Planning
How agents plan and manage complex tasks with to-dos
SignalPilot agents don’t just react to single requests—they can create a plan. When you give a higher-level instruction, the agent will break it down into clear steps, track progress, and update the plan as it executes.
How Planning Works
-
Analyze context – the agent reviews your current notebook, available datasets, and connections.
-
Draft a plan – it proposes a step-by-step sequence to achieve your goal.
-
Ask for confirmation – before running anything, the agent shows you the full plan so you can approve, edit, or reorder steps.
-
Execute step by step – each step is marked as done once completed.
-
Update dynamically – if results change or you refine your request, the plan is revised in real time.
Why Planning Matters
Planning makes complex workflows easier to manage. Instead of juggling multiple instructions, you can give one high-level goal and let the agent:
-
Handle dependencies (e.g., load data before analysis).
-
Keep track of progress.
-
Adapt when errors or missing context appear.
-
Ensure you stay in control by approving the plan upfront.
FAQ
Q: Why does the agent ask for confirmation first?
To give you control. You can approve, reject, or adjust the plan before any code runs.
Q: How is planning different from chat?
Chat handles single requests, while planning organizes multiple steps into a structured workflow.
Q: What happens if a step fails?
The plan pauses, shows the error, and suggests fixes before continuing.
Q: Can I update the plan mid-execution?
Yes. You can stop the agent at any moment, edit the plan, refine your instructions, and the agent will regenerate the plan with the changes.
Q: Does the plan persist across sessions?
Yes. The plan is saved as part of the notebook and the agent will remember the plan the next time the notebook is loaded.