The Agent Broke It, The Agent Fixed it

Not that you would do this of course, but we all have "a friend" who may not review every line of code produced by their trusted coding agent. At the same time, again not you, but another "friend", may not review every single line of code before it goes to production.

Despite our best efforts in reviewing, testing and perhaps even letting AI agents verify our code, sometimes production releases don't go as planned.

In this session, we'd like to demonstrate how we came up with one way to ease production rollout pain. We created an agentic analysis and remediation system that ties in to a tried-and-true progressive delivery solution (Argo Rollouts) as a plugin featuring agents that:
* Fetch and analyse canary and stable logs/metrics and other sources
* Make intelligent decisions on release quality
* Make code corrections based on the failure analysis and create automated PRs and/or issue descriptions.
* Embrace Human-in-the-Loop oversight by allowing developers to review and course correct if needed.

This strategic combination of agents leads to significantly reduced Mean Time To Recovery (MTTR) and a more resilient, productive developer experience. And all open source!