Planning techniques let a model break a goal into steps before or while acting, from chain of thought prompting to plan-then-execute architectures and built-in reasoning modes. Plans and reasoning traces help with review and audit, but they are model output too and can be steered by injected content.
Read in this order
- OfficialExtended thinking (Claude Platform Docs)Start here: how built-in reasoning works in a production API.
- PaperChain-of-Thought Prompting Elicits Reasoning in Large Language ModelsThe paper that showed step-by-step prompting improves model reasoning.
- ArticlePlan-and-Execute Agents (LangChain)Plan-then-execute designs, which separate planning from acting for tighter control.