In 2024, GitHub Copilot crossed 1.8 million paid subscribers. Figma's AI features are now used by the majority of active Figma teams. Legal has Harvey. Finance has Planful. Sales has every AI CRM tool ever built.
Product managers got AI note-takers.
That's not nothing — but it's not the same thing. An AI note-taker transcribes your meeting. GitHub Copilot writes your code. There's a meaningful difference between a tool that captures work and a tool that accelerates it.
Why PMs got left behind
The gap isn't accidental. It reflects something structural about the PM role.
Engineers write code — a highly structured, formal artifact. AI can be trained on billions of lines of it. Designers create visual assets in constrained tools with well-defined grammars. Lawyers write structured documents with established conventions.
PMs synthesize. They take unstructured input — interviews, support tickets, Slack threads, sales calls, product analytics, executive intuition — and produce decisions. The input isn't structured. The output (a prioritized roadmap, a PRD) is semi-structured at best.
That's genuinely hard to automate. But "hard" isn't "impossible," and the field has spent the last 18 months catching up.
What's changed
Three capabilities have matured enough to actually help PMs:
Retrieval over large corpora. Modern embedding + vector search means you can actually find the relevant customer quote across 200 interviews. Not keyword search — semantic search. "What do customers say about our onboarding?" against every transcript you've ever collected.
Structured extraction from messy text. LLMs are now reliable enough to extract themes, sentiment, job-to-be-done signals from interview transcripts without hallucinating at a rate that destroys trust. They still make mistakes — but the error rate is low enough to be caught in review.
Evidence linking. The hardest part of PM synthesis isn't finding insights. It's maintaining the chain from raw customer quote → theme → decision → feature. AI tools can now hold that chain and surface it on demand.
What the PM AI stack should look like
Not one tool. A stack:
Signal ingestion — Every source where customers express themselves: interviews, support tickets, app reviews, Slack, sales calls. Everything in one place, indexed.
Intelligence layer — Automatic theme extraction, opportunity clustering, sentiment tracking over time. Not manual tagging.
Decision layer — PRD generation with citations back to the raw evidence. Opportunity scoring against your strategic priorities. A living document, not a static snapshot.
Execution bridge — Jira/Linear tickets generated from the PRD. Status synced back. The loop closed.
Most PM tools today cover one of these layers. Some cover two. The gap is usually the intelligence layer — the synthesis step that turns signal into prioritized direction.
The PM's advantage
Here's something worth saying: PMs who learn to use these tools well won't be replaced by them. They'll be able to process 10x more signal with the same human judgment applied at the end. The judgment — what to build, for whom, why now — still requires a person.
What AI removes is the synthesis tax: the 15 hours a week of reading, tagging, clustering, and writing that currently crowds out strategic thinking.
That's the version of AI that matters for PMs — not one that makes decisions for you, but one that clears the path so you can make better ones.