What's the point of a PM if any stakeholder can prompt their way to a working prototype?

The shift

Producing a plausible, working prototype from a stated intent goes from scarce — gated by a PM writing a spec and an engineer building it — to abundant: near-zero cost, minutes, doable by anyone who can describe what they want.

The axioms

• Stakeholders can't build, so they route intent through a PM who translates it into something engineers can execute. Rests on: technical execution capacity being scarce.
• Whoever can produce the spec, mockup, or prototype controls what actually gets built. Rests on: production capacity as a gatekeeping lever.
• The PM is the single throat to choke for what's being built and why. Rests on: accountability for the outcome.
• Prioritization means reconciling competing stakeholder demands against limited engineering time. Rests on: engineering capacity being scarce, and judgment about tradeoffs being scarce.
• Users don't know what they want until they see it, so someone has to guess right before real build cost is committed. Rests on: correct judgment under uncertainty being scarce.
• A prototype that demos well is evidence the idea works. Rests on: verification beyond the happy path being scarce and expensive to obtain any other way.

Invalid axioms

1. Stakeholders need a PM to turn their intent into a tangible artifact. Producing the artifact was the scarce step; a PM's ability to spec something buildable was the moat. That's flipped — a stakeholder can now prompt a working click-through or even a functional prototype directly. The habit-trap: orgs still insist every idea pass through a PM-authored spec before anyone's allowed to see if it works, when seeing if it works is now the cheap part.
2. Producing the prototype is what earns you the right to decide what ships. When only the PM (or a designer, or an engineer) could make the idea concrete, "I built the mockup" was a reasonable proxy for "I've thought this through." Now anyone can generate that same artifact in an afternoon regardless of how well they've thought it through. The habit-trap: treating a slick self-service prototype as validated judgment rather than as a first draft with the same evidentiary weight as a paragraph of text.

Unchanged axioms

1. Someone has to own the tradeoff across competing demands. Ten stakeholders can each prompt their own prototype for what they want built next; engineering capacity to actually ship and maintain any of them is still finite. Deciding which one is worth real investment is a scarce judgment call, not a production problem.
2. Someone accountable has to answer for why this and not that, and for what happens after launch. A model can generate the prototype; it can't be on the hook when the feature cannibalizes revenue, breaks a workflow for an enterprise customer, or fails a compliance review. That answerability doesn't get cheaper because drafting got cheaper.
3. A demo surviving in someone's hands is not evidence it survives in production. A stakeholder's prompted prototype proves the happy path renders. It says nothing about auth edge cases, data at scale, error states, support burden, or whether it holds up against the fifteen other things already promised to that same user segment. That verification gap is still real work, and it's the same work a PM already did before AI — it just now has to run against a much higher volume of incoming prototypes.
4. Deciding what's worth building at all is a taste and judgment call, not an execution call. Stakeholders prompting a prototype answers "can this be built," not "should this be built, now, for this reason, at the expense of what else." That's the part that was never about production capacity.

New axioms

1. When every stakeholder can generate an equally credible-looking prototype, whose gets built? Confidence of presentation stops correlating with quality of the underlying idea — a beautifully prompted demo from someone with no product sense looks identical to one from someone with real insight. The org has to solve for signal versus noise across a flood of self-service prototypes it never had to filter before.
2. Adjudicating between N competing, already-built prototypes is a new kind of conflict. Previously, a PM said no to an idea before it cost anything. Now the idea already exists as a working thing someone is emotionally invested in, and saying no means killing a built artifact, not a slide. That's a harder conversation than the old one, and nobody's designed a process for it yet.

Where it breaks

"The PM's value was writing the spec that turns intent into a buildable thing" collides directly with "the flood of self-service prototypes still needs someone to decide which is worth real investment." Orgs that let PMs keep spending their time on spec-writing and prototype production — the now-cheap step — are starving the one thing that got harder: triaging and adjudicating a pile of equally plausible, already-built ideas competing for the same finite engineering slot.

ai-axioms.vercel.app/axioms/product-management/product-pm-point-ai-axioms