The hourglass economy

I keep seeing the same shape.

Not in a vision. I’m not that dramatic. I mean in meetings, in job posts, in investor decks, in what friends tell me after a second beer.

It looks like an hourglass.

At the top: huge companies. Bigger than companies, really. Capital structures with logos. They own compute, distribution, compliance, contracts, and enough lawyers to make reality negotiable.

At the bottom: tiny shops. Two people, a laptop, a stack of AI agents, a Stripe account, maybe a physical workshop, maybe a local reputation that still means something.

In the middle: the part most of us were told to aim for. The agency. The regional software company. The forty-person consultancy. The internal department where a normal person could build a normal life.

That middle is getting squeezed.

The first end: the giant room with almost nobody in it

The easy version of the AI story says companies will get bigger because the systems are expensive. That is true, but incomplete.

The stranger version is that the companies get bigger while the teams inside them get smaller.

You can already feel it. A product surface that used to need a designer, two backend engineers, a frontend engineer, a data person, a QA pass, and a project manager now has one sharp person steering agents through the first draft. Not the final draft. Not magic. But enough of the slope changes.

Now put that inside a company with the cash to buy compute in bulk, the data to train on its own exhaust, and the distribution to ship to millions before lunch.

The unit of leverage stops being “team” and becomes “operator plus agents plus infrastructure.”

That sounds empowering until you notice where the infrastructure lives.

If the model, data, payments, deployment, legal cover, support channels, and customer trust all belong to the platform, then the human operator is powerful and fragile at the same time. A very expensive fingertip on a very large hand.

I do not mean the work is easy. The people inside those rooms are often excellent. They have to be. A tiny team running a huge surface needs taste, systems judgment, stamina, and the ability to hold an absurd amount of state in their head.

But it changes the career bargain.

The old corporate promise was: give us your years and we will give you stability. The new one is closer to: give us your nervous system and we will give you leverage, until the leverage points somewhere else.

The second end: the two-person business with ten invisible employees

The other end is more interesting to me.

A couple runs a design studio. One handles clients and taste. The other handles production and operations. Around them: an agent that drafts proposals, one that triages invoices, one that cleans up CAD files, one that keeps the website alive, one that watches inventory, one that turns meeting notes into next actions.

Or a small bakery. Not “AI bakery” in the cursed LinkedIn sense. Just a real bakery using agents for supplier emails, rota planning, margin checks, local ads, allergen paperwork, maybe a little demand forecasting when the weather turns.

Or a two-person robotics shop. One engineer, one operator, a bench full of parts, a small fleet of software agents doing documentation, test harnesses, sourcing, quoting, and customer support.

The point is not that AI replaces the business. The point is that AI removes the layer of clerical gravity that used to force a small business to hire before it was ready.

That is a real social change.

For the first time, a tiny company can have some of the coordination surface of a medium company without becoming a medium company. It can answer faster. Quote faster. Prototype faster. Look more polished than its headcount should allow.

There is a dark side, obviously. More pressure. More noise. More tiny firms pretending to be bigger than they are. More people working alone inside a dashboard at midnight while an agent politely reminds them that cashflow is bad.

Still, I think this end of the hourglass has dignity in it.

The value is not only scale. The value is short accountability. If the chair is bad, you know who made it. If the software breaks, you know who shipped it. If the bread is perfect, there is a person within shouting distance who can be proud.

AI agents can make that kind of business less administratively suffocating. That is not utopia. It is a better set of tools.

What happens to the middle

The middle has a problem: it was built on coordination costs.

A fifty-person agency exists partly because clients need more than one freelancer but less than an enterprise vendor. A regional IT department exists partly because software is annoying and someone needs to glue systems together. A mid-market consultancy exists partly because expertise is scarce, documents are messy, and meetings need humans who can translate.

AI attacks that glue.

Not all of it. Never believe a demo too quickly. The world is full of edge cases, politics, bad data, angry customers, procurement rituals, legacy systems, and codebases that smell like wet carpet.

But enough.

Enough first drafts. Enough summarization. Enough boilerplate. Enough tests. Enough reports. Enough “can you make a quick version of this by Friday?” Enough of the work that justified a layer of juniors, coordinators, analysts, account people, and internal process.

The middle does not vanish in a flash. It thins.

One company acquires another “for the client base.” A department is “streamlined.” A vendor is replaced by a platform. A team of twelve becomes three senior people and a private agent stack. Everyone says it is a productivity story because that sounds cleaner than admitting it is a labor story.

The hourglass tightens one quarter at a time.

This is not humans versus machines

The lazy framing is humans versus AI. I do not buy it.

The real contest is between forms of organization.

AI makes large organizations more scalable. It lets them push policy, product, content, support, monitoring, and analytics through fewer human bottlenecks. It turns the corporation into something more automatic.

AI also makes small organizations more capable. It lets two competent people punch above their weight without immediately hiring a manager, a copywriter, a bookkeeper, a junior dev, and a support inbox.

So the question becomes: which structure absorbs the gains?

Do we get a few enormous corporate organisms with tiny elite crews controlling vast surfaces of life?

Or do we get many small, weird, competent businesses run by a couple of people and their agent systems?

Probably both. That is the hourglass. The top and bottom grow while the waist gets meaner.

Where I would stand

I work in AI systems, so I am not writing this from a cabin in the woods. I like the machinery. I like runtimes, kernels, robot policies, graphs, ugly interfaces, and the satisfying click when a system finally runs end to end.

But if I were choosing a life strategy, I would not aim for the middle by default.

I would either get close to the deep infrastructure of the giant rooms, where the hard constraints live: chips, compilers, runtimes, robotics, energy, security, regulated systems.

Or I would build toward a small business with a real edge: local trust, physical skill, taste, maintenance, craft, domain knowledge, a niche where customers can tell the difference between alive and automated.

The dangerous place is polished mediocrity. The agency whose work is mostly decks. The SaaS wrapper with no distribution. The team whose main asset is that clients have not yet realized a senior operator with agents can do 70 percent of it.

That sounds harsh. It is harsh. The economy is not famous for being sentimental.

The optimistic version

There is a version of this future I actually like.

Big companies build the impossible infrastructure: chips, models, power systems, medical tooling, logistics, aerospace, climate hardware, serious robotics. They have to be large because the coordination burden is real.

Tiny companies use that infrastructure without being swallowed by it. A two-person studio can make beautiful things. A local repair shop can run like a grown-up operation. A small lab can sell a precise tool to a precise market. A family business can spend less time drowning in admin and more time doing the thing people pay them for.

That future has fewer fake middle layers.

It also has less forgiveness. People who used to hide in coordination work will be exposed. People with taste, courage, and real skill may get a strange new kind of leverage.

I do not know which side wins. I do not even think “wins” is the right word.

I just know the shape keeps appearing.

A tower with five people inside.

A workshop with two people and ten invisible employees.

And between them, the narrowing place where a lot of careers used to fit.