How much of a manager are you?

Key takeaways

1. Write a spec before you prompt: pick the framework, define the data types, and sketch the API or pages, so the model has somewhere to start instead of a blank canvas.
2. Tighten your feedback loop. Instant page reloads, unit tests the model can re-run, and cheap automated checks like lint or CI let you judge the output instead of reading every line of code.
3. Build bottoms-up in small steps: database types and methods first, then API routes, then frontend pages, then critique the styling. Models rarely nail a one-shot build.
4. Sort work into cheap tasks (boilerplate, unit tests, small regressions, docs, one-off scripts) and expensive ones (big refactors, mission-critical APIs, core data models, database migrations), and hand off the cheap ones first.
5. Keep system design and complex work like race conditions and performance-critical code for yourself. Nudge the model to refactor too, since it tends to keep adding code rather than consolidating it.