Agents/Tools that act, not talk

[TantalusVCV]

Agents are amazing, but they appear to be limited to the paradigm that their purpose is to provide an output of information. I'm looking for a way to build an Agent and Tools whose purpose is to perform actions. Such an Agent could use its reasoning capability to determine the proper steps to take then activate Tools designed to, say, control other apps, print out documents, potentially even control various types of hardware. This would be useful well beyond just the ability to provide information.

I've had some luck creating Tools that can do some of these things, but the limitation I keep running into is that the Agent is always expecting a substantive information result from the Tool, and insists on sharing that with the user. For instance, I can create a Tool that pulls a SQL result, then provides it to the user as a report or saves it to file, but it has to respond back to the Agent when it's done. If I set it to simply reply with "done" or "success" or something similar, I can't seem to get the Agent to understand that it doesn't need to do anything - it typically will then hallucinate an answer to the original request and present that as "the final answer". What would be best is for it to simply accept that the tool completed the requirement and stop.

Anyway, my question is: does anybody have ideas about how to achieve this, or know of some work already done in this area? I can conceive of a whole new family of Agents that are performative rather than informative, and potentially Agents that are both as needed. That would be awesome, yes?