ZenML puts TypeSafe's Jev in production as an evidence reviewer for its SRE agent

strickvl · x · 2026-09-22

ZenML is testing TypeSafe's Jev model as an evidence reviewer for its SRE agent—checking whether the agent's conclusions are supported by collected diagnostics—with mixed results so far. Kitaru PR #1138 adds configurable yes/no, choice, and ordered-level judge questions for recorded sessions without custom evaluators; deterministic checks stay for exact rules, provider calls stay out of the offline bundle.

Original post →

More from coding & agent

coding & agent channel →