The problem is that most business use cases are non-axiomatic; you cannot auto-prove them, even with the vast knowledge of an LLM.

The current discussion about AI agents focuses heavily on the tech, not the use cases. It is a classical tech push vs. market pull situation (which often happens during the early adoption phase).

Much more research from a business perspective would be interesting. Otherwise, there is a risk of trying to solve potential problems without success.