The execution loop
1) Define a challenge Pick a painful user task with clear success criteria.
2) Build a feelable prototype Convert ideas into something users can experience, not just read.
3) Push to dogfood fast Internal usage catches workflow failures early.
4) Leadership review passes Use real behavior and metrics, not abstract slides.
5) Generate docs from reality Create PRD/spec from working code and observed behavior.
6) Launch experiment Instrument outcomes and iterate on observed frictions.
Two PM visions to choose from
Aggressive path
- PM drives end-to-end build to production.
- AI handles implementation acceleration.
- PM owns outcome metrics directly.
Stepping-stone path
- PM starts with prototype plus real usage data.
- Reviews with leadership happen early and often.
- Specs/docs are generated from code and user behavior.
What to track
- Time from idea to first real user touch
- Time from feedback to shipped iteration
- Decision quality from prototype rounds
- Outcome lift versus baseline behavior
Practical prompt discipline
Don’t start by writing the perfect prompt manually. Start with rough intent, ask AI to draft the spec, review it, then execute. Better loop: rough sketch → AI spec → review → build → measure.