chapter nine
9 Advanced prompt patterns
This chapter covers
- Chain of thought: making multistep reasoning explicit when no reasoning model is available or the pattern is domain specific
- Meta prompting: generating prompt templates from task descriptions using the elements and patterns from earlier chapters
- The ReAct pattern: structuring thought-action-observation loops for agentic workflows
- Knowing when each pattern earns its added complexity and when simpler prompting wins
Chapters 2 through 8 built the core toolkit: prompt structure, language shaping, reusable templates, few-shot examples, and multigeneration aggregation. Each technique improves what a model produces in response to a well-formed request.
This chapter introduces three patterns that operate at a higher level of abstraction. Each one changes not just the content of a prompt but how reasoning or generation itself is structured.
- Chain of thought (CoT) makes intermediate reasoning steps explicit, producing auditable logic that engineers can review, challenge, and trace.
- Meta prompting uses a prompt to design a prompt, applying the framework built across earlier chapters to accelerate template creation for recurring tasks.
- ReAct (reasoning and acting) structures agentic workflows as interleaved thought-action-observation loops, giving a model a principled way to incorporate real-world observations before committing to a conclusion.