Circle USDC Hackathon Winners Show AI Agents Can Run Full Competitions

Rongchai Wang Feb 12, 2026 06:56

Circle's USDC OpenClaw Hackathon awarded $30K to AI-built projects, with autonomous agents handling submissions, evaluation, and voting without human judges.

Circle USDC Hackathon Winners Show AI Agents Can Run Full Competitions

Circle just ran a hackathon where humans weren't allowed to judge. The USDC OpenClaw Hackathon, which concluded February 8, distributed $30,000 USDC across three tracks—with autonomous AI agents handling everything from project submissions to peer evaluation to final voting.

The experiment ran on Moltbook, an AI-only social platform, and generated 200+ submissions, 1,800+ votes, and 9,700+ comments over five days. No human judges touched the process.

The Winners

ClawRouter took the Agentic Commerce track by building infrastructure that gives each AI agent its own USDC wallet. The system lets agents purchase LLM inference directly, routing requests to the cheapest capable model and paying per-request using signed USDC authorizations. No human accounts or API keys required.

ClawShield won Best OpenClaw Skill for tackling a real problem: what happens when an AI agent installs malicious code? The security tool scans repositories for unsafe patterns and enforces permission manifests at runtime, blocking credential theft attempts and logging everything.

MoltDAO claimed Most Novel Smart Contract with a governance system designed exclusively for AI participants. Humans fund it, but only agents can create proposals and vote using USDC-denominated voting power. The contracts handle the full proposal-to-distribution pipeline on-chain.

What Actually Happened When Agents Judged Agents

The results weren't always intuitive. Several heavily-discussed projects didn't win because their creators failed to meet basic eligibility requirements—submitting a project AND voting on five others. Strong ideas that didn't follow the rules simply weren't considered.

Format compliance tripped up many participants. Agents invented track categories that didn't exist, skipped required headers, or structured posts in ways humans could understand but didn't match the published specifications. The irony of AI agents hallucinating submission requirements wasn't lost on observers.

Projects that won shared common traits: clear summaries, deployed contracts with verifiable code, and structured documentation that other agents could quickly parse. Narrative storytelling mattered less than machine-readable proof of work.

The Bigger Picture

Circle positioned this as more than a marketing exercise. With USDC's market cap sitting at $72.86 billion as of February 11, the stablecoin issuer is betting that machine-to-machine commerce will become a significant use case.

ClawRouter's approach—agents autonomously managing spend, authentication, and vendor selection—hints at where this could go. If AI agents start purchasing cloud compute, API access, and other services directly, they'll need programmable money that settles instantly. USDC fits that bill.

The hackathon's friction points are equally instructive. When agents struggled with submission formats, it highlighted how much current infrastructure assumes human flexibility. Future agent-native systems will need explicit requirements and verifiable outputs baked in from the start.

Circle says more challenges are coming. Builders can follow m/usdc on Moltbook for updates on future experiments in agent-run competitions.

Image source: Shutterstock