Intent Preview
Review the Plan Before It Executes
Before an AI agent modifies your code, it shows every planned step with risk levels, affected files, and scope estimates. You can reorder, skip, or approve before a single line changes.
Framing
The problem
Users cannot safely delegate actions they cannot inspect before execution.
The pattern
Show the plan, scope, and risk before the system takes action.
Why chat breaks here
Reassuring assistant language does not expose concrete steps or blast radius.
Risks
Preview friction can become excessive for trivial or low-risk actions.
Avoid when
The action is cheap to reverse and pre-review costs more than correction.
Use when
Before any agent action that costs money, sends a message, or changes shared state.
DOPE evaluation
- Directability
- Reorder, skip, or modify steps before running
- Observability
- Every planned step is visible before execution
- Predictability
- Risk levels and file impacts shown upfront
- Explainability
- Risk levels include plain-language rationale, not just color codes
In the wild
- Claude Code Plan Mode (Anthropic) — Plan Mode (Shift+Tab+Tab or /plan) drafts files-it-will-touch, side effects, and risks before any execution. /ultraplan since Apr 2026 deepens the review. Strongest example in 2026.
- Cursor Composer (Cursor) — Composer ships ask-before-act + security review + permission scopes. Caveat: structured pre-execution plan with risk badges is not the hero surface here — Plan-mode parity is weaker than Claude Code.
- Replit Agent Plan Mode (Replit) — Plan button in chat input produces an ordered, approvable task list before Agent starts building. The plan is a real artifact, not a paragraph.
- OpenAI Deep Research Plan (OpenAI) — Feb 2026 update: users create and edit a research plan before execution starts. Editable plan is the contract before tokens get spent.
FAQ
When should I use the Intent Preview pattern?
Before any agent action that costs money, sends a message, or changes shared state.
When should I avoid the Intent Preview pattern?
The action is cheap to reverse and pre-review costs more than correction.
What problem does Intent Preview solve?
Users cannot safely delegate actions they cannot inspect before execution.
Why is chat the wrong fit for this?
Reassuring assistant language does not expose concrete steps or blast radius.
Related patterns
- Often paired with: Editable Plan — Preview makes scope visible; the editable plan lets you change it before kickoff.
- Often paired with: Approval Gates — Preview at the start, gates inside the run — full coverage for delegated work.
- Alternative to: Reasoning Trace — Plan upfront vs reasoning chain after the fact for the same auditability.