Chapter 1. Why Harness Engineering Matters
Over the past years, we’ve all watched models get better and perform better across various benchmarks. But similarly to us humans, an IQ test (or intelligence alone) doesn’t determine whether we can actually complete a given task reliably. If you think about it, even exceptional and intelligent engineers depend on their surrounding system: repositories, documentation, libraries, test suites, execution environments, review processes, permissions, deployment rules, and feedback from users and production. These structures define what you can see, determine which actions you can take, help you detect mistakes, and turn proposed changes into reliable software.
AI agents operate in a very similar way. The underlying model may propose a plan, write code, choose a tool, or interpret a result. But the model doesn’t independently determine which context it receives, which tools it may call, what state survives the next step, where its code executes, how an output ...
Become an O’Reilly member and get unlimited access to this title plus top books and audiobooks from O’Reilly and nearly 200 top publishers, thousands of courses curated by job role, 150+ live events each month,
and much more.
Read now
Unlock full access