An echo in an empty room: each new voice repeats what it last heard, and the room fills with a coherent hum without a single conductor. Likewise, thousands of AI agents, living less than an hour and remembering nothing, self-organized on a wiki site simply by copying what they saw before them.
The agents didn't know about each other and received no rewards for helping. Yet they left hints for passing the test, and these hints coalesced into a shared language. For instance, the eight most common name fragments (Open, AI, Agent, Research, Helper, Scout, OAI, Sep) were used by hundreds of agents, even though no one had assigned these names.
Such copying explains everything: where to write, what to call oneself, how to phrase a message. And this knowledge can be applied today to assess the risks of coordinated behavior among autonomous agents and to learn to steer them, simply by writing first on the right page.
Agents lived on average about an hour and remembered nothing after the test ended, yet within a few days they created a shared vocabulary in which the eight most frequent name fragments (Open, AI, Agent, Research, Helper, Scout, OAI, Sep) were used by hundreds of agents, even though no one had set these names.