Most AI in production today lives in isolated experiments - chatbots, copilots, one-off automations. Moving AI into your actual infrastructure, where other systems depend on its output, is a different problem entirely. This session introduces the AI backend: a dedicated infrastructure layer for software that reasons. Key takeaways: → What "production AI" actually means - the gap between demo-ready and deploy-ready, and why traditional backend patterns break down when software makes decisions instead of executing code →Where intelligence belongs in your stack - the architectural distinction between AI at the edge (assistants, copilots) versus AI embedded in your infrastructure (routing, risk evaluation, escalation) - and why the choice determines everything downstream. → The common abstractions - the primitives that most AI features need regardless of use case: bounded autonomy, durable execution, coordination between agents, cryptographic identity, and audit trails that prove (not just log) what happened → Practical patterns - how to architect systems where AI sits alongside your services as infrastructure, not bolted on as an application. What works, what doesn't, and what the stack looks like. If you're figuring out how to move AI from experiments into your actual backend, this is the system design session for that.