These aren't things I read somewhere. Each one is a rule I've broken once, learned from, and now hold to.
Start with users. Understand systems before building features.
User interviews beat assumptions.
Technology should reduce cognitive load, not add a new interface to learn on top of an old problem.
AI should augment humans before it replaces them. On AgriLokal, that meant keeping an agronomist in the review loop permanently, not as a gap to close, but as a deliberate design choice, because the model's output affects a real farming decision.
Every metric represents user behavior. If I can't trace a number back to something a person actually experienced, I don't fully trust it yet.
Growth comes from better products, not hacks. A conversion lift confounded by a cheaper acquisition week isn't a win I'll claim cleanly.
Every feature introduces complexity. The question is never "can we build this," it's "what will this cost the next person who has to reason about the system."
Product management is mostly saying no, to scope, to premature build, to the version of a feature that's easiest to ship instead of the one that's right.
Curiosity compounds. The best product decisions I've made started as a question I didn't need to ask.