The PM isn't dying. It's returning to what it always should have been.
Keith Rabois said it on Lenny's podcast a few weeks ago. "The idea of a PM makes no sense in the future." The internet ran with it.
He's not entirely wrong.
The version that's dying is the PM as information mover. Translating business intent into specs. Managing the engineering queue. Running sprint planning. Bridging people who wanted things and people who built them. A lot of PMs spent most of their time on this — not because it was their calling, but because someone had to.
AI does it better now. That's fine. Brian Armstrong wrote a letter to a new Coinbase PM more than a decade ago that I keep coming back to. He listed the operational stuff — roadmap, sprint planning, reviews — as the baseline to get through. Then said:
Once you master all of that, you will need to develop product vision… and strive to make something truly great.
Brian Armstrong
The scaffolding was never the job. It was what the job required before you could do the actual job.
What I'm seeing right now isn't a role dying. It's a bottleneck moving.
A founder I spoke with recently put it plainly. His developers were shipping four times faster than eighteen months ago. Five engineers, one PM, one designer. The engineers were fine. The product judgment wasn't keeping up. He wasn't complaining about speed. He was complaining about something harder to name — building a lot, deciding too little, losing coherence.
That's the new constraint. Not velocity. Judgment.
And I know this from the inside. I heard a version of this constantly at Multis. Our PMs never had enough time for the work they actually cared about — discovery, sitting with users, thinking hard about what to build and why. The logistics ate everything. The cruel irony was that they were too busy managing the product to actually think about it. What AI hands back isn't just efficiency. It's the time that was always supposed to go toward judgment. If the job was meant to be 80% thinking and 20% logistics, most PMs were living in an inverted world.
When the cost of shipping approaches zero, the cost of shipping the wrong thing approaches everything. No engineering bottleneck to blame. No sprint capacity excuse. If you built the wrong thing, it was a judgment failure. Pure and simple.
The judgment function doesn't disappear when AI takes over the scaffolding. It becomes the whole game.
Some people have product taste. Others genuinely don't. It's not a skill you train in a workshop. It's the ability to look at something half-built and know it's wrong before you can articulate why. You can't prompt your way to it. It lives in the person.
Customer obsession is the other thing that stays — not in a values-poster sense, but in the sense that the best PMs I've known were genuinely troubled by their users' problems. Not just informed. Troubled. AI narrows the information gap. It doesn't close the empathy gap.
At Multis, the judgment calls I got wrong share the same root. We built things because we could. Because the technical lift wasn't too heavy, because a customer had asked, because it felt like progress. Not because we had genuine conviction it was right. In a world where building is nearly free, that's a much more expensive mistake.
What AI is doing to the PM role isn't elimination. It's purification. The scaffolding falls away, and what's left is what always mattered: taste, judgment, obsession, the discipline to say no when saying yes costs nothing.
There's a distinction worth making here. What's actually dying is the Product Owner — not the Product Manager. The PO was always the logistics role: backlog grooming, ticket writing, sprint ceremonies. It made sense when the machine needed heavy tending. It needs less now. The PM — the one who thinks like the CEO of the product, who makes the hard calls, who holds the product coherent when everything is pulling it apart — that role isn't shrinking. It's finally getting the room it always deserved. For great product people, this isn't a threat. It's the first real chance in a long time to do the job properly.
Rabois is right that the title might not survive in its current form. I think he's wrong that the function disappears. Teams are moving faster than ever. Someone still has to decide where they're going.
I'm curious whether any of this will hold up. What felt true about AI six months ago can already look naive today. Maybe I'm wrong about all of it — things are moving that fast. But we're living in extraordinary times. I'm sure of that much.