The PM Role Is Dying: Why the Work Isn't

I've written about AI across engineering roles — No Junior Engineers?, The Senior Engineer's AI Trap, The AI-Augmented Engineering Manager, and How Ops Engineers Can Stay Relevant. I kept skipping product managers.

At work, my team has a PM. I kept thinking I was the wrong person to write about a role that belonged to someone else. Then I started building my own projects on the side. I noticed I was spending more time on product decisions than on implementation.

That was the point.

The Wrong Frame

"The PM role is dying." That's been the story since AI coding tools went mainstream. Some companies are quietly shrinking PM headcount. The coordination-heavy version of the job is getting automated away.

That framing is wrong.

The PM role isn't dying. It went everywhere.

What I Actually Do With My Time

My personal projects run on OpenSpec + Superpowers. The model: I write specs. Claude builds. I review. Claude fixes. Implementation runs itself.

What I spend time on: why this feature and not that one. Who it's for. What the UX should feel like at each step. Whether agent behavior matches the intent. Where the gaps are between what I specified and what got built.

That's product work. All of it.

My title is engineer. Since late last year, I've been my own PM.

What's Actually Disappearing

The PM role that's shrinking is the coordination version.

Nikhyl Singhal put a name to it in Why Half of Product Managers Are in Trouble | Nikhyl Singhal (Meta, Google): Information Movers and Builders. Information Movers coordinate — translating requirements into specs, aligning teams, managing roadmaps across quarters. Builders define what gets built and can validate it quickly.

"The information mover is going to be a dinosaur." AI handles summaries, status updates, alignment documents, roadmap maintenance — faster and cheaper than a full-time headcount. Some companies have already stopped backfilling these roles.

The Builder PM is fine. Better than fine.

The Anthropic Signal

Anthropic runs hackathons. The Opus 4.6 hackathon and Opus 4.7 hackathon results had the same unexpected pattern: the winners weren't engineers. A personal injury lawyer. A cardiologist. A road engineer from Uganda. An electronic musician.

Four out of five Opus 4.6 winners had no software engineering background. They beat 500 selected participants — including professional developers. Not by coding better. By knowing something engineers didn't.

The lawyer knew exactly where California architectural permits fail — 90% get rejected for corrections before approval. The cardiologist knew which parts of residency training break down. AI handled implementation. Domain expertise determined what was worth building.

Cat Wu, the PM behind Claude Code, was asked how she hires. Almost everyone on her team came from engineering or writes code directly. She came from engineering herself. Her definition of the irreplaceable PM skill: product taste. The ability to judge which problem is worth solving, and what good looks like when it's solved. Not coordination. Judgment.

Three Things Happening Simultaneously

These aren't sequential stages. All three are happening right now, in different companies, at different speeds.

Path 1: The coordination PM disappears. AI handles status, alignment, and documentation. Companies quietly stop backfilling these roles. This is already underway.

Path 2: PMs become prototype machines. Vibe coding tools let non-engineers build working demos in an afternoon. The ceiling is prototyping, not production. But closing the gap between idea and testable artifact within a day is a real capability jump.

Path 3: Engineers absorb product thinking. The Anthropic model. Engineers with product taste ship end-to-end. The PM function moves inside the engineering role. This is what I've been doing, without naming it.

What's Worth Keeping

The coordination work is going away. That part's not a prediction anymore.

If you've been a PM for ten years, that's a hard thing to sit with. The skills that got you here — aligning stakeholders, writing PRDs, managing roadmaps — are exactly what's getting automated. But the core of the job isn't.

Two things AI consistently can't do: develop product taste, and carry deep domain knowledge.

Product taste is what Cat Wu called the rarest skill in the room — the ability to judge which problem matters, which solution is right, which tradeoff is worth making. It grows from time with users, from shipping things, from being wrong about what users want and noticing. The hackathon winners had something close to it: deep domain knowledge. A lawyer who knew exactly where permits fail. A cardiologist who knew which parts of residency training break down. That knowledge is the moat. AI amplifies it — it doesn't replace it.

Three more things worth building:

Spec thinking. Defining a problem precisely enough that a builder can execute on it. Not writing PRDs — making decisions. Exactly what to build, for whom, measured how. Implementation is getting cheaper. Knowing what to implement is getting more valuable.

Build to validate. Not production code. A testable artifact in front of a real user within a day. Vibe coding makes this accessible. Ten experiments a week beats one perfect quarterly roadmap.

Calibrated model expectations. Cat Wu called it "the right amount of AGI pilled." Building for current model capabilities while staying a step ahead. Too conservative and you miss what's already possible. Too optimistic and you design for a model that doesn't exist yet.

What to let go of: treating cross-team alignment as the primary output. Multi-quarter roadmaps as the unit of planning. PRDs as proof of value.

The Title Is a Lagging Indicator

Job titles change slowly. The work changes fast.

PM headcount will keep shrinking. The companies reducing it aren't wrong — the coordination layer genuinely needs fewer people now. But the product thinking layer — defining what to build, judging what good looks like, carrying domain expertise — that's expanding, not shrinking.

It's expanding into engineering roles. Into technical program management. Into anyone who builds with AI.

The PM role didn't die. It lost a job title and found ten others.