74% of organizations running AI agents in customer channels have rolled one back or shut it down. Rollback works when the change is the damage. When an agent has already sent the wrong answer to thousands of customers, the deploy reverts and the messages stay delivered. This talk covers the production patterns that keep customer-facing agents running: limiting what an agent can do before it acts, noticing when it's wrong before customers tell you, and responding when the agent itself is the incident. Some of it will be familiar from running any service. Some of it changes, because an agent that's confidently wrong still returns 200 OK. You'll leave with patterns you can apply to the agents you run today, whatever framework or model they're built on.
Gunnar Grosch is a Developer Advocate at Sinch and an engineer with over 25 years of experience building software, scaling developer communities, and making complex technology accessible. An AWS Serverless Hero and former Principal Developer Advocate at AWS, he shares code, insights, and lessons through conference talks, blog posts, and open source.
