Overview

Production agents fail in specific, repeatable ways. Infinite loops. Context wipeouts. Hallucinated tool args. Retry storms that DDoS your own APIs. Silent model quality drops after provider updates. Prompt injection through RAG. State corruption across users. Token runaways that turn into a $5,000 weekend. If you're a production AI team building or running autonomous systems, this book is built for one job: detect failures early, recover automatically, and keep the system operational without waking a human. This is not a theory book and it's not meant to be read cover-to-cover. It's a field manual-a catalog of 100 failure modes with pragmatic recovery logic, defensive engineering patterns, and operational heuristics you can apply under real constraints. You jump to the failure that matches your symptoms, stop the bleeding, then harden the architecture so it doesn't recur. What's inside (failure-first, production-focused): Detect infinite loops and dead-end delegation Contain hallucinated tool arguments safely Prevent destructive or unsafe tool actions Stop retry storms and self-inflicted outages Design crash-compatible state and recovery Harden RAG against prompt injection paths Control costs, timeouts, and latency cliffs How you'll use it Monday morning: Start with the failure mode you're already seeing-""context window exhaustion,"" ""session state corruption,"" ""streaming partial JSON,"" ""connection exhaustion,"" ""model deprecation,"" ""vector drift,"" or ""confused deputy."" Treat each chapter as a standalone diagnostic unit: identify the mechanism, assess the risk, apply the remediation, then convert it into an automated defense your system enforces by default. Who this is for: software engineers, DevOps/SRE, and production AI teams operating agentic systems in high-stakes environments-where downtime, unsafe actions, or runaway costs are unacceptable. If you're responsible for an autonomous system in production, you don't need more optimism-you need defenses. Buy this book and keep it within reach of your on-call rotation: the next time an agent starts crashing, looping, leaking, or lying, you'll have a failure pattern to match-and a recovery plan to ship.

Full Product Details

9798242186729

Table of Contents

Reviews

Author Information

Tab Content 6

Countries Available