Chapter 2. The Harness as the Stateful Agent System Substrate
In the last chapter, you learned that the model provides reasoning intelligence during inference, but it can only propose what should happen next. Your AI agent harness governs the agent system around that decision: it determines what information reaches the model, which tools it can call, whether a proposed step executes, what happens after an action completes, and whether the resulting state can move forward. In addition, Figure 1-1 gave you an overview of the agent harness, showing you which layers you author, own, and improve.
These components may live in different parts of your system, across separate services, or even in remote infrastructure. The challenge is that they still need to communicate and interact in ways that preserve the intended control flow across your harness. What one component produces can determine what another component sees, which actions can execute, how state changes, whether verification can establish ...
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