AI is a teammate, not a toy
I have rebuilt my own workflow around AI: drafting specs, summarizing research, interrogating data and pressure testing decisions. The point is not novelty. It is compressing the boring 60 percent of PM work so I spend more time on judgment.
Automate the drafts, own the decisions
I let models produce first drafts, such as PRDs, meeting summaries and competitive scans, then I edit hard. The model is fast and confidently wrong sometimes. My job is to bring the judgment it cannot. Never ship what you did not verify.
Build small workflows, not big platforms
Most value comes from unglamorous chained steps: ingest the transcript, extract decisions, draft the follow up, flag the risks. I design these lightweight workflows for a team rather than waiting for a grand AI platform that never arrives.
Keep a human in the loop where it matters
For anything customer facing or irreversible, AI drafts and a human approves, always. I am explicit about where automation ends, so speed never quietly turns into an unreviewed decision that ships.
Real example
I use AI daily across discovery and delivery: turning user interviews into themed insights, roadmaps into stakeholder ready narratives, and raw metrics into a first pass analysis I then verify. It roughly doubles my throughput on the written half of the job.