Question 1

What does good AI agent observability look like?

Accepted Answer

It looks like one timeline for the run: user message, tool calls, approval events, resumes, and final outcomes. That is much easier to reason about than distributed logs alone.

Question 2

Why is observability especially important for production agents?

Accepted Answer

Because production agents can take actions. When something goes wrong, teams need to know exactly which run, which tool call, and which approval event led to the outcome.

Question 3

Should observability live inside the product or in external tools?

Accepted Answer

Both matter. The product should provide the run-native operational view, while exports and APIs let teams send the same data into their broader logging or security stack.