← AI Terminology
Planning (LLM)
Planning in LLM agents is the process of decomposing a goal into ordered steps, subgoals, or tool sequences before or while acting.
It ranges from implicit chain-of-thought to explicit planners and classical AI planners in the loop.
It ranges from implicit chain-of-thought to explicit planners and classical AI planners in the loop.
Why It Matters in AI
Complex tasks fail when models act myopically. Explicit plans improve multi-step reliability, enable user approval of steps, and structure tool use. Planning is the difference between a chatbot reply and a competent agent.
Key Points
| Aspect | Description |
|---|---|
| UX | Show plan for confirmation in enterprise agents |
| Styles | Plan-then-act, interleaved plan/act, hierarchical goals |
| Benefit | Fewer missed dependencies; clearer control |
| Failure | Brittle plans; plan not updated when world changes |
| Methods | CoT, ReAct, ToT, LLM+PDDL, task graphs |
| Related | Reflection, multi-agent delegation |
Simple Analogy
Writing a travel itinerary before booking trains and hotels — sequence the trip, then execute, revising if a train is full.
Common Usage Examples
- “Make a plan, then execute step by step” prompts
- LangGraph / state-machine agent flows
- Classic planners + LLM translators
- User-approved plan checkboxes in copilots
Summary
In short: Planning decomposes goals into steps so agents act with structure instead of one-shot guesses — essential for multi-step work.