← Back home

Fewer Roles. Heavier Ones.

Listen to this article
--:--

In 1776, Adam Smith described a pin factory. One man draws out the wire, another straightens it, a third cuts it, a fourth points it. Eighteen distinct operations. Ten workers, each doing one narrow thing, made close to fifty thousand pins a day. Working alone, he wrote, each of them might struggle to make twenty.

That idea built the modern economy. Later, it built the modern software company. Designer, product manager, frontend engineer, backend engineer, QA. Each one a station on the line. Each one passing the work to the next.

We split the work because skills were expensive. Learning to design took years. Learning to build took years. Nobody could hold all of it in one head, so we divided it and accepted the handoffs as the price.

AI changed the first half of that equation. It did not touch the second.

Earlier this year I wrote that one multidisciplinary person can now build products alone. Since then I have built a city simulation, a chess game and a real-time strategy game, mostly in the evenings. The work didn't disappear. It stopped requiring a team.

That was the individual. The organisation is next.

When a designer can ship the frontend herself, the handoff from design to engineering is no longer a necessity. It is a cost. And handoffs were always expensive: each one loses context, adds a queue, and spreads accountability a little thinner. When building was hard, that cost was buried under the cost of building. Now building is cheap, and the seams are most of what is left.

So the roles collapse. Not because people suddenly became more talented, but because the seams now cost more than they save.

The designer becomes a designer who builds. The product manager stops writing specs for others to interpret and starts living inside the product: prototyping, using it every day, finding what is broken before anyone writes a comment about it. Engineering splits in two. Product builders own a feature from idea to production. A thin platform layer underneath owns the ground everyone stands on: architecture, security, performance, data.

And the roles whose whole job was checking other people's work, QA most obviously, start to disappear.

This is the part most people get wrong.

Handoffs were waste, but not only waste. Each one was also a second pair of eyes. The engineer who read the spec and said this won't work. The designer who saw the build and said that's not the flow. The reviewer who asked why. Some of that friction was judgment.

QA is the clearest case. Deming made the point decades ago: you cannot inspect quality into a product, you have to build it in. A separate QA department institutionalises the opposite belief, that quality is someone else's job. As a role, it should go. As a function, it has never been more needed.

AI produces more work, faster, with fewer human eyes per line than ever. A developer who wrote two hundred lines could explain every one of them. A builder who generated two thousand in an afternoon cannot. More gets shipped. Less gets understood.

Collapse the roles without rebuilding the checks, and you get the closed loop I described in How AI Actually Takes Over. One person frames the problem, builds the answer, and approves it. Approval becomes ritual. The signature survives. The decision doesn't.

So the question is not whether to collapse the roles. It is where the checks go when you do.

If the role disappears and the need stays, the function has to go somewhere. It goes to three places.

Ownership. One person owns the outcome. Not the design, not the code, not the ticket. The outcome: it works, people use it, it does not break. There is no wall to throw it over, because there is no one on the other side.

Machines as the second pair of eyes. The best reviewers and testers were never valuable because they followed a checklist. They were valuable because they looked at work they had not made, trying to find what was wrong with it. That stance can now be automated: tests written alongside the code, regression on every change, and a separate agent that never saw how the work was produced and is pointed at it to break it. The builder does not grade their own work. Something independent does.

Humans define what correct means. What good looks like is the first draft that matters: the acceptance criteria, the design principle, the definition of done. Let AI write those too, and nobody is verifying anything. They are just confirming. Whoever decides what "working" means decides whether it works.

There is one more requirement, quieter than the others. A collapsed role only works if the person in it optimises for the whole. When I named Polis, I borrowed a colleague's contrast between the polis and the idiōtēs, the private individual concerned only with their own affairs. A builder who owns design, code and quality but ignores support, onboarding or performance is an idiōtēs with better tools. The point of collapsing the roles is that the whole becomes everyone's job.

Here is the uncomfortable part.

Role collapse does not transform people. It selects them. And the selection is not mainly about skills. AI closes much of that gap. It is about appetite for accountability.

Many people chose specialist roles partly because the boundaries protected them. "Not my job" was a feature: you did your part, passed it on, and the outcome belonged to the system. Collapse removes that hiding place. No tool can make someone want to own the whole result.

For those who do not want that deal, three paths remain. Go deep: the platform layer concentrates expertise, so some specialists become more valuable, not less. Become the independent eye, the critic who checks work they did not make, though machines are taking much of that seat. Or leave. What shrinks is the middle: the comfortable station on the line.

Nobody needs to announce any of this. The work decides, by contrast. When one person ships in a day what used to take a chain of four people a week, the comparison is visible to everyone. Selection by contrast is quieter than selection by decree. It is also harsher.

Which puts a burden on leaders. The tempting move is kindness: keep people in roles that no longer have a reason to exist. But a role kept alive for comfort becomes a ritual role, a job whose only function is to approve. That is not kindness. It is a slow cruelty, to the person and to the organisation.

What leaders owe people is different. Honesty about where the work is going, early enough to act on it. Real time and support for those who want to make the jump. Respect for those who choose depth instead. And a fair exit, not a quiet sidelining, for those who don't. The jump is the individual's. The transition is the company's responsibility.

I see this in our company. The realisation is everywhere: people know the shape of the work is changing. The inertia is just as real. Not because anyone is lazy, but because the change asks for painful personal choices, like giving up a role you are good at, or owning what you used to hand off. And because of fear: that something will break in production. That fear slows everything down.

The fear is not irrational. It is the correct response to removing seams without rebuilding the checks. People sense that the safety net lived in the handoffs, so when the handoff goes, they hold on tighter. That is why the checks are not a footnote to this change. They are what makes it possible. Nobody lets go of the old structure until they trust the new one to catch what falls.

Understandable or not, inertia has a cost. As I argued in What's Left to Defend?, inertia is not loyalty. It is not strength either. An org chart is a set of assumptions about what is expensive. Those assumptions changed. The companies still running the pin factory will find every station busy and nothing moving.

The ones that collapse the roles will find something harder. The work of the old roles does not vanish. It moves into fewer people, each now carrying the design, the build and the judgment at once.

Fewer roles. Heavier ones.