Autonomous agents are being deployed without runtime control. No open, vendor-neutral way to change what they do. No visibility into why they decided. We can fix that. A physical party game. A Raspberry Pi. An agent that adapts and evolves the game in real time. Without a single proprietary platform.

Sounds fun. But it's also the argument. We've been here before.

Every vendor built their own observability stack. We paid the price - so we built OpenTelemetry. Same with feature flags - we built OpenFeature.

Now agents are here. The control layer is emerging again, often as vendor-specific infrastructure.

Nobody designed OpenTelemetry to be an agent's eyes. Nobody designed OpenFeature to be its conscience. Turns out they work anyway: OpenTelemetry becomes observation, OpenFeature becomes control.

This talk explores agent control on open standards — physical system control, infrastructure management, live model switching, permission changes, and rollback without redeployment.

Every autonomous agent needs a control layer. Vendors are racing to own it. You already have it.