Skip to content

Welcome

A robot and a human pair-programming at a shared monitor

Thanks for checking out this talk.

This site was prepared by Cody Lee Cochran for Texas Linux Fest 2026, and it is the talk itself - there is no separate slide deck. Everything projected in the room is also a page here, so you can read it again afterward, link to it, or copy straight out of it.

The purpose of this talk is to show what it actually takes to let an AI agent change your infrastructure repository safely - not just write code that works once, but code that survives update, test, and teardown too.

The project driving every example is a small demo application, apps/demo-app, managed end to end by a generic lifecycle dispatcher called appctl. It is deliberately boring: one container, one port, one env file. Every point in this talk is made by pointing at real code in that repository, not a diagram of one.

If apps/demo-app ever grows into "the reference implementation of agentic infra lifecycle management," that is a problem for a future talk.