Does Agent generator need to also be top of hierarchy?

[JennyHorner]

I'm wondering if it can be structured differently? In my 'team' I have Luna who I think of as being like an AI staff recruitment agent, she can create personas with reasonably complex personality, skills, backstory and also create teams with a shared task who complement each other, wear different hats in meetings and have different job roles. But she doesn't have any other role. She doesn't supervise other 'staff' and she isn't present in team meetings, it's more like she's Freelance. Fatima is our Chief Operating Officer and also an ESG expert, which means she keeps an eye on ethics. She can call a team meeting (which worked better before recent updates). For this type of thing I'm building on Professor Synapse ideas but with a stable AI workforce. Below Fatima there's a loose system of subteams eg a marketing sub team with Seth the Marketing Manager, Jamal working on Social Media and Akshay the writer. So that's just 3 tiers, with Luna as a separate Freelance entity.

I don't know much coding/github type stuff, but since AI went mainstream a few months ago Ive been working on low code ideas for this type of project. I'm in the process of adding the team as GPTs - the GPT creator says I can add a whole team of personas within one GPT but so far I haven't got it to work. I'm also intending to move Luna to GPT (she's currently an Assistant) and I could make her available via link.

[daveshap]

Agent Generator is a function that each agent should have e.g. every agent can spawn more agents if needed. That's how you get exponential proliferation. But we also need principles e.g. if you're tier 5 you can't spawn any more agents. Something like that.

[dusdb1]

What about to separate the Generation of Agents task to a specialized entity that has all the necessary (profile or skills) to do that.
Like in real life, for ex.: if in a company the IT Department Director requires a new software engineer for a new project, it will coordinate and rely on HR Dept to manage all the recruitment process.
That way, we move apart this critic task so not mixing with core agent functions.
In my projects i use to call it "bot-factory" and receives as input the bot (agent) main profile and tasks that has to do, so the factory that does know a lot (specialized prompt engineering knowledge base) how to construct bots (agents) proceeds to create it.

[daveshap]

Drone Factory. Not a bad idea and worth considering and trying.

[Czajnikuss]

IMO Agent Generator would be a freelancer... so any Agent may request a ne agent created and requestor becomes an executive for its own small project.

And I love the permanent workforce approach that You re using, this we allow us to iterate on agents and improve them in time, not sur about human names though... on the other hand if we would implement different personality traits this makes sense but is not as easy to chose from ... let say a Sarcastic Sociologist, or Friendly Coder maybe something like that and then we may see does a Friendly Coder perform better then Sad Introvert Coder... I wander what the difference would be...

[JennyHorner]

I'm not sure because my ones all have different personalities and different roles. It's an interesting question, perhaps I'll add an additional one, same role and 'qualifications' but different personality.

But I do notice that even within ChatGPT when they're role playing within the same response (same training etc) they will give different quality answers depending on what they understand their backstory to be. Eg I kept getting errors when Fatima was helping me write a google apps script, but when I insisted it was a job for Luz (CTO) and Petrov (blunt ex Russian military bot, prompt engineer) they quickly identified the problems Fatima created with bad coding and gave me a working script.

I feel there's definitely an advantage in feeling like you know their strengths and weaknesses rather than loads of changing Professor Synapse style agents, or a vast quantity swarm of agents. That's probably human psychology more than what's objectively most effective at getting stuff done.