The PM job is being rewritten. Most PMs haven't noticed.

In February 2025, Marty Cagan published a quiet bombshell on the SVPG blog. His prediction: the typical product team will compress from roughly eight people to three within a few years. A fifteen-team scale-up that runs at about $24M in fully-loaded headcount today gets to the same output for closer to $1.8M. He named the role most exposed by name. Product owners and feature-team PMs, in his words, are "very vulnerable."

That is not a hot take from a futurist. It is from the person who wrote the book most PMs were handed on their first day.

The reflex response in PM circles has been comforting and, I think, wrong. The standard line is that AI handles the busywork so PMs can finally focus on strategy. It sounds right. It tests well in a LinkedIn post. The actual data tells a different story.

The "AI gives PMs time for strategy" line is mostly cope

Productboard and UserEvidence surveyed 379 product managers in October 2025. The headline number was real: PMs reported saving about 33 hours a month using AI tools. Then the study asked where the saved hours came from. The top categories were writing PRDs, producing competitive research, summarizing customer calls, building presentations, and drafting roadmaps.

For a senior PM at a mature company, those tasks are overhead. For a feature-team PM at a typical scale-up, those tasks are the job. When the work you were hired to produce becomes a one-prompt output, the question is not what you do with the extra time. The question is whether the role still justifies a seat.

Adam Judelson, who runs product at Palantir, put it more bluntly than most executives are willing to: the less technical parts of PM work are now done better by generative systems than by humans. He is describing his own hiring bar.

The work that survives is judgment, not process

Lenny Rachitsky ran his own skill-mix survey in August 2024 and revisited it in 2025. The skills PMs and their managers now rank as most valuable are data literacy (58%), synthesizing customer insight (54%), systems thinking (53%), and strategic prioritization (52%). Writing specs, running standups, and maintaining roadmaps all dropped.

This tracks with what is happening on the engineering side. Internal studies at Microsoft, GitHub, and several large banks now put engineering productivity gains from Copilot-class tools in the 20 to 30 percent range for routine work, with steeper gains on greenfield code. When delivery cycles compress, the binding constraint on a product org is no longer "can we build it." It becomes "do we know what to build, and do we know fast enough to be right about it."

That is a judgment problem. It rewards PMs who can sit with a messy customer call and pull out the one insight that reframes the roadmap. It punishes PMs whose primary contribution is artifact velocity.

PMs who are thriving

The doom framing is easy. I believe it is also incomplete. A small group of PMs has already adapted, and their work is worth studying because the pattern is consistent.

Shreyas Doshi, formerly of Stripe, Twitter, and Google, has spent the last two years writing publicly about using LLMs as a thinking partner rather than a drafting tool. His pre-mortems, strategy memos, and product critiques use models to stress-test arguments, not to generate them. The leverage shows up in the quality of his decisions, not the volume of his output.

Lenny Rachitsky himself is the cleanest example of compounding through AI. One person runs the newsletter, the podcast, the job board, the community, and the course platform. The operation would have required a team of ten a decade ago. He uses AI across research, editing, transcription, and analytics, and the business has grown into one of the largest independent product media properties on the internet.

Claire Vo, former CPO at LaunchDarkly, went a step further. She built ChatPRD, an AI tool now used by tens of thousands of PMs to draft and pressure-test product documents. She did this while operating as a senior product leader. PMs who build with AI rather than just consume it move from being users of the new tooling to authors of it, and the career ceiling is meaningfully higher there.

Aman Khan at Arize AI represents the new archetype most hiring managers actually want when they post an "AI PM" role. He writes openly about eval design, agent observability, and the messy work of shipping LLM features that behave reliably in production. The job is half product, half applied ML, and the people doing it well were almost all conventional PMs eighteen months ago.

Ravi Mehta, formerly CPO at Tinder and a product leader at Tripadvisor and Facebook, has turned his AI-augmented workflows into an advisory practice. Senior PMs who can articulate how AI changes the operating model of a product org, with specifics, are commanding the kind of rates that used to belong only to former VPs of Product.

The pattern across all five is consistent. They treat AI as a force multiplier on judgment they already had, and they invest in the parts of the craft that do not commoditize. None of them spend much time worrying about prompt libraries.

What to do if you are a working PM

My honest read on the next two years is that the median PM job description will change faster than most PMs change with it. Compensation will bifurcate. The PMs whose value sits in producing artifacts will compete with a Claude subscription. The PMs whose value sits in calling the right shot in an ambiguous room will be more in demand, not less.

If you are doing the work, pick one domain and go deep enough that a general-purpose model cannot retrieve what you know. Get fluent enough with the current generation of tools to evaluate their output critically, which is a higher bar than using them. Take the problems no model can define for you: the multi-stakeholder negotiations, the decisions made on incomplete data, the calls that require sitting with real risk. That is the work that still pays.

Cagan's three-person team is not a threat. It is a description of who gets to be on it.

Sources: Marty Cagan, "Product Teams in the Age of AI," SVPG, February 2025. Productboard and UserEvidence, "The State of AI in Product Management," October 2025 (n=379). Lenny Rachitsky, "What product managers actually need to be great," Lenny's Newsletter, August 2024. Adam Judelson, Palantir, public remarks 2025. Shreyas Doshi essays at shreyasdoshi.com. Lenny's Newsletter at lennysnewsletter.com. Claire Vo at chatprd.ai. Aman Khan at arize.com. Ravi Mehta at ravi-mehta.com.