Notes

Every metric needs three things: the number, the mechanism that caused it, and the counter-metric (what you sacrificed). Without all three, you're guessing.

The 80/20 rule applied to AI: use deterministic rules for predictable problems, reserve ML for genuine judgment calls. Most features don't need a model.

Payments require a different PM mindset. Safety before possibility. One week reading the API and tracing money flows prevents months of production fires.

Remote PM fails when you solve timezone overlap with better scheduling. The real skill is async-first: persistent docs, written decisions, shared context.

When nobody disagrees, nobody's thinking. A PM with no position turns the roadmap into a parking lot of unrelated features.

Diagrams hide the hard parts. Building a real system, even a small one, exposes the architectural constraints that should drive product decisions.

Build vs buy is four strategic questions in sequence, not a technical one: constraint, differentiator, total cost, reversibility.

Seven years freelancing taught one rule above all: shipping is what matters. Clients care about delivery dates and visible results, not elegant process.