Production-Grade AI SRE vs. ‘It Works on my Laptop’: What’s Missing & How to Add It

Building a local AI agent that can troubleshoot a Kubernetes cluster is the easy part. Turning it into something enterprise teams actually run — serving hundreds of users simultaneously, against infrastructure you don’t own, holding up when things go wrong — is a different discipline, and surprisingly little of it is about the AI itself.

We’ve spent the last few years building and operating a production-grade AI SRE at Komodor, and this session distills what we’ve learned. We’ll cover monitoring and tracing an agent whose behavior is never quite the same twice, creating a lab for measuring agent quality (checking if the agent “actually works” and proving whether each change is an improvement), and the other unglamorous pieces required to take a locally working prototype to a prod-grade system people can depend on.

You’ll leave with an experience-backed picture of what “production-grade” actually demands from an AI SRE — the considerations to weigh whether you’re building your own or judging someone else’s.