AI development is still software development, just with some uniquely frustrating twists. You don't need to throw out everything you know about engineering, but you do need better patterns for building, debugging, and actually seeing what your AI is doing.

In this talk, we'll walk through Pydantic's opinionated approach to building reliable AI applications in Python, combining practical engineering patterns with what we call "human-seeded evals", a way to create meaningful evaluation systems that start small and scale up (without needing a PhD in data annotation).

We'll show you how to build AI apps with Pydantic AI that don't fall apart in production, create evaluation systems that start with just 5-10 examples and grow from there, and use Pydantic Logfire to actually understand what's happening under the hood. Live demos, real code, and patterns you can use tomorrow.