AI Needs Departments, Not Superheroes

The popular picture of AI is a lone genius. Real organizations have never worked that way, and neither will their AI.

In the last article, I argued that a single AI answering a question is rarely enough to finish the work. Once you accept that, a harder question follows.

If one AI isn't enough, what is?

The instinctive answer, the one baked into most of how we talk about this technology, is to build a bigger one. A single system that knows everything, sees everything, and can do everything. The all-knowing assistant. The superhero.

It's a compelling image. It's also the wrong one.

The superhero we keep imagining

Watch how AI is usually described and you'll notice the same shape appearing again and again.

One system. One prompt. One mind that reads your contract, checks your inventory, understands your finances, drafts your email, and closes your deal. The user speaks, and a single intelligence handles the rest.

It's the model of a hero. Brilliant, tireless, endlessly capable, and entirely alone.

I understand the appeal. It's simple to imagine and easy to sell. But I've never seen a real organization that works this way, and I don't think that's an accident.

No company hires one extraordinary person and asks them to do everything.

How work is actually organized

Walk into any functioning business and you'll find the opposite of a superhero.

You'll find departments.

Marketing has people who understand campaigns, and different people who understand analytics. Finance has someone who handles accounts payable and someone else who handles compliance. Operations has specialists for procurement, for logistics, for quality. Legal reads the contract. Security reviews the access.

Nobody expects the person who runs payroll to also negotiate the supplier agreement.

We organized ourselves this way for a reason. Real work is too broad and too deep for any single mind to hold at expert level. So we divided it. Each person became genuinely good at one part, and we built structures to move work between them.

A specialist who knows one domain deeply will almost always outperform a generalist who knows everything shallowly.

This isn't a limitation we tolerate. It's a design we chose, because it works.

Someone still has to hold the whole

Specialists alone don't finish anything, of course. A room full of experts who never coordinate produces very little.

So organizations added something on top.

Someone who understands the goal. Someone who breaks it into parts, decides who does what, sequences the steps, and carries the outcome. We call them managers, coordinators, project leads. Their expertise isn't any single domain. Their expertise is the whole.

Think about how an insurance claim actually moves.

Someone intakes it. A specialist assesses the damage. Another checks the policy for coverage. Finance calculates the payout. A reviewer confirms it all holds together before money moves. And somewhere in the middle, when the case is unusual or the number is large, a human being makes a judgment that no rule could make for them.

No single person did all of that. A coordinated group did, each contributing their part, with someone making sure the work moved cleanly from one hand to the next.

That is what finishing work looks like. It has always looked like this.

The shape this suggests for AI

Once you see it, the pattern for AI stops being mysterious.

If organizations don't rely on a single all-capable person, why would we expect them to rely on a single all-capable model?

A more natural shape mirrors the organization itself.

At the center sits something that reasons about the goal. It reads the request, understands what "done" actually means, and plans a path to get there. It doesn't try to do everything. Its job is to understand and to coordinate, the way a good manager does.

Around it sit specialists. One that knows how to work with the financial system. One that understands the customer records. One that handles documents, or scheduling, or a particular regulatory check. Each is good at its narrow part, and none pretends to be good at all of them.

Somewhere in the flow there is a check. A step that reviews the work before it becomes real, that asks whether the numbers reconcile and the policy was followed, the way a second set of eyes catches what the first missed.

And at the points where judgment genuinely matters, there is a human. Not to babysit the machine, but to make the calls that should belong to a person, exactly as a manager steps in only when the case demands it.

The coordinator understands. The specialists execute. The check verifies. The human decides where deciding matters.

What comes out the other end isn't a reply.

It's an outcome.

Why the team beats the hero

There's a deeper reason to prefer this shape, and it isn't only about capability. It's about accountability.

When a single system does everything and something goes wrong, you often can't tell what happened. The refund was miscalculated, or the wrong policy was applied, or a step was skipped, and all you have is one confident output that turned out to be wrong. The failure is silent. It's unaccountable. There's no seam to point to.

A superhero who fails leaves you nothing to inspect.

A team is different. A team is legible.

When work moves through a coordinator, specialists, a review, and a human, you can see it. You can see which specialist handled which part. You can see where the check passed or failed. You can see the exact point where a person made a decision and what they decided.

If something breaks, you know where. And knowing where is most of knowing why.

This is how we already run organizations. We don't trust important work to a single unaccountable actor. We distribute it, we make it observable, and we place judgment where judgment belongs. We do this not because our people are weak, but because legible systems are the only ones we can actually trust at scale.

The pattern we already know

The more I think about it, the more I believe we've been overcomplicating the future of enterprise AI.

We keep searching for a smarter hero when we already invented something better a long time ago.

We invented the organization. A structure where specialists do what they're best at, coordinators hold the goal, review catches the errors, and humans decide what only humans should. It is one of the most reliable machines humanity has ever built, and it works precisely because no single part carries the whole.

The emerging generation of enterprise AI, the one people have begun to call the autonomous enterprise, looks far less like a lone genius than like this. Not one mind that knows everything, but a coordinated group that finishes something. The most honest ongoing account of this shift is worth following as it develops.

The superhero was always a fantasy. Capable, yes, but opaque, unaccountable, and alone.

Real work has never been done by heroes.

It's been done by teams.

In the next article, I want to look more closely at the member of that team we're most tempted to leave out, and least able to afford to: the human, and the specific kinds of judgment that should never be automated away.