When development velocity increases 10x through AI-assisted coding, operations becomes the bottleneck — not because we're doing it wrong, but because our tools weren't designed for this pace.
Traditional Infrastructure as Code won't work with AI agents. It's dominated by taste and principles, with no architectural constraints for agents to follow. Teams that build software with AI have learned this the hard way: architecture leads to stability, vibes lead to chaos. The same pattern applies to operations.
We need automation designed for agents as the primary consumer: transparent type systems they can introspect, extensible primitives they can compose, schemas that match the domain (not custom abstractions), inline validation (not after-the-fact review), and coverage across the full operational lifecycle — not just provisioning.
In this talk, you'll see live demos of AI agents using swamp — an AI-native automation CLI — to build workflows from intent, discover production state through typed models, and compose operations through versioned data. Then we'll discuss what this means for the SRE role: the evolution from operator to architect, from reactive to proactive, from tribal knowledge to encoded knowledge.
You'll leave with a framework for evaluating AI-native tools and concrete steps to start encoding your operational knowledge into typed, composable, executable models. The people who figure this out will define what operations looks like for the next decade.