article --summary
The summary
Jonathan Bryce, head of the Cloud Native Computing Foundation, says existing systems were designed for human-paced use and cannot yet handle the volume and rhythm of AI agent workloads. He believes the cloud-native community should build security, control, and elastic scaling into the infrastructure itself.
Nvidia's OpenShell and Google's Agent Substrate both joined the CNCF as Sandbox projects. OpenShell enforces policies over what files, networks, credentials, and model endpoints an agent may reach, while Agent Substrate isolates agent sessions, suspends idle ones, and restores them on available workers to avoid wasting compute.
Bryce also points to efficiency work such as llm-d, a distributed inference framework, as a way to get more from scarce hardware. He notes that security teams are now dealing with floods of duplicate, unverified vulnerability reports produced by AI tools, and that agent adoption outside software development is still low, which may give the community time to harden these projects before enterprises rely on them broadly.
Editorial note: This is a summary of an article first published by The New Stack. The full piece is on their site.