AI Agent Governance

Chapter 2. Failure Patterns in Autonomous Agent Systems

A note for Early Release readers

With Early Release ebooks, you get books in their earliest form—the author’s raw and unedited content as they write—so you can take advantage of these technologies long before the official release of these titles.

This will be the 2nd chapter of the final book. The GitHub repo will be made active later on.

If you’d like to be actively involved in reviewing and commenting on this draft, please reach out to the editor at [e-mail].

Chapter 1 left the AIOps Assistant running in production with four tools and real system access. It also left three things missing. The assistant acts under a shared service account, so in the logs its work is indistinguishable from every script and cron job that uses svc-platform-automation. It writes no durable record of its runs; once a run finishes, piecing together what the assistant did means combing through log files and Slack messages. And nothing sits between the model deciding on an action and that action reaching production. Run #4288 escalated instead of remediating because the model chose caution during that run, nothing in the loop required that choice. 

This chapter shows what those absences cost. The failures ahead develop while the agent runs exactly as configured. Typically, when production breaks, you look for the code change or config push and roll it back, but in agentic systems, they emerge during the normal operation of an agent without ...