The product manager role isn't disappearing. The generalist middle is.

A May 2026 report from Products That Count and Mighty Capital surveyed more than 1,500 product leaders and produced a headline built for panic. Traditional product manager roles are down 30% across industries, and as much as 70% in SaaS. The role will be obsolete by 2030. That is the part that got quoted everywhere.

The same report has a second number that got quoted almost nowhere. Product Builders, the hybrid product and engineering role, are up 10x. Forbes put real headcount on it a few weeks later: roughly 600,000 to 1.7 million. So the work did not vanish. It moved into a job with a different name.

I read job postings every day, because I built an agent that surfaces them for me. The split shows up in the raw listing text before any analyst names it. One posting wants eval design, model selection, and the cost math on every API call. The next wants stakeholder alignment, roadmap sequencing, and a number you owned end to end. Five years ago those were the same job with different emphasis. Now they read like two careers sharing a title.

The cause is not complicated. AI took the expensive part out of building. When a working prototype takes an afternoon instead of a spec and a sprint, being the person who can describe what to build stops being scarce. Larry English put it plainly in Forbes: deciding which opportunities deserve attention becomes the harder challenge. A PM whose main value was translating between business and engineering loses ground on both sides at once.

Userpilot names the two sides better than I would. The Builder-PM ships the prototype themselves. The Integrator-PM owns the roadmap and the relationships inside a complicated B2B environment. Both are real jobs. Both pay. The problem is that most people have not picked one.

If you want the builder track, there is a concrete entry test, and it is not a course. Take one real product surface and write the eval for it. The dataset, the scoring rubric, the target pass rate, and the name of the person who owns the postmortem when it comes in wrong. I have written here before that evals are the new acceptance criteria. Writing one is how you find out whether you can actually do that job.

The integrator track has its own test, and it is less forgiving than it used to be. Own one number all the way through. The baseline before you touched anything, the specific thing you changed, and the movement you can defend in a room full of people who would rather it were bigger. Building a 0-to-1 analytics product taught me that a roadmap slide with no number on it is a guess in a nice font.

What does not survive either track is the undifferentiated middle. The PM whose main asset was a clean PRD and a well-run standup. AI writes the PRD now. It does not own a rubric with a defensible threshold, and it does not sit in the room when the number comes in wrong.

That said, 2030 is a marketing hook, not a prophecy. Roles this broad do not hit zero. They get reclassified, the way webmaster turned into six different jobs instead of disappearing. But the pressure is real, and it is already in the postings. The title will probably survive the decade. The version of it that does not specialize in anything will not.