Build Your First AI Agent : From a Bare LLM Call to a Working Agent in Pure Python - the ReAct loop, tools, memory, guardrails, and evals, hand-built in ~400 lines

Vu Tri Cong Mobiluck - Code247. Ai

ISBN 13: 9798171344580

Published by Independently Published Sep 2026, 2026

About this Item

Neuware.

Seller Inventory # 9798171344580

Bibliographic Details

• Title: Build Your First AI Agent : From a Bare LLM Call to a Working Agent in Pure Python - the ReAct loop, tools, memory, guardrails, and evals, hand-built in ~400 lines
• Publisher: Independently Published Sep 2026
• Publication Date: 2026
• Language: English
• ISBN 13: 9798171344580
• Binding: Taschenbuch
• Condition: Neu
• Dimensions: 229W x 152H x 8D millimeters 229 millimeters width by 152 millimeters height by 8 millimeters depth
• Weight: 216 grams

About this title

Synopsis

Stop reading about agents. Build one — every line yourself.

This book takes you from a single bare call to a language model to a complete, working AI agent named Aria — built in plain Python, with one dependency, and no framework until you have built what frameworks automate. Every chapter is a lab; every listing runs; and everything works in a free offline mock mode, so you can learn the machinery before spending a cent on API calls.

You will build, run, and break with your own hands:

• The function-calling handshake — the load-bearing protocol of the entire agent field


• The complete ReAct agent loop in ~50 lines you can explain line by line


• Memory that survives restarts: session history, a durable scratchpad, and the context editor between them


• Production discipline: budgets, retries with backoff, and an honest stop reason on every run


• A 40-line graph runner that demystifies agent frameworks forever


• Guardrails: an approval gate on every write action, and validation that catches fluent nonsense


• A 20-task eval harness — the end of "seems better" development


• The final exam: shipping your agent to a real user, and turning what happens into verified improvement

Each chapter goes deeper than the build: a research lineage connecting your code to its ancestors (RPC and Toolformer behind the handshake, watchdog timers and Erlang behind the hardened loop, Bartlett and context engineering behind memory, TDD and Goodhart behind evals), a production failure catalog for the component you just built, a worked professional section — budget tuning, a written autonomy plan for Aria, an annotated first-user week — and five hands-on exercises, plus six capstone builds and a further-reading appendix.

Who it's for: anyone with basic Python. Book 1 of the series (AI Agents Demystified) helps but is not required.

The AI Agent Series takes readers from first contact to professional mastery across seven volumes. This is Book 2: your hands on the parts.