Papaya.

Papaya examines production agent traces to identify recurring problems with prompts, context, tool use and model choices. It groups runs by workflow and ranks proposed changes by their expected effect on quality, latency and cost. Engineers can inspect the supporting runs, approve a fix and receive a pull request, while monitoring whether the change holds up in later traffic.

Engineering teams spend time manually reviewing traces to find agent failures and performance regressions.