Standard progressive delivery is an essential safety net, but in most cases, a rollback is just the beginning of a long investigation. This session explores how to evolve from simple automated rollbacks to a self-healing release cycle by integrating an agentic AI system via the A2A protocol.
We will demonstrate an architecture where specialized agents analyze canary logs and metrics in parallel to provide high-fidelity health checks. If an issue is detected, the system immediately signals a rollback while simultaneously launching an asynchronous remediation workflow. This "coding agent" analyzes the specific error against the app's source code to either create a detailed issue description or propose a fix in the form of a pull/merge request.
To maintain performance and keep costs low, the system utilizes a hybrid model strategy. We run Small Language Models (SLMs) directly on a Kubernetes cluster for fast, local analysis and escalate to larger frontier models only when complex reasoning is required. Attendees will learn how to build this modular system and get access to the open-source implementation to start automating their own remediation loops.