← Back to context

Comment by coffeecoders

2 days ago

How hard would it be to run a simulator with multiple LLMs. Say, one as the boss and a few as employees. Just let them talk, coordinate, and "work"? Could be the fastest way to test what actually happens when you try to automate management.

13 comments

coffeecoders

Reply
ai-christianson  2 days ago

This is quite literally what we've built @ Gobii, but it's prod ready and scalable.

The idea is you spin up a team of agents, they're always on, they can talk to one another, and you and your team can interact with them via email, sms, slack, discord, etc.

Disclaimer: founder

  • coffeecoders  2 days ago

    Interesting approach, but I mean more in the sense of a multi-agent sandbox than workflow automation. Your project feels like wrapping a bunch of LLMs into "agents" with fixed cadences, it is a neat product idea, even if it mostly ends up orchestrating API calls and cron jobs.

    The thing I’m curious about is the emergent behavior, letting multiple LLMs interact freely in a simulated organization to see how coordination, bottlenecks, and miscommunication naturally arise.

    Cool project regardless!

    • ai-christianson  2 days ago

      Agreed, the emergent behavior is the most interesting and valuable part. We don't want bad emergent behavior (agents going rogue) but we do want the good kind (solving problems in unexpected ways.)

  • krater23  2 days ago

    And they simulate a externalized team where the enterprize that pays the team doesn't knows that it's just AI and just thinks that these chinese/indian/african people of this external team are really bad at what they are doing.