Engineering AI Agents Design and Ship Single- and Multi-Agent Workflows That Survive Production [#1042165]

Engineering AI Agents: Design and Ship Single- and Multi-Agent Workflows That Survive Production
English | October 14, 2026 | ASIN: B0HN1JP3YM | 400 Pages | Epub | 3.15 MB
The demo worked. Then the agent met production.It handled twenty claims flawlessly in front of the leadership team. In its first real week it paid the same customer twice, looped through thirty-one tool calls on a single request, and forgot every task that was waiting for a manager's approval when the server restarted. The model was fine. The engineering around it wasn't.Engineering AI Agents is a production-first guide for senior developers and tech leads who have been asked to put an agent into a real product. It shows you when an agent is the wrong answer, how to build the tool-calling loop and the core patterns in plain code, and how to make single- and multi-agent workflows durable, approvable, retry-safe and deployable.Written by an engineering manager with more than 15 years of building and running production systems in e-commerce and logistics, this book treats agents as what they are in production: distributed systems with a language model inside. Patterns first, frameworks second, hype never.Inside, you'll learn how to:Decide between an agent and a deterministic workflow with a framework you can defend in a design reviewBuild the tool-calling loop from scratch, with stop conditions, step limits and structured tool resultsApply routing, planning and reflection where they pay off, and skip them where they don'tMake agents durable: checkpoints, journaled steps, idempotency keys, and runs that survive crashes and deploysDesign multi-agent systems with a supervisor, specialist agents and handoffs, and know when one agent is betterAdd human-in-the-loop approvals that can wait for days without losing stateHandle failure on purpose: retry budgets, backoff, fallbacks, compensation and sagas for side effectsChoose between frameworks and your own code with an honest decision matrix and an exit planDeploy agents as workers and queue consumers, version them, and roll out model changes safelyOne running project ties it together: a fictional parcel carrier's claims agent that reads tracking data, checks contracts, asks partners to trace lost parcels, waits for approvals and issues credits. Every chapter adds one piece, in short, runnable Python you can adapt on Monday, and the final chapter assembles the complete system.No framework lock-in. No autonomous-future speculation. Every number in the book comes with the arithmetic behind it.If your agent works in the demo and you need it to work on a Tuesday night in production, this book is for you.Get your copy and ship an agent that survives its first week.