Interoperability between Super Agents
future
K
Kam A
allow super agents to communicate with other super agents as another category below triggers, tools, and Knowlege & Memory. I'd love for one agent to be able to ask another agent questions for what it may have access too where the first one does not. This allows for comment tracing within the ticket in case an agent deviates in the conversation.
Log In
D
Dirk Riesenbeck
It works, I discovered it by chance. When a chat is started with a super agent, other agents can be invited in your threat with @... superagent 1,2,3, etc. They then have access to the entire threat. You will then have a kind of mini team room. That works pretty well, I've just made a contribution about having a team room created. Maybe something will come of it
Nicolas Potier
It would be interesting if a higher-level agent (executive assistant) could delegate tasks to specialized agents for specific uses or with a restricted scope of action (spaces, files, etc.) In the same way, agents at the same level could ask for summary information on territories in which they cannot intervene (e.g. CRM agent asks the accounting officer for information).
D
Dirk Riesenbeck
Nicolas Potier Hello, that's okay. This is due to the allocation of rights. Agents need clear instructions on what they can or cannot, and in the profiles, you can narrow down or expand the work areas.
Michael Van Doorn
marked this post as
future
Michael Van Doorn
Hey, everyone!
We'd love to hear more about how your use cases and how Super Agents would work together in your Workspace. Can you share more details on what you're doing and what you'd like to see?
Thanks!
Oleksii Segodin
Michael Van Doorn, here is the use case I would like to have:
- A higher level ("A") agent (which has not so deep but broader knowledge about the company) reads my emails and intelligently decides which project current email belongs to.
- Then agent A gives an order to lower level project manager ("B") to work on something stated in the email and/or just resends the email to him.
In this case agent A acts like a smart router for emails and tasks in the company. And while there is an AI Assignee feature, I'm not very confident that it may handle everyting correctly always, because I may have a big list of fine-grained instructions on how to process specific emails and some other metadata which should sit in the Super Agent's memory.
Oleksii Segodin
Experienced the above issue with AI Assignee just now. I instructed it to assign a certain agent on any emails which include invoices in attachements. It understands if attachment has "invoice" word in the filename, but doesn't recognize it when it's named differently. To recognize invoice it must actually look into it and read PDF/Doc/image.
Kyle Hannon
Michael Van Doorn I simply would like for agents to recognize when another agent has already answered a message or question. For instance, we have an agent specifically for one type of request, and after it answers a question, the default Ambient Answers agent re-answers the question, quoting the specialized agent which has already answered it.
I already have gripes about how the Ambient Answers agent cannot be customized, given instructions or memory, or turned off universally, and how it seems to "wait" for someone else to answer the question, then parrots back that answer in the same thread, but I'll settle for not overlapping with existing specialized super agents so my channels aren't littered with the same information rephrased several different ways.
D
Dirk Riesenbeck
Hi that should be narrowed down. It's just like us humans, new employees don't know our processes and ideas and have to get used to it first. If mistakes happen or an agent forgets something, I explain to him what the mistake is and how it shouldn't work and also the company's goal on the subject, or even briefly the solution approach usually with the addition to your rules so that he doesn't forget it. This has worked out very well for us so far, and in the case of parroting, tell the agent that someone else has reported something that he should not just repeat it, that this is not effective with rules, then the problem should be resolved.
Kyle Hannon
Dirk Riesenbeck At the moment, the Ambient Answers agent cannot be given instructions or rules. It simply automatically triggers when it detects a question, regardless of if the question has already been answered, if the question is directed specifically at a human team member, or if another super agent is tagged. You can "thumbs down" and provide feedback all you want for a specific answer, but it will not remember guidelines that you give it.