The Role Everyone Is Cutting

Across the industry right now, one move is spreading fast enough that it has started to feel like conventional wisdom: thin out the product and project managers, hand both hats to the engineers, and call the leaner org chart a dividend of AI. The story is tidy. Engineers can build faster than ever, so surely they can absorb the deciding and the coordinating too. I have come to think it is one of the more expensive mistakes a leadership team can make this year, and the reason is a single word most of us stopped thinking about after our first operations class: constraint.

For most of the history of software, building was the constraint. It was the slow, costly, uncertain part, and whether we said it out loud or not, we organised whole companies around protecting it — the roadmaps, the sprint rituals, the careful triage of what engineering time got spent on. Then AI walked in and, more or less overnight, stopped building from being the bottleneck. And here is the thing about a bottleneck: when you finally remove one, the work it was holding back does not disappear. The pressure simply moves to whatever is now the slowest thing in the line.

So where did it go? It went to the two things AI did not touch. One is the product question: deciding what is actually worth building, and understanding a customer deeply enough to tell the difference between a request and a need. The other is the project question: taking a dozen suddenly-fast workstreams and keeping them aligned, sequenced, and actually delivered, instead of sprinting in twelve directions at once. Neither got cheaper when the code got cheap. Both got heavier, because there is now far more to decide and far more to coordinate, and far less time to do either. Those two jobs have names. We have been calling them product management and project management, and we picked this exact moment to start deleting both.

Gregor Ojstersek laid the numbers out well, and the most instructive one is not from the companies cutting PMs — it is from the one that arguably sits closest to the frontier. While much of the industry moves toward thirty engineers per product manager, or none at all, Anthropic is going the other way and hiring more, roughly one embedded in every cross-functional team and reportedly looking to add more still. When the people building the most capable models decide the scarce resource is product thinking rather than engineering hands, it is worth asking why the rest of us concluded the opposite.

The error underneath the trend is an old one in new clothes: confusing the roles with the work. You can absolutely delete the titles. You cannot delete the work they were doing — the deciding and the coordinating just get quietly redistributed to engineers who are now also expected to ship twice as fast, and who will, reasonably, optimise for the thing they are measured on, which is shipping. The decisions still get made and the projects still get run. They just get made faster, more thinly, and by people whose attention is pulled toward the build. You did not remove the cost. You hid it, and moved it downstream, to a quarter where the features were the wrong ones and the launch quietly slipped.

I have watched enough teams up close to know how this failure actually looks, and it is rarely dramatic. It is not a crash. It is a slow drift — a roadmap that fills with things that were easy to build rather than things that mattered, projects that quietly collide because no one is holding the dependencies, a team moving at genuinely impressive speed toward a destination no one has recently questioned. Motion is easy to celebrate, because you can see it. Direction and coordination are easy to lose, because for a while their absence looks exactly like progress.

So the reframe I keep offering the leaders I talk to is simple. When hands get cheap, aim and alignment get expensive. The organisations that come out of this era ahead will not be the ones that trimmed their product and project functions to look efficient on a headcount slide. They will be the ones who noticed the constraint had moved, and had the discipline to staff the new one instead of the old. AI did not make the questions of what to build and how to ship it together less important. It made them the only questions that are still genuinely hard.