Description
People having been asking for a while for a roadmap for PydanticAI, here goes.
For the most part, I'll drink the GitHub cool ade and add the issues that represent our ordered priorities using the new sub-issue feature.
In terms of timeline, I'm terrible at predicting how long things will take, so I'm not going to dare to guess when features will land. I'll just say that PydanticAI is becoming a bigger and bigger priority for Pydantic, so we'll have a few people working full time on the library, especially once Pydantic 2.11 is released (pydantic/pydantic#11194).
Feedback here very welcome, also as per #905 I'll be in NYC and SF in the next couple of weeks, so come and say hi and demand your chosen feature in person if you can!
The philosophy here is as follows:
- build the things people actually want/need in production use cases
- build the things that have to exist in the framework (e.g. Add GraphRun object to make use of
nextmore ergonomic #833 is a big refactor, but it allows patterns that are hard or impossible without it) - break APIs ASAP so we can get to a stable state ASAP
- make the experience of using PydanticAI with Pydantic Logfire as good as it can possibly be while remaining as open and standards-based as possible, e.g. adopting the GenAI semantic conventions for OTel even if they're still in flux and difficult to use in places
- Build the sugar on top when the underlying fundamentals are right, to encourage people to adopt PydanticAI early in their development even if things like type safety, native async support and test coverage don't woo them