Fifty agents booted, zero secrets injected

Your orchestrator spawns fifty worker agents, each in its own container, for a two-hour crawl. On the old stack step one was secrets distribution: a vault policy, an init sidecar, an env var per pod, and one shared provider key that fifty processes could now leak.

On Hoody step one doesn't exist. AI access is on by default when a container is created. Each worker calls the gateway as itself and the host resolves which container it is. Fifty identities, no provisioning step, no key that fifty workers share. When the crawl ends you delete the containers, and fifty identities end with them.

The blast radius of one compromised worker is one worker, and it held nothing to begin with.

The workers never received a credential, so they can't lose one.