Toolcall planning

[clojj]

Several statements here
https://spring-ai-community.github.io/spring-ai-agents/#_the_paradigm_shift
make me wonder, if embabel's planning feature relates to latest/upcoming models?

Maybe I don't understand that overview's context and it is not talking about "agents vs. reasoning-models", any opinions?

Anyway, the Actions of an Agent can do anything/a lot more than mere Tool calls, and I guess that Agent frameworks provide determinism, no matter what (tool related) abilities Models may have (?)

[johnsonr]

I strongly disagree with that content and the whole direction of Spring AI Agents, which I think is completely wrong for the JVM.

I think there are tasks that are well suited to this kind of heavy tool-calling model autonomy. Coding is an obvious example. However, I think it's unsuited to critical business tasks, where explainability is vital and determinism paramount.

This approach is essentially untestable and reliant on God models, whereas deterministic planning can allow the use of multiple (potentially local) models and the safe execution of updates against critical data.

Even Open AI seem to recognize the limits of model-driven autonomy with the introduction of AgentKit.

[dvag-joerg-winter]

After re-reading and seeing that I misunderstood these vague opinions, I totally agree with you.