From Signal to PR: Anatomy of a Self-Improving Agent — Jason Lopatecki, Arize

Signal, Arize's debugging agent, is designed to assemble evidence about application failures before an engineer starts investigating. In Arize's talk, it combines source code with logs and traces, which record the steps an application actually took. That helps connect a reported error to the execution that produced it.

The described workflow runs periodically or in response to events and can open an issue containing its findings. The speaker reports a small fix for a canceled-stream error, while explaining that larger repairs often still need a person to guide the investigation. The supported benefit is a better-informed starting point for debugging, with human involvement continuing through difficult fixes.

“observability used to be for humans. Used to be a UI you click, a graph you click, something you look at”

Jason Lopatecki

“You can build at agent speed, but today you can't improve your systems really at this agent speed.”

Jason Lopatecki

“your job kind of moves from responder to reviewer”

Jason Lopatecki

“Getting that data into the repo in a file format. These harnesses are magical with files.”

Jason Lopatecki

“by logging and tracing orders and orders of magnitude more than we do today, we can actually create these continuous loops”

Jason Lopatecki