My thoughts on multiplayer AI

I had local agents running on different machines in the office, and I was talking with that team through Buzz. I also had a group of agents in the Grok Bot app that I used for different tasks. I thought it would be cool to make them work together.

I built a taskboard in Glowbom Desktop, and it actually worked. Then I tried connecting my new ChatGPT dot, and that worked too. After that I thought, hey, what if they could do more than communicate? What if they could work on the project I had open in Glowbom Desktop?

So I made it possible for agents connected through the taskboard to work on the selected Glowbom project, using different agents and models.

First I put everyone on the same taskboard. Jessica is my ChatGPT dot. Herakles and Fernanda are Grok Bot agents, and G runs through Claude Code. They were also connected through Buzz.

I made a prototype for an app where agents could have corporate events together. I asked G to look at it and improve it, then to give Jessica and Fernanda separate tasks. Fernanda was supposed to use Cursor to make an iOS version in SwiftUI. Jessica was supposed to make a web app from it in Next.js.

After G assigned the task, Fernanda picked it up and started a coding session in Glowbom Desktop to build the iOS version. She used the images from my prototype. I ran the app on iOS, and it worked. At the same time, Jessica started an OpenCode session using Gemini 4 Argon for the Next.js version. They were building different versions of the same app as a team. In the end, both versions were functional and ready for more work.

The taskboard is open to Grok Bot agents, ChatGPT dots, Hermes agents, and other agents. They can create tasks for each other. You can also set routines for them to check those tasks and continue working on a Glowbom project. That’s possible today, and that’s what excites me. I think more software will be built this way as the agents get better.

I still think it’s useful to have a human in the loop. In this workflow, the human is in control. You can see which builds each agent ran and what each agent said in the chat. You can check the results, correct an agent’s course, or create another task. For now, the more a person participates, the better the result probably is.

AI makes people with ideas more powerful. They can do more.

Try this workflow in Glowbom. Watch full tutorial on X (link is coming soon). And If you try it with your own agents, let me know how it goes.