The AI-era PM is not the smartest typist; they're the sharpest decider. AI does the writing and the summarizing; the PM does the vetting, the scoping, and the "what do we actually promise users" call. On shipped projects, that's moved weeks into decisions and half the typing into judgment.
There is a neat-looking "the PM is dead" stream on LinkedIn. It's wrong, but it's pointing at something real: the spread between a good and a fine PM is widening, and the drying part is now delegable.
Before the AI era
A classic PM day churned through drafting tickets, summarizing calls and decks, and polishing spec language for an audience of one. It looked busy, right.
The AI-era shape
The modern-day actually spends:
- Drafting delegated — AI drafts; the PM edits for truth and tone
- Judgment centralized — the PM decides which problems are worth a build
- Evidence kept human — measurements still come from your own funnels and interviews
- Scope defended — the PM says "no" more than "yes" and everyone survives
What I actually got from it
The most concrete shrinker: daily report automation. The previous post replaced a 30-minute manual slice of the day with a script — I got back to the thinking those 30 minutes used to crowd out.
And call summarizations: 20+ discovery calls became an insight map, not a binder of transcripts — exactly what I describe here at the extreme.
The skills I copied from being PM
- Scope is your best product — a tight enough build that AI can staff, and the client trusts what's real
- Human veto — nothing ships that the model wrote unchecked
- Metrics before meetings — the checklist still excludes a direct methodology
The honest limits
- AI drafts conclusions too confidently — a PM sorts the confident but wrong
- Model output is best as a skeleton, never a tombstone
- Product strategy is still yours: where the yard has value
For where my head really is now: the deep-dive method for evidence, and agents in prod for the build side.