Putting agents to work: Agent-driven orchestration you can govern, trust, and repeat

The value of AI agents in infrastructure isn’t speed at any cost. It’s a capable tool that reads your code, follows your conventions, previews before changes, and gets approval when it matters. In this session, we’ll give an agent the same things a good engineer gets: context about what’s deployed and short-lived access scoped to the job.

We’ll use Pulumi’s harness that gives any capable agent the context and scoped credentials it needs. We’ll make real changes to infrastructure declarations, watch an agent plan and preview them, and see policies and approvals apply to the agent exactly as they would to a person.

You’ll leave knowing how to make agents a predictable part of your infrastructure workflow, with every change visible, reviewable, and governed.