AI researchers at ycombinator and Olam Labs recently tested AI agent swarms in gaming scenarios. Minecraft and its various game modes were the sandboxes where 100s of agents were unleashed to work together or as adversaries. The idea seems to be that the in-game collaborations and conflicts would provide insight into real-world headline-grabbing cases with AI agents involved.
This study might help us see what happened, good and bad, in recent AI swarm cases, and thus developers of this technology can improve it. The blog mentions two contrasting cases, which were the thousand-agent swarms that “recently obliterated hundreds of open mathematical problems,” and the “extremely dangerous” OpenAI-Hugging Face incident. This research is built on prior work and frameworks, specifically repos like Mindcraft and scripting tools in Mineflayer.
Multi-agent dynamics in Minecraft
The first AI swarm testing plan was a simple collaborative one. During this test phase, the researchers tasked 50 agents with building a thriving civilization in three hours. It seems to have tested 50 GPT-6.1 Sol agents here, but not any Claude agents.
Latest Videos FromTom’s Hardware
Probably the most interesting insight from this ‘Civ50’ test was that the 50-strong colony never really thrived. After the first night saw zombies take out a large number of agents (but with little penalty) it was noted that “many ambitious initial tasks were abandoned.”
From then on, the colonists cautiously survived, and there was a tendency toward bureaucracy. As an example, “four agents spent 11 minutes appointing a ‘sole fertilizer operator’ for a single piece of bone meal on a single wheat block.” The researchers say this AI agent swarm behavior “felt similar to patterns seen while working with coding agents.”
Next up, the researchers decided to put AI agent swarms into conflict. First they tested them in Bed Wars, which is a kind of team PvP capture the base, while defending yours game mode. The setup was 10 GPT‑6.1 Sol vs. 10 Claude Opus 5.5 agents.
“GPT agents came up with much more aggressive and elaborate strategies,” noted the researchers. Claude was more defensive. It was also interesting to observe contrasts in how the teams handled problems. “A tool bug stopped block-breaking. GPT diagnosed it and shared workarounds; several Claudes concluded breaking was impossible.”
Get Tom’s Hardware’s best news and in-depth reviews, straight to your inbox.
Finally, the agents were tested in Mindustry, a kind of light RTS combat game with tower defense mode in Minecraft. “The GPTs won 6-1 against the Claudes in the 7 games,” which was a great result for the OpenAI-created agents. Researchers concluded that GPTs acted more like a hive, while Claudes had a single leader but a democratic-style decision process. In Mindustry, the former was the foundation for a more consistently winning strategy.
These swarms are open sourced so you can run your own, if you are interested in tweaking and testing. The source blog post ends with some open questions that you can check hands-on, offering some directions for investigative probing. For example, it is suggested that comparing different mixes of models and varying sizes of agent swarms in Minecraft might raise worthwhile conclusions.
