AI agents are reaching production, but the controls around them are still catching up. From KubeCon + CloudNativeCon in Salt Lake City, this discussion examines AI-agent security, AI SRE and the governance needed to operate agents responsibly.

The opening panel looks at why sandboxing alone cannot resolve agent risk, how permissions should match the task, and what cloud-native teams still need to address.

The second conversation explores AI SRE: what changes when agents help run production systems, why runtime policy matters, and how shared context and memory affect operational decisions.

The final segment introduces the RAMP test: replay, audit, move and patch. The discussion connects those controls to platform-team responsibilities and the challenge of increasing delivery speed without losing accountability.