Product Managers Are About to Be Found Out

Most product work is language, and AI just learned to produce the language without the judgment underneath. The product manager most exposed is the one who confused the two.

For two years, engineers argued about whether AI was coming for them. AI writes code, so it threatens the people who write code. Product managers mostly watched from the side, and a few felt a quiet relief. The same tool seemed to hand them power instead. You can prototype now without waiting in the engineering queue. You can build the thing instead of just describing it.

So if you are a product manager who felt that relief, the question underneath it deserves a worse answer than the reassuring one you keep getting. Will AI replace product managers? Most of the internet rushes to pat you on the head. No, it says. Your job is human and strategic and safe.

The relief is half a truth.

(If you want to know how your specific situation stands, use the quick, interactive, free app at the end of this article.)

The dangerous edge was never pointed where you think

Engineers got one thing half right. They saw the threat aimed at code, because code is what they make. It was. What they missed is that it never stopped at code. PMs are walking straight into the mirror version. They see the threat aimed at engineers, because engineers make code and AI makes code now too, and they file it under someone else's problem.

I have already written about the upside, the handoff dissolving, the PM who shows up with a working prototype instead of a spec. That part is real, and it is good.

But AI did not learn only to write code. It learned to write the things you write.

PRDs (Product requirements documents). Roadmaps. User stories. Prioritization arguments. Competitive teardowns. Customer summaries. Launch plans. Strategy memos. The stakeholder update that turns a messy, political decision into something that reads like it was always the plan.

That is most of the visible surface of the job. A machine that has read a million of them can now produce a passable version of any one in seconds.

This does not mean AI is a good product manager. It means AI is very good at the surface area by which average PM work gets recognized as PM work. The threat was never that AI becomes Marty Cagan.¹ The threat is that AI becomes good enough to pass for the PM three desks over.

AI does not need to be a great PM to threaten PMs. It only needs to be good enough at the visible outputs of the average one.

Why a bad PRD is more dangerous than bad code

The prototype was never the scary part.

Code has a brutal honesty to it. Generate a function that looks right, and the test suite, the build, the 2am incident will tell you it is wrong. The feedback loop is short, and it does not care how confident you sounded in the pull request.

Product work has no such reflex. A roadmap can be wrong and still sound mature. A prioritization matrix can be arbitrary and still look rigorous. A persona can be invented at a desk and still feel user-centered. A strategy memo can be completely generic and still get nodded through a room of smart people.

You have sat through this one. Every planning cycle produces it. Here is the genre in miniature:

Enterprise grew 38% last year, but net revenue retention has slipped two quarters running, and the churn concentrates in accounts that never turned on the collaboration features. The read is clear: we won on individual productivity and we are losing on team stickiness. Three bets follow. Deepen collaboration to defend retention, expand into adjacent departments to grow seats, or build the integration layer that makes us infrastructure. Retention is the open wound, so collaboration is where we lead. We should reallocate now, before a competitor turns our stickiness gap into their wedge.

Read it like it landed in your inbox from the desk next to yours. A number, a trend, a named risk, a defended choice. It sounds like someone did the work. I wrote none of it. A model produced it in one pass from a one-line prompt. Set it beside the strategy memo your coworker wrote last quarter and you cannot tell them apart, and neither could the room. Now look at what it is actually made of. Metrics and inference, and not one customer anywhere inside. That is the tell, and almost nobody catches it in the meeting, because the meeting is not looking for it.

Now run the real one forward, the version a person actually presents and the room approves. A quarter later the bets have gone nowhere, the memo is forgotten, and its author has written three more like it. The document was never tested against reality. It was only ever tested against the taste of the people in the room, and they liked it.

PMs like to call ambiguity their moat. It was never that. Ambiguity is the medium product work happens in, and the medium has no test suite. The good PM resolves the ambiguity and reality eventually agrees. The mediocre one hides inside it, producing work that sounds correct right up until a quarter is gone and reality finally files its report.

That slow verdict is the exact gap AI thrives in. It produces the convincing surface faster, cheaper, and without end. The thing that let a weak PM survive, that product failure arrives late and quietly, is the same thing that makes their output trivial to imitate.

Call it the falsification gap. The slower your work is to prove wrong, the easier it is for a machine to fake.

The product manager AI actually replaces

Not "PMs." That word is too big to be useful here.

It comes for the PM whose value was writing clean documents. The one whose value was running the process. The one who summarized the meeting, applied the framework, and turned a pile of stakeholder opinions into roadmap language. The one with no direct line to a customer, who cannot tell plausible demand from real demand because they have only ever met demand through a dashboard.

