Why agents are hard to observe

A traditional request is mostly deterministic: same input, same path, same output. An agent turn is none of those. The same prompt can take different paths (different tools, different order), the model's output varies run to run, and the quality is subjective ("was that a good answer?"). So agent observability needs to capture not just that something happened, but the full trajectory: what context went in, what the model decided, which tools ran with what inputs and outputs, and what came out.