Why AI Won't Replace PMs (But Will Replace Bad Ones)

The question of whether AI will replace product managers resurfaces every time a new model drops. The answer is always the same: not the good ones, and not anytime soon. But the more useful question isn't whether the role disappears — it's which version of the role becomes redundant.

There's a version of product management that is highly automatable. There's a version that isn't. The difference is worth understanding clearly.

The Automatable Version of PM

Some PMs spend the majority of their time on work that is structurally identical to what AI does well:

Document production — Writing PRDs, updating wikis, drafting release notes, formatting roadmaps. This is pattern-heavy writing that benefits from a first draft but doesn't require deep contextual judgment.

Information aggregation — Pulling together status updates, synthesising meeting notes, creating dashboards from metrics that already exist. This is coordination work that doesn't add unique value.

Generic requirements writing — Writing user stories that could apply to any product in the category. "As a user, I want to be able to filter results so that I can find what I'm looking for" is valid but doesn't require knowing your specific users.

Meeting facilitation that substitutes for actual decisions — Running alignment meetings that produce consensus documents but defer the actual call. AI won't be running these meetings, but it will make them less necessary if documentation is automated and decisions are made with better information.

Research synthesis without interpretation — Summarising what users said without addressing what it means for the roadmap. AI can produce the summary; the interpretation still requires a human.

If a significant portion of your current role looks like this list, that's worth paying attention to.

The Non-Automatable Version

The parts of product management that AI cannot replicate — and is not approaching — are:

Strategic synthesis under uncertainty — Deciding what matters most when you have incomplete information, competing stakeholder priorities, and a market that's moving faster than your research cycle. This isn't a pattern-matching problem; it's a judgment call that depends on context AI doesn't have.

User empathy at depth — Understanding not just what users say they need but what they're actually trying to accomplish, why they've stopped reporting a problem they still have, and what would genuinely delight them rather than just satisfying a stated preference. This requires human-level presence in user conversations.

Organisational navigation — Knowing which engineers will push back on a technical requirement and why, which stakeholder needs early involvement before they become blockers, and how to frame a deprioritisation so the affected team doesn't feel dismissed. This is relationship work that requires a presence and a history.

Novel problem framing — Defining the right problem to solve when nobody's done it before. AI is trained on solved problems; genuinely new problem categories require thinking that deliberately breaks from existing patterns.

Trust-carrying — Being the person whose judgment engineers trust enough to build without a full spec, whose read of a customer situation a CEO defers to, whose assessment of a market shift lands with credibility. Trust is earned through demonstrated judgment over time. AI doesn't carry it.

What "Replace Bad PMs" Actually Means

The provocative framing is intentional. The PMs who are most at risk aren't necessarily junior PMs or generalists — they're PMs who've built their professional value around the automatable parts of the role.

The PM who is valued primarily for producing comprehensive documentation quickly will find AI produces similar documentation faster. The PM who is valued for the depth of their user understanding, the quality of their strategic calls, and the trust they've built with their cross-functional team has something AI can't replicate.

This isn't a prediction about layoffs in any specific timeframe. It's an observation about where durable value accumulates. The skills that will matter in three years are not the same as the ones that look impressive in a quarterly review today.

The Skills Worth Investing In Now

Conducting high-quality user research — Not just running interviews, but designing research that uncovers mental models and latent needs rather than confirming assumptions. The quality of AI synthesis depends entirely on the quality of the inputs; being excellent at the human side of research becomes more valuable, not less.

Developing strong strategic judgment — Making clear calls with incomplete information, being right about what matters more often than not, and being able to explain the reasoning. AI can model scenarios; it can't make the call. The PM who consistently makes good calls is irreplaceable.

Building genuine cross-functional trust — Engineering leads, designers, and data scientists who would go to bat for your prioritisation decisions because they trust your judgment. This requires being right, being honest about uncertainty, and following through. No tool shortcut this.

Becoming fluent in AI tools — Understanding the failure modes, knowing how to get useful output, and being able to evaluate AI-generated work critically. The PMs who get the most from AI tools are the ones who understand what AI doesn't do well. This fluency is itself a differentiating skill.

Developing clear communication — The PM whose written and verbal communication is clear, concise, and well-structured for the audience produces PRDs that get built from, roadmaps that get aligned behind, and updates that generate confidence. AI drafts the documents; the clarity of thought behind them comes from the person.

The Honest Version

AI is compressing the time cost of being a mediocre PM. It's not compressing the cost of being an excellent one — because the excellent parts require judgment that AI doesn't have.

The medium-term risk isn't that AI replaces PMs. It's that the supply of "good enough" PM work increases (because AI can produce it), while the demand for excellent PM work — judgment, empathy, strategy, trust — stays constant or grows.

That's a good situation for PMs who've invested in the non-automatable parts of the role. It's a difficult situation for PMs who've coasted on the automatable ones.

The answer to "will AI replace PMs?" is: it will replace the work, not the judgment. Whether it replaces you depends on which one you've built your career around.