Overview
Action Plan
Step-by-step execution plan with approval controls
planstepsapproval
Docs
Action Plan makes the AI's intended steps visible and controllable before and during execution. Rather than letting the AI act autonomously, it presents a structured list of planned actions that users can review, approve, modify, or reject. This pattern is essential for agentic workflows where the AI performs multi-step tasks with real-world side effects — file modifications, API calls, data transformations — that users need to understand and authorize.
Variants
- Pre-execution plan — A list of proposed steps shown before any action begins, requiring explicit user approval to proceed.
- Live execution tracker — A real-time view of steps as they execute, with status indicators showing progress, completion, and failures.
- Editable plan — An interactive plan where users can reorder, remove, or modify steps before or during execution.
Use Cases
- AI coding agents that modify multiple files in a project
- Data pipeline tools that transform and move data across systems
- Automated workflow builders that chain multiple AI operations
- DevOps tools that plan and execute infrastructure changes
- Research assistants that gather and synthesize from multiple sources
Best Practices
- Always show the plan before executing — users should review and approve before any side effects occur.
- Use clear status indicators for each step: pending, in-progress, completed, failed, skipped.
- Show estimated duration or cost per step when available.
- Allow users to pause, cancel, or roll back execution at any point.
- Surface the tools or APIs each step will use so users can assess risk.