How to use AI agents on a real codebase

Using on an existing depends on the size and risk of the change. A small fix may be safe to hand over right away, but a larger feature or risky change usually needs a first step where the agent studies the code and makes an .

There is no single settled workflow yet. The practical question is how to choose between asking questions, planning the work, and letting the agent edit code immediately.

Key points

  • should change based on the size and risk of the task.
  • Small fixes can often go straight to code changes.
  • Larger or riskier work benefits from questions and an first.
  • There is not yet one standard workflow that fits every project.
Read original