You use Claude Code or GitHub Copilot every day: pulling in a pull request, having it draft a plan, letting it apply a fix. Handy, but everyone on your team does this separately, by hand, over and over again. What if you turned that workflow into an agent of its own that does this without you, triggered by, for example, a failing pipeline?

Using a concrete example, an agent that automatically resolves breaking changes from Renovate updates, I’ll show you how to build such an autonomous agent. We’ll walk through the building blocks of your own “agent harness” using a code-first framework: orchestration, observability, guardrails, and hosting. For each component, I’ll translate what you already recognize from your IDE agent into how it works when an agent has less supervision and more autonomy.

With autonomous agents like these, you can take the automation of your development work much further than what you can do with Copilot or Claude alone. After this talk, you’ll know how to start building this kind of agent yourself.