← Journal
8 min readAI Dev Review

Prompting patterns for large refactors

The refactor prompts that actually work — plus the ones that produce plausible garbage.

PromptingBest Practices

Refactors are the highest-leverage AI use case and the easiest to get wrong. A handful of prompt patterns account for most of the wins.

Patterns that work

  • Freeze the interface. Tell the agent what the public API must remain, then let it rework internals.
  • Ask for a plan first. Approve or edit it before any file changes.
  • Constrain the blast radius. 'Only touch files under src/billing.' Agents obey scope surprisingly well.

Patterns that fail

Open-ended 'clean this up' requests. Multi-goal refactors ('extract this and also add tests and also update docs'). Agents will do all three badly.

Comments

0 replies

Join the conversation

Sign in to leave a comment on "Prompting patterns for large refactors".

Loading comments…