AI Engineering: A production story, from first LLM call to multi-agent chaos

You've seen the hello worlds. This is what happens after.

In this talk, we follow the evolution of a real production AI system: from a simple chained LLM call to a full multi-agent architecture. At each step, a new requirement forces a new decision: routing, agents, supervisors, orchestrators... Each pattern unlocks something. Each pattern introduces something new to worry about.

We'll go deep on the parts that matter: context engineering in multi-agent systems and evaluating a non-deterministic system... We'll also touch on frameworks, security, and the very human challenge of shipping AI features when everyone (including you) is figuring it out in real time.

No hype, no magic. Just the decisions, the tradeoffs, and a few spoilers.