Which Product Manager Is AI Replacing?

Your company has three different jobs called product manager, and AI is about to treat them very differently. One of them spends the week moving work through delivery. One of them manages a backlog for a sprint. The third decides what is worth building. Agents are already good at the first two.

That is why the question is the PM role dead? has two loud camps, both convinced they are right. Some leaders say the role is finished. Others say it has never mattered more. Marty Cagan wrote on 6 November that each very likely assumes the other is clueless, and that they are talking about different jobs. I agree, and I would add that the line between those jobs runs through each company.

Three jobs, one title

Cagan’s point is that the operating model decides the job, not the title. A company that funds projects and features needs people to get those features shipped. A company that hands teams a problem and an outcome needs people to find the solution. He counts three product roles:

• The feature-team product manager, who coordinates the design, build, test and launch of a solution someone else already chose.
• The product owner, a role inside Scrum and SAFe delivery, who manages the backlog and works with engineers before, during and after each sprint.
• The empowered-team product manager, who is given a problem and a desired outcome, and has to discover a solution that works for users and for the business.

The first two are mostly coordination. The third is mostly judgement.

Product owners should have gone years ago

Cagan says the product owner role is already being phased out. I would put it less politely. It should have gone years ago, together with the rest of the hard framework agile industry: Scrum, SAFe and the certification industry built around them.

The product owner was a handoff dressed up as ownership. The title promised that someone owned the product. The job was to translate other people’s decisions into tickets, groom a backlog and keep a sprint fed. Many of the people in the role were good, and the best ones were already doing discovery work the title didn’t describe. But the role existed because the process needed a person at the join between the business and the engineers, and agents now sit at that join for free. AI is only finishing a job the industry should have done itself.

Coordination is going

The feature-team product manager is next, and the evidence is already public. Figma’s PMs run a shared Claude Code toolkit they call PMOS, kept in sync across 60 to 70 PMs. Ezra Mechaber, the PM who built it, told How I AI: I like to tell my engineers that my goal is to be lazy in the right places. The toolkit finds the right stakeholders for a review, pulls status from several Slack channels and tools into one daily digest and makes sure support hears about a launch. He describes it as raising the floor: we have to do just a little bit better on our baseline.

That is the coordination work, and it is the same shift I described for middle managers: routing goes to the machine, and judging stays with people. A PM whose week is mostly routing is in the same position as a manager whose week is mostly routing.

The mood on social media has caught up. A post doing the rounds this month, seen 128,000 times, lists product owner and project manager among the cost centres that AI is eating up. The list is crude, but about these two roles it could be right.

Discovery grows

The empowered-team PM is the opposite case. Cagan puts it plainly: the bottleneck is no longer product delivery, the real question is determining what to build. Every hour an agent takes off delivery is an hour that should go to talking to users, testing ideas and killing the weak ones before they ship.

A company that removes its coordinators and keeps the project model gets features faster, without knowing whether they work, which is the discovery debt problem at a higher speed. The skills that matter are the ones AI didn’t replace, and the test is the one I set out in the real PM test: clear thinking when anyone can produce anything.

The same company also piles up measurement debt. The project model counts outputs: features shipped, dates met, tickets closed. It does not count outcomes, so nobody can say which of those features earned their keep. That matters more as AI makes features cheaper to ship and more expensive to run, because the PM is the person who has to answer for what each one makes and costs, and you cannot answer for profit with a scoreboard that only counts deliveries.

Why I haven’t decided yet

I haven’t changed a single PM role yet, and that is deliberate. It will happen. But this is a decision that is important rather than urgent, and I think things through for a long time, until the answer emerges. This one hasn’t yet.

Part of the reason is that the line runs through my own organisation. The teams I lead aim for the product model, and I let project elements in when they help, especially when we coordinate with teams outside product and with the businesses in our franchise network. I am a pragmatist. That means some of the same people do coordination on Monday and discovery on Thursday, and a headcount decision made from the job title would remove the wrong ones.

So before deciding, I am sorting the jobs, person by person, with three questions:

• What did they spend most of last week on: moving work through delivery, or deciding what should be built?
• If an agent took the first part, what would be left?
• Could they test an idea with users this month, and do they want to?

There is no prize for deciding first. Cutting before you have sorted the jobs removes the people who know your customers and keeps the ones whose work an agent already does.

The industry may finally rename the roles, as Cagan suggests. One new name is already in circulation, product engineer, and it is older than it sounds.

Before you ask whether you still need product managers, ask which job your product managers are actually doing. I’m still asking, and I’m in no hurry to stop.