Typing a longer prompt feels slower, yet it is the fast path. Loose instructions to a coding model do not stay loose; the model fills every gap with a guess, and each guess is a coin flip against what you actually pictured. The rework from those misses costs far more time than writing the details up front ever would.
The Details:
- A vague request forces the model to invent answers for naming, error handling, edge cases, and what counts as finished. Writing a spec is linear work: thirty minutes in, thirty minutes of output out. Fixing a misaligned generation is not linear; you read, run, find the mismatch, re-prompt, and regenerate, and by the third loop you often discover the whole approach was aimed at the wrong target.
- Precision also exposes gaps in your own thinking before any code exists. Trying to pin down exact retry counts, backoff timing, or which errors are safe to retry often reveals that you had not actually decided those things yourself. Catching that early costs seconds; catching it after three regenerations costs an afternoon.
- The real skill shift is from describing desired code to stating the conditions that code must satisfy. Junior-style prompts name a target; senior-style prompts name boundaries the output must never cross. A model given firm boundaries has a small, well-defined space to fill correctly, while a model given only a target can wander anywhere plausible.
- The tell that you have slipped into the guessing loop is saying "close, but" more than once on the same task. That phrase marks unpaid thinking debt. The fix is to stop generating, write down the required outcome and the forbidden outcomes in plain terms, and start over rather than patch a wrongly shaped result.
Bottom Line: The time spent stating constraints does not add to the task. It replaces the much larger time spent undoing a guess that missed.
