I Put a Bunch of AI Agents in Charge of a Fantasy Soccer Team. They’re Not Better Than Humans—Yet
The writer used AI agents to assemble and manage a Premier League fantasy football team, testing whether a group of bots could make better choices than a person. The system improved after being split into agents with different tasks, but it still made poor decisions and did not show that AI could outperform skilled human forecasters.
The agents read news and blogs, analysed recent results, ran prediction models and critiqued proposed choices; using more than 100 agents added little beyond API costs. Researcher Haifeng Xu says newer models can make stronger forecasts by processing large amounts of real-world information and coordinating complex systems. In separate prediction-market tests, models sometimes beat average people, but another trial found frontier models lost money, and skilled “superforecasters” still have an edge through deeper understanding of events.
In short
- AI agents improved the fantasy team’s suggestions.
- More than 100 agents added little beyond cost.
- Skilled human forecasters still have an edge.
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
AI agents could become useful forecasting tools because they can process large amounts of news and data, compare evidence across specialist roles and coordinate complex decisions. Early tests suggest newer models can sometimes outperform average forecasters, and the system improved when tasks were divided among agents, so continued development may narrow the gap with skilled humans.
The case against
The reported fantasy-football trial did not show that AI could outperform skilled human forecasters, and adding more than 100 agents brought little benefit beyond higher costs. Other tests found frontier models losing money, while superforecasters retain an advantage through deeper understanding of events; this suggests that more information and coordination alone do not ensure reliable judgement.
Explained simply
Fantasy Premier League is an online game in which players select real footballers for a virtual team and earn points based on their performances in Premier League matches. Managers make choices such as whom to buy, sell or name as captain, often using injury news, form and expected results.
The test used AI agents, computer programmes given different tasks, to gather information and make those choices. The wider question is whether AI can combine large amounts of information and work in groups well enough to make reliable forecasts.
That matters because forecasts influence decisions in areas beyond football, including prediction markets, where people trade on the likelihood of future events. Studies have produced mixed results: AI has sometimes beaten average participants, while skilled human forecasters have retained an advantage.
AI Football Research Science Sport Technology