Chapter 3

Agent Core

ReAct · Tools · Intent · Context

Why

LLM ≠ Agent

Chat = 1 pass · Agent = plan + act + observe loop

ReAct

ReAct loop

Thought → Action → Observation → repeat

Example

Flight booking trace

search_flights → book_flight → confirm PNR

Tools

Tool calling

JSON schema · one job per tool · RAG as tool

Intent

Routing

QA → RAG · transaction → agent · confidence fallback

Context

Token budget

rules > state > RAG > history > summary

Lab

Build ReAct agent

2 tools + router + 10 test cases

Eval

Agent metrics

Tool accuracy · task success · SWE-bench

Agent Core done

Scroll: #agent-practice · #agent-eval

Next: Retrieval Engineering

Intro
1 / 1 Hub Space F