Inside the Man vs. Machine Hackathon (www.wired.com)

At a recent San Francisco hackathon dubbed “Man vs. Machine,” 37 teams competed to see whether developers armed with AI coding tools outperform those relying solely on human skill. The event, cohosted by AI research nonprofit METR, sought to move beyond traditional productivity metrics by focusing on new project creation rather than contributions to existing codebases. A panel of judges evaluated entries on creativity, real-world utility, technical impressiveness, and execution, with a $12,500 prize plus OpenAI and Anthropic API credits awarded to the winning AI-assisted team. The competition revealed a surprising balance: finalists split evenly between AI-supported and human-only teams, with many projects blurring the line between man and machine. Notable demos included an AI-powered code review heat map that identified security risks, which ultimately won first place, and a human-built writing tool that tracked narrative consistency across novels, earning second. Participants reflected on the tension and trade-offs, with human coders admitting they wished for AI’s help for speed but proud of their resilience. Judges and attendees recognized that while raw productivity gains from AI remain nuanced—echoing prior METR research showing AI slowed experienced devs on existing code—combining human creativity with AI assistance currently offers the strongest advantage in fast-paced, novel software development challenges.