This seems promising - more than anything I like that there's very little magic. With that said I honestly find it significantly easier to reason about / debug traditional control flow logic. As soon you introduce these frameworks it raises the complexity of your app 10X.
This might be warranted when you have swarms of agents but I don't think anyone believes the models are close to that level of capability yet. But the LogFire integration does seem useful.
Just for fun I rewrote their example using a basic query classifier + regular control flow. Personally I find it much easier to reason about.
Interesting, but this doesn't support multiple tools with different arguments, or an arbitrary number of steps in the conversation. If you add that, you basically get PydanticAI.
This can pretty easily be generalized to any number of steps - you're just passing the entire conversation when routing the query. I agree it's more constrained - intentionally. With each step being stochastic, it becomes incredibly challenging to troubleshoot agents when they're doing unconstrained work. Regardless, I look forward to trying this out at some point.
This might be warranted when you have swarms of agents but I don't think anyone believes the models are close to that level of capability yet. But the LogFire integration does seem useful.
Just for fun I rewrote their example using a basic query classifier + regular control flow. Personally I find it much easier to reason about.
https://gist.github.com/Filimoa/bd883bbbd6475993f3f6d2621275...