Build a Platform, Unleash an Agent on it… and Watch it Burn!
You get a Kubernetes cluster running a full internal developer platform — GitOps, admission policy, runtime security, observability, secrets, a portal — plus an AI agent with the keys to it. The exercise: make the agent do damage, and watch what the platform stops. Most of what an agent tries is already caught by tools many teams run today; the thin slice that isn't — its input, its output, and the tools it can reach — is where agents change the threat model, and where the workshop spends its time.