t easily monitor an outside supplier, you might bring the activity in-house.

If information was difficult to transfer, you might hire the person who already had the knowledge.

If managing a large number of people was expensive, you might split the organization into departments and create layers of management.

AI changes those calculations.

Information can move more easily.

Work can be monitored more continuously.

One person may be able to coordinate far more activity, but AI creates new costs too.

Someone still needs to decide what the system is allowed to do. Someone needs to determine whether the output is correct. Someone has to be accountable when the system makes a mistake. Someone needs to decide what should happen when the AI’s objective conflicts with the company’s actual interests.

AI may reduce some coordination costs while creating new governance costs. That could become one of the defining management problems of the next decade.

The result probably won’t be a world where every company becomes a one-person company. Instead, the boundary of the company may become much more fluid.

What Is the Optimal Firm?

Coase gave us a useful way to think about the firm nearly 90 years ago.

Companies exist because markets have costs, but companies have costs too.

The optimal organization depends on which is cheaper.

AI is now changing both.

That makes the question of organizational design more interesting than simply asking how many jobs AI will eliminate.

The better question is: How much economic activity can a given number of people coordinate?

If the answer keeps increasing, the relationship between headcount and company size starts to break down.

A company with 50 people might eventually be capable of doing what once required 500. A company with 5,000 people might be capable of doing what once required 50,000.

And a company with millions of customers might need surprisingly few humans to coordinate the systems serving them.

We usually think of technology as making individual workers more productive.

AI may be doing something bigger.

It may be changing the economics of the organization itself.

And if coordination is one of the fundamental reasons firms exist in the first place, then AI could change the optimal size, shape, and boundaries of the firm.

The question isn’t how big the company should be.

It’s where coordination should happen.