It is tempting to read that as the junior, the entry-level PM with the least to fall back on. But this was never really about seniority. A senior PM who spent ten years getting fluent at producing the artifacts, and never built the muscle underneath, is more exposed than a junior who talks to customers every week. The question is not how long you have done the job. It is whether the job was ever more than the documents.

That PM was already running on borrowed time. AI just mailed the invoice early.

The same machine makes another PM far more dangerous. The one with taste. The one who actually talks to customers and hears the hesitation, the contradiction, the thing they go quiet about. The one who can prototype, run an experiment, read distribution, and kill nine good ideas to protect one. AI hands that PM a research engine and a drafting engine and gets out of the way. Their judgment now moves at machine speed.

Same tool. Opposite effect. The split runs along one line: whether your judgment was the point, or just the packaging.

This is not about who adopts AI fastest, whatever the loudest version of the advice says. The weak PM who masters every new tool just generates their forgettable documents faster. Adoption is not the line. It only speeds you toward whichever side of the line you were already standing on.

That split is already showing up in who gets hired. People watching the PM job market describe it pulling apart into two piles, strong demand for product managers who can build and decide, thinning room for the generalist whose output was mostly documents.²

Here is the test, if you want to know which pile you are in. Look back over your last quarter and find one decision specific enough that reality could prove it wrong, with your name on it when the verdict came. Not a document you authored. A call you owned. If you reach for a roadmap or a deck instead, you already have your answer, and soon everyone else will too.

The job starts where the narrative stops

Safety moved. It used to live in producing the artifacts. Now it lives in the things the artifacts only ever pointed at: real contact with customers, the constraints, distribution, timing, taste, and the nerve to cut.

But notice what AI just learned to do. It writes the research summary. It generates the persona. It produces the slide that says here is what customers told us. Contact with reality has a surface too, and the surface is the next thing to get faked. Plenty of PMs will show up with a deck full of quotes and call it proof.

So pointing at the customer is not the safe ground either. The one thing that cannot be counterfeited is the bet you make off the back of it. A decision specific enough that reality can prove it wrong, with your name on it when it does.

No plan survives contact with the enemy, the old military line goes.³ No roadmap survives contact with a customer either. This was always true, and the good PM always knew it. The document was a hypothesis, not a verdict, and the job was to go find the contact that would settle it. The weak PM mistook the writing for the work, and AI just made the writing free.

The strong PM manufactures that test on purpose. Not by producing a better document, but by forcing the verdict to arrive early, in days instead of quarters. A fake door and a count of who clicks through. The thinnest slice of the thing shipped to ten real users. A pre-order taken before a line of it exists. Each one drags a verdict back from next quarter into this week. They build the test suite product work never had, and the falsification gap that sheltered the weak PM, they close on themselves.

Is product management dead? No. The average version of it is the part that dies. AI will not erase product management. It will redraw it. The weak PM, the artifact-maker, becomes easy to route around. The strong PM, the one who owns the bet and its consequence, becomes more dangerous than before.

The engineer's job, in the end, starts where the AI prototype stops. Yours has the same shape. AI can generate the whole convincing product narrative now: the problem statement, the roadmap, the memo that makes all of it sound inevitable.

Your job starts on the next line. The one the narrative cannot write, because it asks for someone willing to be wrong on purpose, in public, with a real customer on the other end.

Want to find out what your actual specific situation is?

Answer 6 quick questions to get a clear analysis and personalized next steps (with hundreds of possible results, including one that matches your case), all here- no gate, no signup, completely free.

Rabbit Holes:

If you want the upside of this same shift, I wrote about the PM who already built it and found the bugs, where the prototype becomes the PM's real output. The engineering mirror of this whole argument lives in the uber-engineer who doesn't write code. And if the hard part is now deciding what deserves to exist at all, that is the new bottleneck. And if the thread you keep pulling is how to tell real demand from the plausible kind, your signups are lying to you.

Footnotes:

1. Marty Cagan is widely regarded as the leading thinker on technology product management, founder of the Silicon Valley Product Group and author of INSPIRED and EMPOWERED. He has a phrase for exactly this, "product management theater," the role performed convincingly without the substance underneath.
2. For one account of this divide, see The Great Reshuffling: How AI Is Polarizing Product Management Roles, which describes a K-shaped split in PM hiring between AI-fluent builders and disappearing generalist roles.
3. The line is usually traced to Prussian field marshal Helmuth von Moltke the Elder, whose actual formulation was that "no plan of operations extends with any certainty beyond the first encounter with the enemy's main strength." See Helmuth von Moltke the Elder.