Definition
Plan mode
Plan mode is an explicit phase in which an AI coding agent breaks a complex task into structured steps before implementing anything. In Claude Code it is reached with Shift+Tab or /plan, with /effort adjusting how hard the model thinks. Cursor and Codex expose planning through their own agent interfaces.
Why it matters
A plan is reviewable before it costs anything. Reading five steps and correcting the second is cheaper than reading a diff across forty files and working out which step went wrong — and it is the difference between supervising an agent and cleaning up after one.
How each tool handles it
- Cursor
- Ask mode is read-only and serves the same exploration phase before Agent mode executes.
- Claude Code
-
Shift+Tabor/plan;/efforttunes the depth. - Codex
- Planning is exposed through the current agent interfaces.
Frequently asked questions
Is `think` a command that enables planning?
No. In Claude Code think is ordinary prompt text, not a command. Plan mode is entered with Shift+Tab or /plan, and /effort is what adjusts how much reasoning the model spends.
When is plan mode worth the extra step?
When the change spans files you have not read, when the approach is genuinely uncertain, or when a wrong turn is expensive to undo. For a single well-understood edit it is overhead.
Is Cursor's Ask mode the same as plan mode?
It serves the same phase. Ask mode is read-only: the AI reads your codebase and answers questions but cannot modify files, which is the exploration and planning step before Agent mode executes. The separation is by permission rather than by an explicit plan artefact.