Free Lesson
The Anatomy of a Self-Improving Agent w/ Arize AI
Part of The AI Builders Summit
60 min
Sep 9, 2026 1:00 PM
What you'll learn
Instrument agents to surface failures
What to capture in traces and evals so a production failure is findable, not buried in dashboards
Move from dashboards to diagnosis
How observability is shifting from a human combing through traces to agent-driven investigation.
Close the loop from break to fix
The anatomy of a self-improvement loop: detect, diagnose, and propose a change.
Verify improvements before they ship
How evals turn a plausible change into a proven one, so an improvement does not regress once it is live.
Why this topic matters
As agents get more capable, the bottleneck shifts from building them to improving them. Observability and evals surface what an agent is doing, but a human still reads the dashboards and combs the traces to fix what broke. Aparna walks through how observability is moving from human-driven investigation to agent-driven execution, and the loop that turns a production break into a verified fix.
