CHAPTER 02 · The Building Blocks of Agentic Intelligence · 5 / 5
Key takeaways
- An AI agent is built from an LLM, memory (short and long term), planning, and tools.
- Four core agentic patterns: reflection (self-critique), planning (task decomposition), tool use (external actions), and multi-agent (specialized collaboration). Reflection and tool use are the most mature; planning and multi-agent are more powerful but less predictable.
- Five workflow patterns wire these together: prompt chaining (sequential), routing (classify and direct), parallelization (concurrent independent work), orchestrator-workers (dynamic delegation), and evaluator-optimizer (generate-and-critique loop).
- Every agentic RAG architecture in Chapter 3 is just a particular combination of these building blocks.