What Is Agent Planning?
Agent planning is the process of deciding what steps to take to achieve a goal before execution begins. A plan converts a high-level goal into an ordered sequence of concrete actions.
Why Planning Matters
Without planning, an agent is purely reactive: it takes one step, observes the result, and decides the next step. This works for simple tasks, but fails for complex goals that require coordinating many steps in the right order.
Planning allows the agent to:
- Think through the full task before starting
- Identify dependencies between steps
- Allocate the right tools to the right subtasks
- Anticipate problems and prepare fallbacks
- Estimate how long the task will take
Example without planning:
Goal: 'Prepare a competitor analysis and send it to the CEO.'
Unplanned agent: searches randomly, may forget to email, may repeat steps.
Example with planning:
Step 1: Identify top 5 competitors (web_search)
Step 2: Research each competitor's product (web_search × 5)
Step 3: Analyse pricing for each (extract_data)
Step 4: Write the analysis report (summarise)
Step 5: Send to CEO (send_email) — requires human approval
Plan Properties
- Ordered: steps have a clear sequence. Each step can depend on results from previous steps.
- Bounded: the plan has a finite number of steps. Open-ended plans become infinite loops.
- Validated: each step is checked for feasibility before the plan is accepted
- Flexible: the plan can be updated if a step fails or produces unexpected results
- Auditable: the plan is stored in state so it can be reviewed after execution
Key Takeaways
- Agent planning is the process of deciding what steps to take to achieve a goal before execution begins.
- Ordered: steps have a clear sequence. Each step can depend on results from previous steps.
- Bounded: the plan has a finite number of steps. Open-ended plans become infinite loops.
- Validated: each step is checked for feasibility before the plan is accepted