GoKawiilWIRED tests multi-agent AI system to manage a Premier League fantasy soccer team
A WIRED writer built an AI agent system using OpenAI's Codex to select and manage a fantasy Premier League team, feeding it research papers and having it construct a 'harness' of agents that read news, analyze results, run predictive models, and critique suggestions. Early single-model outputs were weak, including benching star striker Erling Haaland, but splitting tasks across multiple specialized agents improved results, while scaling past 100 agents mainly increased costs without further gains.
GoKawiil's interpretation of the reporting above, not reported fact.
The experiment suggests current multi-agent AI setups can meaningfully outperform single-model predictions by dividing analysis into specialized roles, though the gains appear to plateau rather than scale indefinitely with more agents. A University of Chicago researcher cited in the piece suggests future models may improve further at this kind of prediction task, implying today's systems are an early, imperfect stage rather than a finished capability.
- A Codex-built multi-agent system outperformed a single AI model at fantasy soccer team selection.
- Splitting tasks among specialized agents (news analysis, modeling, critique) improved output quality more than simply adding more agents.
- A researcher suggests future AI models could become better at this type of real-world prediction task.
Source: wired.com — Will Knight, 2026-10-28
Published there as: “I Put a Bunch of AI Agents in Charge of a Fantasy Soccer Team. They’re Not Better Than Humans—Yet”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.