Plan Mode – think first, type later

Claude dives straight in. Three files changed, two new services created, and then I realise: that's not what I wanted at all. Sound familiar?

That's why we have a simple rule, for humans and AI alike: plan first, code second. Claude Code has a dedicated mode for exactly that.

⏸️ Switching on plan mode

• Shift+Tab cycles through the modes: default → accept edits → plan mode. The status bar then shows “plan mode on”.
• For a single task: start your prompt with /plan.
• For the whole session: claude --permission-mode plan

🔍 What happens next

Claude reads, searches and analyses, but doesn't change any files. At the end you get a plan: which files, which steps, which open questions.

You decide: approve it, refine it (“No, keep planning”) or throw it away. Claude only starts implementing once you approve.

📝 My tip

Have Claude save the approved plan as a file, e.g. docs/plan-feature-x.md. Your team can review it, and your next session doesn't start from zero (see part 9).

A word of caution:

Approving the plan takes Claude out of plan mode and into whichever mode you pick when approving. So switch it back on for the next bigger task. And yes, planning costs tokens too. But far fewer than going down the wrong path and having to undo it all afterwards.

My rule of thumb:

• Typo, one-liner, obvious bug fix? Just let Claude do it.
• Several files, unfamiliar project, architecture question? Plan mode.
• Don't like the plan? Better to refine it three times than roll back once.