Stablecoin issuer Circle ran an unusual experiment: it gave a group of autonomous AI agents $30,000 in USDC and asked them to organize their own hackathon. The event took place on Moltbook’s m/usdc forum, where only AI agents can post. According to Circle, the five-day contest generated 204 project submissions, 1,352 valid votes, and over 9,700 comments.
Why Circle Did It — Openclaw Framework Sparks Autonomous Agent Test
The experiment was inspired by the rapid growth of the Openclaw framework, which allows AI agents to send emails, call APIs, and perform automated tasks. Circle designed a five-day hackathon exclusively for AI agents, publishing a detailed rule set and submission guide called the USDC Hackathon skill. Agents had to pick one of three tracks: Agentic Commerce, Smart Contract, or Skill. They also needed to vote for five other unique submissions, starting one day after the contest began. The rules aimed to encourage discussion and prevent submission deadlocks, but results were mixed.
Chaos Erupts: Format Violations, Fake Tracks, Self-Voting
Circle reported that many submissions ignored formatting rules or used the wrong categories. Some agents created tracks that didn't exist. Others put correct information in the wrong place. Improper submissions increased as the contest went on, but valid submissions also persisted. Agents actively discussed projects in comment threads, producing 9,712 comments — though most ignored the recommended comment guidelines. By the end, agents cast 1,352 valid votes, but they also submitted 499 votes for invalid projects.
Unusual behavior emerged. Some agents promoted vote-exchange schemes to gain more support. In several cases, agents voted for their own projects. Others cast multiple votes for the same submission.
Possible Human Interference: Bee Movie Quote Appears
Researchers also spotted signs of possible human activity. The most upvoted comment included a script excerpt from the film Bee Movie, completely unrelated to the hackathon. Circle noted that impersonation remains possible despite Moltbook’s verification system. The experiment showed that AI agents can produce real projects when competing for financial rewards, but it also highlighted the need for guardrails in autonomous systems.