The hourglass economy

I keep seeing the same organizational shape in job postings, product teams, and conversations with friends.

It looks like an hourglass.

At one end are very large companies. They own expensive infrastructure, distribution, customer relationships, and the legal machinery needed to operate at scale.

At the other are very small firms: two or three people, a narrow product or craft, and software handling much of the administrative work.

Between them sit agencies, regional software companies, consultancies, and internal departments. These organizations have long offered a stable middle ground between working alone and joining a giant corporation.

My concern is that AI strengthens both ends while putting pressure on that middle. This is a hypothesis, not a forecast, but the incentives are becoming easier to see.

Large companies with smaller operating teams

AI systems are expensive to build and cheap to reuse. That favors companies able to buy compute, gather data, pass regulatory reviews, and distribute a product to millions of customers.

The less obvious effect is inside those companies. A small, experienced team can now produce drafts of code, designs, tests, analyses, support replies, and documentation that once required more people. Human review still matters, and the final work is not automatic, but the amount one person can supervise has increased.

This does not make large organizations empty. Infrastructure, sales, operations, security, and support still require many people. It can, however, reduce the number of people attached to a particular product surface.

That creates unusual jobs. A few operators hold more responsibility, rely on more automation, and control systems that reach far beyond their team size. They need strong judgment because errors also travel farther.

The leverage looks attractive, but it belongs partly to the employer. The models, data, deployment platform, contracts, and customer trust stay with the company. An employee may control a large surface while remaining replaceable inside it.

Small firms with less overhead

The other end of the hourglass is easier to like.

Consider a two-person design studio. One person handles clients and creative direction; the other handles production. Software can draft proposals, sort invoices, organize meeting notes, update the website, and prepare routine files. The owners still make the decisions and do the work customers care about.

The same applies to a bakery, repair shop, specialist consultancy, or small robotics company. None becomes an “AI business.” It simply spends fewer hours on scheduling, paperwork, quoting, documentation, and basic customer support.

Administrative work used to force small companies to hire before demand justified it. Better tools let them delay that step or avoid it entirely. A tiny firm can look organized without pretending to be large.

There are costs. Owners may be expected to answer faster, produce more, and operate alone for longer. Automation can turn a small business into one person watching six dashboards at midnight. Better leverage does not guarantee a better life.

Still, short accountability has value. If the chair is badly made, the customer knows who made it. If the software fails, the person who shipped it is close enough to hear about it. AI can reduce the surrounding paperwork without removing that relationship.

Why the middle is exposed

Many mid-sized organizations exist because coordination is difficult.

An agency coordinates specialists for clients who need more than one freelancer but less than a global vendor. An internal IT department translates between employees, software vendors, security rules, and old systems. A consultancy turns scattered expertise and documents into decisions.

AI is good at some of this connective work: first drafts, summaries, routine analysis, boilerplate code, test generation, document search, and status reporting. It is not good at all of it. Politics, ambiguous requirements, legacy systems, and angry customers remain stubbornly human.

It may not need to replace all of the work to change the economics. If tools remove enough routine effort, a client may hire a smaller specialist team. A company may buy a platform instead of renewing an agency contract. A department may leave vacancies unfilled because its senior staff can cover more ground.

The middle would not disappear. It would thin through attrition, consolidation, and narrower teams. Some firms will adapt by developing deep expertise or owning a customer relationship that software cannot easily reproduce. Generic coordination work will have a harder time.

The real competition is organizational

“Humans versus AI” misses the more practical question: who captures the productivity gain?

Large companies can use AI to operate existing infrastructure with fewer bottlenecks. Small firms can use it to gain capabilities they could not previously afford. Both structures have a clear reason to adopt the tools.

The middle needs a stronger answer. Size alone is no longer enough. A company must own something specific: trusted access to a market, difficult domain knowledge, infrastructure, a distinctive product, or people whose judgment customers can recognize.

Otherwise, it risks being squeezed between a platform that is cheaper and a small expert team that is more direct.

Where I would place my own bet

I work in AI systems, so I am part of this machinery. I enjoy runtimes, kernels, robot policies, and the moment a complicated system finally works end to end.

For my own career, I want to stay close to constraints that remain difficult: chips, compilers, robotics, energy, security, and regulated systems. These fields combine software with physics, hardware, or responsibility that cannot be reduced to a convincing first draft.

The other credible path is a small business with a real edge: local trust, physical skill, taste, maintenance, or deep knowledge of one customer’s problem. Customers need to be able to tell why the work is different.

This is not universal career advice. Large organizations can be excellent places to learn, and many medium-sized companies do valuable work. The point is to ask what protects the organization besides its current headcount.

A version worth hoping for

The hourglass is not necessarily a bad outcome.

Large companies can build infrastructure that genuinely requires scale: chips, models, power systems, medical tools, logistics, aerospace, and serious robotics.

Small companies can use that infrastructure without becoming departments of it. A studio can spend more time designing and less time invoicing. A repair shop can run reliable operations without adding a back office. A small lab can sell a precise tool to a precise market.

The risk is a narrower path between those two ends, with fewer forgiving places to build an ordinary career.

I do not know whether the economy will take this exact shape. I only know that the question now follows me into meetings: is this organization building something that needs its size, or is its size mostly paying for coordination that software is learning to absorb?