Pydantic By the Bay

Did you know that Pydantic is downloaded nearly 300 million times every month? It began as a weekend experiment, sparked by a simple but persistent frustration: “Why don’t type hints actually do anything?”.  Samuel Colvin, our keynote speaker this year, set out to see if they could—and it turned out they could do quite a lot. What started as a small project has grown into one of the most widely used Python libraries today, quietly powering some of the most critical tools in AI and software infrastructure.

RSVP

What is Pydantic?

At its core, Pydantic is a Python library that validates and parses data using Python type annotations. It allows developers to define the structure of their data using familiar syntax—then automatically checks, coerces, and enforces that structure at runtime.

This means you can trust that your application is working with the right data, even when inputs come from unpredictable sources like APIs, user input, or LLMs.

Pydantic is one of the most-installed packages on PyPI and is now deeply embedded in the GenAI stack.

It’s a core dependency for:

• OpenAI SDK

• Anthropic SDK

• LangChain

• LlamaIndex

• Auto-GPT, Instructor, and many emerging AI-native frameworks

  
  

In these tools, Pydantic handles tasks like schema enforcement for LLM outputs, structured function calls, and safe interaction with external APIs. Without it, a lot more code would break—or be harder to reason about.

With the release of Pydantic v2, the library was rewritten in Rust, delivering major performance improvements—up to 50x faster in some cases—while keeping the same Pythonic developer experience. It's fast, reliable, and now mature enough to serve as infrastructure across a wide range of applications.

Currently, Samuel and his team are working on Pydantic ecosystem, expanding beyond validation with Pydantic Logfire, a new observability tool designed for modern AI-native applications.

Logfire focuses on developer-first experience—fast setup, low overhead, and a more intuitive interface than traditional observability platforms. The backend is powered by Apache DataFusion, chosen specifically for its flexibility and performance, and implemented in Rust like Pydantic v2.

The goal isn’t to replace existing tools—it’s to give developers the visibility they actually need, without the complexity that usually comes with it.

What makes Pydantic remarkable isn’t just its speed or scale—it’s that it makes doing the right thing easier. It rewards best practices. It turns type hints into guarantees. And in an era where data, AI, and distributed systems are growing more chaotic by the day, that kind of reliability is priceless.