One useful Claude Code workflow is simple: let Opus plan, then let Sonnet execute.
This is not magic. It is just a clean role split.
Opus is good for deciding direction
When the task is unclear, planning matters.
Opus is useful for reading the situation, choosing the steps, and deciding what should not be touched.
Sonnet is good for moving fast
Once the plan is clear, execution can be simpler.
Sonnet can follow the plan, edit files, and run checks.
The handoff must be short
The handoff should not be a long essay.
Use a compact format:
- Goal
- Files
- Do not touch
- Steps
- Verification
Why this helped
The work felt faster because I stopped asking one model to do every type of thinking at once.
Original Japanese Article
This article is based on the Japanese post: Claude CodeでOpusに計画させてSonnetに実行させると速くなった話.
