All concepts
Agent Planning
Break a goal into ordered or conditional sub-tasks before acting.
Agentic AI · Intermediate · ~8 min
In plain English
Decide the whole route before setting off, instead of choosing each turn as you reach it. Useful when steps depend on each other and backtracking is expensive.
Why it's worth your time
It's the difference between an agent that wanders and one that finishes — but only where the task genuinely has structure.
If you remember three things
- Plan first, then execute, versus deciding step by step
- Plans must be revisable when reality disagrees
- A plan is also a thing you can show a human for approval
Overview
Planning decomposes a complex goal into ordered or conditional sub-tasks before acting. The agent lays out steps with dependencies, executes each (calling tools or models), observes results, and revises the plan when steps fail or new constraints appear — trading upfront structure for reliability on multi-step work.
How it works
- Start: User Goal The request is too complex for one direct model call.
- User Goal -> Plan The agent decomposes the goal into steps with dependencies.
- Plan -> Execute Step Each step may call tools, retrieve context, or ask a model.
- Execute Step -> Observe Results update the plan state.
- Observe -> Revise / Finish The plan adapts when observations fail or new constraints appear.
In an interview
Planning is decomposing a goal into a sequence of sub-tasks with dependencies before executing, rather than hoping one model call solves it. The agent runs each step, observes the outcome, and revises the plan when something fails. It helps on complex tasks but adds latency and can over-plan simple ones.
Production defaults
- Use when
- steps are interdependent, or a human should approve before execution
- Re-plan
- on failure or surprise, with a cap of 1–2 re-plans. Unlimited re-planning is a loop
- Show it
- a visible plan is the cheapest form of human-in-the-loop control
What breaks
- The plan is beautiful and wrong — It was made without touching reality. Cheap exploratory calls before planning beat a confident plan built on assumptions.
- Endless re-planning — Cap it. After two failed plans, stop and report rather than trying a third.