The PM role in the AI era: more essential, not less.

For decades, the limiting factor in software was engineering bandwidth, and AI flipped that overnight. When one of Andrew Ng's teams recently proposed a 1:0.5 PM-to-engineer ratio — twice as many PMs as engineers — it was a radical break from the old 1:4–7 norm.1 The reason is simple: building sped up dramatically, and product thinking didn't.2

Melissa Perri names the structural catch: the classic 1-to-8 PM ratio was designed for a world where building was the constraint — and that world is gone. Make building 10 times faster without investing proportionally in discovery and strategy, and you simply ship 10 times more of the wrong things.3

As AI democratizes the craft of building, advantage moves from execution to judgment. Ravi Mehta frames it as a ratio shift: product used to be a roughly fifty-fifty split between building well and deciding what to build. He puts it now at ninety-ten in favor of judgment, because building got exponentially cheaper.4

Shreyas Doshi takes this point even further. Product sense, Doshi argues, is the one PM skill that will still matter, because tools commoditize but taste can't.5

“It reveals what actually requires thinking.”

Chip Huyen, on what AI does to the work7

Marty Cagan paints the same idea from the other side: pair strong product sense with generative tools and you get the most exciting leverage in the field; but hand those same tools to people without a product foundation and it turns dangerous fast.6

The most persistent misconception is that faster-moving engineers will just swallow the PM function, but these roles carry different accountabilities. On an empowered team, the PM owns value and viability, the designer owns usability, and the engineer owns feasibility. These are distinct, non-overlapping responsibilities. Cagan's read on the next decade goes further: even as AI automates most of delivery, the trio stays necessary because those competencies are too distinct and too deep to collapse into one person.8

The competencies that resist substitution:

Mehta's image lands it: AI turns product teams into jazz bands where everyone can suddenly play more instruments. Exciting — but noisy without a rhythm.4 Someone still has to conduct.

Beyond whether they can, training your best engineers to do PM work is a strategic own-goal. The opportunity cost is steep — your strongest builders create value through depth, and pulling them into discovery and stakeholder alignment strips them of their highest-leverage work.13 The cognitive modes clash, too: engineering rewards uninterrupted focus, while PM work is relentless context-switching. Force the switch and you get less effective engineers, not stronger PMs.8

Furthermore, product sense isn't a shortcut skill; it's judgment built over years of customer exposure. Without it, people tend to optimize for technical elegance over customer value, and default to polishing whatever the AI proposed rather than doing the hard, creative thinking themselves.14 The predictable result is what Perri calls the “build trap”: teams measure what they know how to measure — features shipped, velocity — and produce output without outcomes.3

AI isn't erasing the product manager, but rather splitting the profession in two. Product creators — high-agency PMs who own value and viability outright — are thriving, and the market is already rewarding them with fast-rising pay. The non-creators — backlog administrators with a PM title — are the ones genuinely exposed.15

Deborah Liu sees the old “entourage” model collapsing: PMs can no longer lean on a bench of specialists for every task. They become athletes who use AI to move faster, prototype earlier, and research deeper, while leading with the soft skills that stay irreplaceable.16 As Elena Verna frames the transition: AI isn't replacing people so much as replacing the choice to keep doing work AI can now do.17