It should work now beyond that your prompt matters way more than the model I can make a crappy model do incredibly complex work or I can make a modern Frontline model stink with my prompt.
You have to tune your prompts and tool descriptions to get what you want you can’t just turn it on… Sorry.
My Friday’s Party post while pong has a lot of info about this.
Thyraz also has a post that should help where they’re using the concepts in thier own way
Itl take a lot of careful planning and instruction but it can be Fina dn worth it.