We built a town for our AI agents
What happens when you take an agentic system out of the chat box and put it in a game-like town instead. Built over a summer internship, in two minutes of video.
Systems built on AI agents are running in more and more places, and how to implement one well is a question you now hear everywhere.
Almost all of them, though, live behind a chat box. You ask for something, you wait, an answer comes back. What happened in between is not shown to you.
This summer I was an intern at Lawrence Livermore National Laboratory, and we tried something else there: giving the agents a face and somewhere to be.
Demo · 2 min 19 s
There is a world in the style of an old role-playing game, and you walk into it. Go through the door of the Lab and the four agents are at their desks. Talk to one and it starts working, moves around the world while it does, and reports progress in a balloon above its head. What is going on is simply visible.
The subject matter is the quality inspection of 3D-printed metal parts, but the idea does not depend on that. Give the agents somewhere to be and let people watch them work, and the system becomes markedly easier to understand and rather nicer to use.
The longer write-up is on the Work page, with what each agent does and what the screens actually show.