The demo always works. Production is where agents break — and they break in ways traditional software testing never prepared you for. The same prompt returns a different answer twice. The agent confidently answers outside its grounding. An action fires against the wrong record. There’s no test harness because there’s nothing deterministic to assert against. This session is the honest field guide to operating agents once the applause dies down.

We’ll walk through the failure modes that actually show up in production and what to do about each: testing behavior that won’t repeat, guardrails that keep an agent inside its lane, action safety, and the monitoring and feedback loops that tell you an agent is drifting before your users do. We’ll close on the governance gates worth enforcing before go-live. You’ll leave able to take an agent that demos beautifully and make it something you’d actually put in front of real users.