My rule of thumb for building products — click any card to expand.
Updated as I learn. These aren't rules, just patterns that tend to work.
Outputs only matter when they move user or business outcomes.
Define the problem crisply, tie every feature to a measurable outcome, and ask if the outcome can be achieved without building. In the AI era, separate human outcomes from agent outcomes and set clear success metrics for both.
Reuse what works, save effort for differentiation.
Start with patterns that already work, then adapt them for AI-native flows. Keep familiar onboarding and permissions, but optimize for agent handoffs, automation loops, and tool interoperability.
Iteration reveals reality faster than planning.
AI lowers development cost, so test more and pivot faster. Release the smallest version that tests the riskiest assumption, instrument it, and iterate. Prefer experiment velocity over big-bang launches.
If it needs a manual, it isn't done.
Favor familiar language and obvious actions. Reduce cognitive load with clean hierarchy, sensible defaults, and predictable patterns. Make the happy path obvious; explain only when needed.
Commit real resources when users commit something too.
Look for pre-commit signals: usage, time, integrations, budget, public requests. Balance this with small, fast experiments for emerging paradigms (agent-first UX, AI-native flows) where demand won’t be obvious yet.