Understand what separates an AI agent from a chatbot or a workflow. Learn the ReAct (reason + act) loop end-to-end. Build a minimal agent with the Anthropic SDK in under 50 lines of code.

Master tool use with strict schemas, parallel calls, and graceful error recovery. Manage conversation state and context window. Add long-term memory using vector embeddings and semantic recall.

Split complex goals into subtasks with planner-executor patterns. Build RAG pipelines with chunking, hybrid search, and re-ranking. Add reflection and critique loops so agents catch and fix their own mistakes.

Coordinate multiple agents with supervisor, swarm, and hand-off patterns. Build state machines with LangGraph. Compare CrewAI and AutoGen for role-based, sequential, and parallel crews.

Build golden datasets and run evals with LangSmith and Arize. Defend against prompt injection and ship safe outputs with guardrails. Deploy with caching, model routing (Haiku/Sonnet/Opus), and latency budgets.