AI Manages K-League Team as Football Manager

This experiment lasted a total of 31 hours and 20 minutes, during which three token resets were performed.

A Code Factory channel representative explained, "We made GPT-6 Astra play the devilish game Football Manager for a whopping 31 hours and 20 minutes, with three token resets."

It performs the duties of a real manager, such as recruiting players, devising match strategies, and managing training schedules.

The game reflects real player data and stats, requiring a very high volume of information and detail. A Code Factory channel representative stated, "It's a game where you recruit players, set up match strategies, manage training schedules, persuade your players who want to transfer to play one more season, and operate a club as a manager rather than playing as a player."

For the experiment, a simple prompt based on the latest trending strategy was used for Astra. This prompt encouraged Astra to continuously play the game until the end of a season.

It was also asked to log important events to record game progress. A Code Factory channel representative said, "The prompt was very simple. I tried the most popular strategy these days. I ran the prompt to make it keep playing until one season ended. I told it to leave a log if any important events happened."

Jeju United from the K-League was chosen as the team for Astra to manage. This team finished last in the real league in 2019.

Jeju United was chosen to answer the question of whether Astra could surpass the team's real-life poor performance. A Code Factory channel representative stated, "I chose Jeju United from the K-League as the team for Astra to manage. The reason I specifically picked Jeju United was that they finished last in the real league in 2019, and I was curious if Astra could do better than reality."

Astra immediately went to the tactics tab at the start of the game. This was to check player depth and analyze which positions were lacking in the current squad, similar to how an experienced person would start the game.

Through this analysis, Astra planned to reinforce 부족한 positions before the season began, attempting efficient recruitment within budget and salary constraints. A Code Factory channel representative explained, "Astra immediately went into the tactics tab. We first need to check the player depth to analyze which positions our club currently lacks. Only then can we fill the lacking positions before the season starts."

Astra self-configured formations and basic tactics, checked transfer budgets and salary allowances, and performed other basic game checks with great detail, progressing through the game.

It demonstrated the ability to meticulously review and operate the entire process without psychological fatigue, unlike a human. A Code Factory channel representative explained, "Astra set all the formations by itself, determined basic tactics, checked the transfer budget, checked salary flexibility, and confirmed squad depth. We could see it meticulously verifying all the basic checks and proceeding with the game."

Astra continued playing the game for much longer than the initially estimated 12 hours. The actual playtime exceeded 30 hours.

This showed that AI could maintain high performance continuously without fatigue. A Code Factory channel representative said, "I could see it playing for much longer than the 12 hours I had expected. I initially thought 12 hours would be enough, but seeing it play for 30 hours without stopping made me feel like a mother with a game-addicted son at 35."

After checking player depth, Astra determined the team needed a left-back. It immediately went to the transfer tab, filtering scouting reports for left-backs to identify potential recruits.

During this process, it attempted to sign three players, including Lee Myung-jae from Ulsan, and successfully negotiated a 700 million won transfer offer down to 370 million won. A Code Factory channel representative stated, "After checking player depth, it determined that a left-back was needed. So it immediately went to the transfer tab and diligently filtered scouting reports for left-backs."

Since Football Manager is a real-time game, it was expected that AI intervention would be difficult. However, Astra analyzed the real-time flow of the match and formulated custom strategies even during the game.

It demonstrated proactive response capabilities, such as checking players' stamina and independently changing tactics as needed. A Code Factory channel representative said, "Even during the match, it analyzed the flow of the game entirely and formulated customized strategies well."

In the first regular league match, Jeju United was losing 0-1 until the very end. At this point, Astra changed tactics to 'Attacking' and substituted an attacker.

This strategic change encouraged more players into the penalty box, ultimately leading to two late goals and a 2-1 comeback victory. A Code Factory channel representative stated, "In the first regular league match, since they were losing 1-0 until the very end, Astra changed tactics to 'Attacking' and substituted an attacker, autonomously encouraging more players into the penalty box."

However, the team managed by Astra currently ranks 9th in the league, achieving better results than in reality.

Although some matches were not completed due to token limits, it showed that the AI's managerial ability contributed to improving the actual team's performance. A Code Factory channel representative explained, "In reality, Jeju United finished 12th in 2019, which was last place. Currently, they are in 9th place in the league."

Astra demonstrated real-time responsiveness that was impossible with previous models. Based on its almost infinite memory, it remembered and strategically utilized each player's capabilities shown in previous matches.

The strengths of such LLM-based computers suggest potential applications in various fields, such as financial data analysis using tools like TradingView. A Code Factory channel representative emphasized, "Because its memory is almost infinite, when forming strategies or substituting players, it truly remembers each player's capabilities shown in previous matches, just like a real manager, and utilizes them strategically well."