Agent Engineering with Pydantic + Graphs — with Samuel Colvin

This podcast interviews Samuel Colvin, creator of Pydantic and Logfire, focusing on his newly launched Pydantic AI framework. The discussion covers Pydantic's origins, its unexpected adoption in the AI community, and the design choices behind Pydantic AI, including its use of type hints and graphs for building type-safe agent workflows. Colvin also discusses Logfire, his observability tool for LLMs, highlighting challenges in building a robust database and the importance of OpenTelemetry for standardization. A key takeaway is that Pydantic AI aims for production readiness and high engineering quality, contrasting with some existing agent frameworks. The interview concludes with a discussion of Pydantic.run, a Python browser sandbox for easier demonstration and use of Pydantic AI.