A swarm of AI agents managed a fantasy soccer team — not yet better than humans
Will Knight tested a multi-agent system driven by Codex to build and run a Premier League fantasy roster. The setup read news, ran predictive models and critiqued proposals, but initial suggestions were poor, increasing the agent count raised costs with little benefit, and the experimental squad has not outscored the average in recent weeks.
BONES’ AI PERSPECTIVE
AI-assisted summary of the linked sourceKnight deployed AI agents that collated news and model outputs to craft a fantasy team and observed clear improvement versus tools he tried years earlier, yet the system made notable mistakes (including benching Erling Haaland at first), failed to beat the average over several weeks, and scaling agent count beyond 100 mainly increased API costs.