Who's Watching the Agents? What AI Code Does in Production

Audience
120
Slide views
29
Video views
81

Presented at GitHub Universe 2026 in San Francisco.

Abstract

GitHub Copilot and coding agents can now write, commit, and deploy code on their own. GitHub gives you control over how software is created and shipped, but once that code runs in production, seeing what it does gets much harder. Static analysis and pull request reviews look at code at rest. Neither can watch a running process.

This session shows how eBPF-based runtime observability with Tetragon gives kernel-level visibility into the processes, file access, and network activity of AI-generated workloads while they run. When agent-written code connects to an unexpected endpoint or touches a restricted file, you see it.

You’ll leave with practical ways to close the security gaps AI introduces: watching not only what agents write, but what their code actually does in a live environment.

Resources

Code