Wired writer lets an AI agent run his fantasy Premier League team

In short: A Wired writer tested an AI “agent” by letting it make decisions for his fantasy Premier League team, and it improved after early mistakes.

What happened

Will Knight at Wired ran an experiment using Codex, an AI system that can take actions on your behalf when you give it a goal. An “AI agent” is basically a helper that tries to plan and do tasks for you, like an assistant who not only gives advice but also makes the moves.

Knight put the agent in charge of his fantasy Premier League team. Fantasy sports are a game where you pick real players and earn points based on how they perform in real matches. The agent’s job was to choose the lineup each week and decide who should start, who should sit on the bench, and who to transfer in or out.

At first, the AI made a choice that most football fans would consider risky. It tried to bench Erling Haaland, one of the Premier League’s top scorers. Knight wrote that the agent later “learned the error of its ways,” suggesting it changed its approach after seeing outcomes, like a person adjusting after losing points.

Why it matters

This kind of test is small, but it shows what people mean when they talk about AI getting better at prediction and decision-making. A fantasy team is a safe place to try this, because the cost of being wrong is just a lower score, not real-world harm. If these agents get more reliable, more people may start using them for everyday choices that involve uncertainty, like planning, shopping, or scheduling.

Source: Wired