Plan Mode
Plan Mode lets you think through a feature, migration, or refactor before the AI writes any code. Outline the scope, decide on an approach, approve a step-by-step plan, then hand it off to Build Mode.
What Plan Mode does
In Plan Mode, the AI won't modify files. Instead, it:
- Scopes the work: What needs to change and what's out of scope
- Proposes an approach: The high-level strategy
- Breaks it into steps: Concrete, ordered implementation steps with file paths
- Identifies dependencies: New packages or integrations needed
- Flags risks: What could go wrong and how to mitigate
The output is a numbered plan you can review, adjust, and approve before anything gets built.
How to access Plan Mode
From the mode toggle
Click the mode selector in the bottom-left of the chat input and select Plan.
With a slash command
Type /plan followed by your request.
Keyboard shortcut: Ctrl+Shift+P (or Cmd+Shift+P on Mac)
With the deep planning command
For complex, multi-domain tasks, type /deep-planning followed by your request. This activates a more thorough multi-step reasoning mode with architecture analysis.
When to use Plan Mode
- The task touches many files: A feature spanning API routes, database schema, and UI benefits from mapping out changes first.
- You're unfamiliar with the area: Have the AI explore and document the relevant code paths before proposing changes.
- The approach isn't obvious: When there are multiple valid ways to do something, use Plan Mode to evaluate trade-offs.
- You want to estimate effort: A detailed plan shows how much work is involved before you commit.
For small, self-contained changes: a typo fix, a style tweak, a single component: skip planning and go straight to Build Mode.
The planning process
- Describe your goal: Be specific about requirements, constraints, and desired behavior.
- Review the plan: The AI presents numbered steps with file paths and rationale.
- Iterate: Ask the AI to adjust steps, add considerations, or change the approach. The plan is a conversation, not a one-shot output.
- Approve and execute: Switch to Build Mode and tell the AI to execute the plan.
Plan steps in the chat
Each step appears as a trackable item. As the AI executes them in Build Mode, completed steps are checked off so you can follow progress.
Tips
- Start broad, then narrow: Give a high-level description first. Let the AI ask clarifying questions before drilling into details.
- Reference existing patterns: Mention how similar features are implemented so the plan follows your conventions.
- Set constraints upfront: Library preferences, file organization, architectural decisions. State them early so the plan accounts for them.
- Split large tasks: If the AI says a task is too large for one plan, follow its recommendation.