Most of a developer's day isn't writing new code — it's understanding old code, reproducing bugs, and reviewing changes. A few sharp prompts take the grind out of all three.
Reproduce before you fix
A minimal reproduction is worth ten guesses. Feed in the error and the surrounding code and ask for the smallest snippet that still fails, plus the most likely root cause.
Let tests write themselves
- Generate happy-path and edge-case tests for a single function
- Ask for tests named after the behavior, not the method
- Have the model flag the cases it could not cover
Treat the output as a first draft you review — not as gospel. The prompt does the typing; you keep the judgment.