At a conference earlier this year, someone asked whether AI would make product managers obsolete.
Half the room laughed.
The real answer isn't "no" — it's "some of them, yes."
The PM market is splitting into a K-shape. The upper half is growing: PMs who orchestrate agents, genuinely understand customer problems, and make hard calls about what gets built. The lower half is shrinking: PMs who write Jira tickets, collect status updates, and "coordinate" between engineering and design.
The second group is already automatable. It's uncomfortable to say directly, but it's accurate.
"What to build" has always mattered. But the "how to build it" side was heavy enough that most PMs spent most of their time there — sprint planning, backlog grooming, handoff docs, capacity math. We convinced ourselves this was the real job.
Now that machinery is moving to automation. So what's left?
Judgment. Intuition. The ability to understand what a customer actually means when they say something.
Which is, if you think about it, what the job was supposed to be from the start.
I manage four products — DentalBulut (dental clinics), Medibulut KYS (general healthcare), Diyetbulut (dietitians), and a marketing site. All four run on the same codebase. Same code, different audiences.
When a feature request comes in, I have to ask: which product does this actually belong to? Do dentists benefit? Will dietitians even use this? What about general practitioners?
I can ask AI for input. I'll get a reasonable answer. But when I ask it to decide — there's a gap. Because the decision isn't just technical. It holds the segment's culture, the customer's workflow, the sales team's current pressure, and this month's company priorities all at once.
I can't fully explain that context to anyone. Not to AI, not to a new team member.
That gap is where a PM still exists.
a16z published something last week called "5 Principles for Product Managers Fending Off Obsolescence in the AI Era." A bit doom-y as a headline, but the core idea is right: as execution shifts to automation, the PM's value concentrates around answering "what should we do?" correctly.
That's not wrong. But it's also deceptively hard. Answering that question well requires time spent with customers, an understanding of how they actually work, and a track record of some things having gone wrong.
There's no shortcut for the reps.
There's also this: managing AI agents is itself becoming a PM-shaped job. Telling a coding agent, a research agent, a support agent what to do — that's more "what should happen" than "how should it happen."
So the role is evolving, not disappearing. But it's concentrating. The gap between PMs who are building judgment and PMs who are maintaining spreadsheets is widening fast.
If you're in the lower half of that K and don't realize it yet — that's probably the more urgent problem.