Why AI May Hollow Out the Middle of the Economy

AI may favour giant context-rich firms and tiny fast-moving teams, leaving medium-sized companies trapped between both advantages

Jul 12, 2026

I went to my first All Blacks game in Wellington this week hence the cover.

Whenever I travel, I always think of random ideas and it’s probably because I read buzzy books. None buzzier than this one.

Why Information Grows - view it on Amazon here.

So I started off by thinking about a dumb question and I find that dumb questions are the best ones, because people don’t ask them.

So my dumb question for the day is, what is a job?

Easy! A job is what somebody does for money! Yes that is true, but if we follow someone around and look at what they do, jobs seem kind of strange.

Let’s take a project manager. Let’s be like David Attenborough and follow them for a week.

They write status updates
They book meetings
They calm down a client who is panicking about nothing
They notice two departments hate each other and make it so that they both agree without somehow hating each more
They remember what everyone agreed three weeks ago, because nobody else wrote it down.

Their job is a series of random tasks. The book references an essay called the transaction cost theory of the firm and it is something that I had never thought about but imagine if instead of hiring a ‘project manager’, you had a hypothetical market and you hired someone for each individual task. Like going to Fiverr.

So you pay someone $15 for some status updates and maybe $57 to book some meetings. You would probably save money but it would be annoying as you would have to vet people each time.

As a result, the tasks aren’t actually bundled together because the work demands it but more because hiring one person is cheaper than going to the market and buying each little task separately.

So we did something cool and created a magic trick. We took thirty unrelated requirements, squeezed them into one ongoing relationship with one human being, and called the whole thing a job.

And now we ask why do firms exist?

Now we know why jobs exist, but then why do firms exist?

Because as we said, coordinating work through the open market costs money. A firm is simply a bubble inside which those costs are lower than outside the bubble.

The size and shape of the bubble is set by those costs. But if you change the costs then you change the bubble and ultimately the firm itself. This is a theory put forward by Ronald Coase in 1937 The Nature of The Firm

AI changes the cost structure of firms in a number of ways.

It makes work cheap to break apart
cheap to describe
cheap to hand out
cheap to check
cheap to reassemble.

It actually makes hiring easier. Now if there is good context, a new staff member can pick up where someone left off. We have been finding this in our company. Because there is context, if someone is sick or we have a new staff member, they are able to pick up and keep moving the project forward. Because we now have this context layer that the AI can leverage.

So then we might say, that’s great, AI is going to help these people doing jobs get faster.

Now the lawyer will get faster!
Now the programmer will write more code faster!
Now the project manager can manage more projects!

But this is where it get buzzy. That economic theory of the firm from 1939 talks about how if the cost structures change then the bubble of the firm has to change too.

When you look at it from this lens, the first bundle to come undone is the job. Once that gets undone, it makes sense to question the function of the department and as soon as you are at that level, you might as well question the existence of the entire firm.

Think about it.

When you query ChatGPT for legal advice, you are assembling what used to be a lawyer rapidly and then it is dissolves. Everything appears around a problem. It’s actually a form of going to the market for the random tasks we need to do without all the annoying inefficiencies.

The “organisation” (our legal problem) forms around the shape of the problem the way water forms around a rock.

Now, you might think this is all speculation about some liquid future. It isn’t, and this is my favourite part, because there’s an industry where this experiment has already ran to completion.

Films.

In 1940 Hollywood looked exactly like the firms I’ve been describing.

The studios were factories.
Actors were employees on seven-year contracts.
Writers clocked in at the writers’ building.
Directors were staff.
Everything happened on the lot, inside one legal boundary, because coordinating a film through the open market was unthinkably expensive.

Then the conditions changed, and the system melted.

Look at what a film is now. A film is a temporary company. It assembles hundreds of specialists, a cinematographer here, a stunt team there, a colourist, a composer, an insurance broker, each arriving with their reputation attached. The company exists for eighteen months, makes exactly one thing, and dissolves. The ending credits are the organisation chart of a firm that no longer exists.

You basically have these strange studios now that are holders of capital, owners of content libraries and controllers of the distribution system. You can sue them as well which is a great thing but you have these small satelite production companies that form around them.

So the industry settled into a peculiar shape. You’d expect a bell curve, most companies medium-sized. You get the opposite. Two humps and a valley between them.

The humps on the left are the massive companies and on the right are the production companies.

This graph looks weird I admit but it has been buzzing me out the more I think about it. I can’t decide if it’s right but here is my argument.

Has AI basically made it so that there is an uncanny valley between those two modes? Now is the optimal strategy to decide if you want to be a context building machine (valley 1?) or a fast moving production company (valley 2?).

Is it really difficult to be a medium sized business now because you sit between both?

Maybe there is going to be a fast world (valley 2) and a slow world (valley 1?). Maybe AI won’t affect both the same?

In my head there is some power to building context, I think about all of the IP my company has in developing software and AI. That would lead me to think that context as a strategy is a really important north star. But maybe it is because we are a smaller company and can move fast like valley 2, that we can service companies in valley 1 that makes it work? If we became too big, then we wouldn’t be fast and we would lose all of our key advantages?

This graph helps though because when you are going to make a decision of what to buy, you can buy from valley 1 (context) or you can now buy from valley 2 (AI and fast but temporary.) For some things like branding and marketing that might be okay?

My intuition is that most people do not want to live permanently in the world of the second hump. It might be economically efficient, but it would also be psychologically exhausting. Few people want to spend their entire lives auditioning for the next temporary organisation, carrying their reputation from project to project and watching every team dissolve as soon as the work is finished.

That might explain why AI has caused less disruption than even Sam Altman seemed to expect by now.

Transaction costs are not only financial. Trust has a cost. Belonging has a value. So do reputation, accountability, institutional memory and knowing that somebody will still answer the phone in six months.

Perhaps, then, the winning strategy is not to choose one hump?

Perhaps it is to become a context-building machine on the left, while developing the ability to temporarily behave like a production company on the right.

The firm would retain the things that should persist: its memory, capital, relationships, reputation, permissions and understanding of the world.

But around each problem it could rapidly assemble the people, agents, software and specialists required to solve it. Once the problem was solved, that temporary organisation would dissolve, while everything it learned would flow back into the permanent core.

Maybe reread that sentence a few times as I think it’s the crux of this whole article.

That would make context more than an asset held by the company.

Context would become the company. This feels closer to the right answer.

I read that Richard Feynman sometimes ended his lectures without tying everything into a neat bow, so I will do the same.

Is this two-humped economy actually emerging? Are the companies caught in the middle becoming structurally disadvantaged? Is context as a strategy actually viable now?

I genuinely cannot decide.

But the graph has been buzzing me out all week.