AI agents don't belong on developer machines

AI agents are becoming a normal part of how developers work, but where and how they run matters more than most teams realize. This talk explores what "good" looks like when agents run on remote compute: isolated execution, clear boundaries, parallel execution, and predictable behavior that platform teams can actually support.

It walks through the core principles behind running agents this way without assuming deep AI expertise. Attendees learn how remote execution improves both security and developer experience, why isolation enables faster iteration instead of slowing it down, and how platform teams can deploy coding agents that are reliable, scalable, and easy for developers to adopt.