The PM Job Didn't Die. It Split Into Three.

"Product management is dead" is the most reliable headline in tech. It runs every year. And every year the people writing it still, somehow, have PM jobs.

So let me offer the more useful version of the claim. The generalist PM — the one who did a little discovery, a little spec-writing, a little stakeholder-wrangling, and a little roadmap theater — is dissolving. Not because the work vanished. Because it split into three jobs that each go deeper than the generalist ever could.

Why the generalist existed in the first place

It helps to remember where this role came from. The PM title traces back to a 1931 Procter & Gamble memo about "brand men," then got its modern shape at Microsoft, where "program managers" existed to coordinate across people who didn't otherwise talk. From the start, the job was glue. Someone had to hold the context and translate between design, eng, sales, and leadership. That coordination justified the role on its own.

Here's the thing nobody likes to say: AI ate the cheap half of that job. Synthesis, summarization, status updates, first-draft specs — the connective tissue that used to fill a PM's calendar — is now a prompt. What's left is the expensive half. And the expensive half is too demanding to do well across the whole surface. So it specializes. The way medicine did once it stopped being one guy with a bag and some leeches.

Role one: the systems thinker

This is the PM who stopped thinking in features and started thinking in feedback loops.

They don't ask "what should we build next." They ask "what is this system doing, where is value leaking out of it, and which single intervention changes the shape of the whole thing." They're comfortable with second-order effects, with metrics that openly fight each other, with the deeply unfun idea that the obvious feature often makes the system worse.

If you've ever watched a team ship exactly what was requested and watch retention drop anyway, you've watched the absence of a systems thinker. This is the role that finally becomes the strategist the generalist always claimed to be on the org chart but never had a free afternoon to actually become. It's Marty Cagan's "empowered team" lead, minus the busywork that used to swallow the calendar.

Role two: the eval owner

This is the new one. It's also the one most job descriptions are still missing entirely.

When your product is probabilistic, defining what good means is a whole job. The eval owner builds the test sets, writes the rubrics, owns the number that says whether the AI feature is actually working or just demoing well. They're the person who can tell you the model regressed last Tuesday — and then prove it with receipts.

Think of it as QA's smarter, more powerful descendant. Three years ago this barely existed. Now it's the line between a team that ships AI confidently and one that ships and prays. I'd argue it's the single highest-leverage PM specialization on the board right now, precisely because almost nobody is good at it yet.

Role three: the orchestrator

This is the PM who directs the work instead of describing it.

When implementation is cheap and fast, the constraint moves to who decides what gets built and who verifies it came out right. The orchestrator runs the loop: frame the problem, scope the task tightly enough that it won't wander, hand it to people or agents, check the output against the bar, iterate. They're measured on throughput and taste, not documents produced.

This is the role that looks least like the old job and most like what happens after the building part gets solved. It rewards agency over credentials, because the entire job is deciding and verifying, at speed, all day.

---

So which one are you becoming?

Most PMs are still trying to be all three at 30% depth, because that's exactly what the generalist role trained them to do. That's the trap. The people pulling ahead picked one and went deep.

So pick. Are you the person who sees the whole system, the person who defines what good means, or the person who runs the loop fastest? You can be excellent at one and competent at a second. You cannot be excellent at all three, and the quiet pretense that you can is the actual thing that's dying.

The title on your badge will say "Product Manager" for a few more years. The job behind it already changed.