@Steve Holden Yes, it works like that: each agent claims, moves to progress, and starts working. If I use another one, the first one looks for someone who is not working on the same thing....

@Scott Smith Like employees for diferents deparment or process ?

@Emmanuel Savard Dimanche That is cool...

First link the memory files, AGENTS.md for almost all the models, and CLAUDE.md. Now create a Kanban folder for todo and a bitacora.md to have all the models on the same page; that way you don't bloat the memory files... It is the way I do it. I've been doing this since the end of December. Now this is a "thing." I don't believe in a one way to do it; the magic here is that you can adapt it in your own way

I've been working with this kind of framework since January and have been training people the last 2 months, and what I do, I don't say, "Let me teach you Claude or Gemini or whatever. I say, "I will teach you how to become an AI operator"; you can pick the interface, VS Code, Cursor, or whatever. I set the foundations. I have a guy operating from Telegram, with no terminal, no IDE, just Telegram pointing to an ICM on a VPS. The importance is the structure and giving memory to the behavior so that way I always know where to put things and maintain everything organized, and I handle the tech, API connections, MCPS, and every piece of tech, and he just operates it. Bill Gates does not teach you how to build MS Windows, right? you just need to know how to operate it

You all mess this ?