Replies: 2 comments 2 replies
-
Great question! Since I had a prospective client ask the same question, I have looked into it today... My conclusion so far is that, despite calling itself a framework, it feels more like a library. Aside from how tool calling works, it feels like it's much closer to being an alternative/replacement for Instructor, rather than Atomic Agents, though. Now, Atomic Agents uses Instructor internally, and PydanticAI serves as a possible replacement for this, I think. I feel, though, as PydanticAI progresses more and more, it might be worthwhile to consider replacing Instructor with PydanticAI as the "internal engine" for Atomic Agents. After all, Atomic Agents is less about providing you functionality, and more about providing you with an organizational layer around Instructor's functionality. For example, how everything works with input&output schemas, how docstrings are enforced through usage of the BaseIOSchema, ... In the end, the impact on Atomic Agents will depend on whether or not Atomic Agents ever goes to other languages. Today, Instructor is available in JS, rust, ruby and python (maybe I missed some) Meaning, the logic behind Atomic Agents could be ported to any of those languages rather easily (given that I actually get the time or the help to get around to it) - This might become less easy if we adopt PydanticAI internally. I guess that'll mostly depend on how much extra benefit PydanticAI will bring over Instructor and whether it's enough to make me want to stick with that... In the end, if I can cut out a dependency, like Instructor, without losing functionality since Atomic Agents heavily uses Pydantic, that's be amazing! |
Beta Was this translation helpful? Give feedback.
-
I prefer Atomic more,pydanticai just like autogen(want's more but more complex),Atomic is more free and light |
Beta Was this translation helpful? Give feedback.
-
Hi,
Pydantic just released PydanticAI.
How does this compare with atomic-agents?
Will atomic-agents look to build on top of PydanticAI or are these two alternate solutions for building agentic AI workflows?
Beta Was this translation helpful? Give feedback.
All reactions