US hosts coding tournament between humans and AI: guess who won

Organizers randomly divided 37 teams into «human-only» and «AI-assisted» groups. The competition offered $12,500 in prizes and API credits provided by OpenAI and Anthropic.

The hackathon was organized by researchers from METR, who previously demonstrated that AI tools can slow down experienced developers by 19%. This time, the conditions were different: participants had eight hours to develop new projects of their choosing. Six teams made it to the finals — three human-only teams and three AI-assisted teams.

Projects included an AI-powered feedback tool for pianists, a book-tracking application, and a neighborhood connection service. Judges evaluated submissions based on four criteria: creativity, practical utility, technical complexity, and execution quality.

The grand prize went to an AI-assisted project — a code review tool featuring heat maps that highlighted critical errors in red and minor changes in green. Second place went to a «human-only» project — a service for writers that automatically tracks characters, their traits, and relationships to help avoid plot inconsistencies.

Despite the AI-assisted team’s victory, the results were nearly equal between human-only and AI-assisted teams. Many participants admitted they wanted to be on the «other side» to play the role of David against Goliath. As the winners acknowledged, «In this format, people always want to believe in themselves, but in reality, the human-plus-machine combination almost always wins.»