AI assistants and agents write a growing share of production changes. Code reviews and tests cannot keep pace with that volume, and regulators expect AI systems and critical changes to stay under human control. This session treats reversibility as a platform service: every change, from a person or an agent, ships behind a runtime control with a tested way back. Live demos show an agent wrapping its own change in a flag and a rollout that stops when metrics drift, or a kill switch that moves an AI feature to a rules-based fallback, with the audit log recording who pressed it. All of the above are presented as structured recommendations and best practices they can apply to their platform.

Level: 200 (Intermediate)
Category: Agentic AI
Audience: Software Architects, Software Engineers, DevOps/Infrastructure Engineers
Projects: Kubernetes, OpenFeature, OpenTelemetry, Unleash (open source), MCP