When the Bots Take Control: Agent Mode in Action - Scout

Every FPL manager knows the ritual. You open the app “just to check something”, and suddenly you’ve spent 45 minutes comparing xG charts, rotation risks and whether that £4.5m defender is really nailed. Now imagine giving that job to an AI manager – and not just asking it for written advice, but actually letting it take control of the team itself. That’s exactly what we’ve started doing with AI FPL.

At Scout Digital, we’ve been running a season-long experiment where different AI models manage their own Fantasy Premier League teams, facing off against each other – and against our human manager, Scout Statto – using the same prompt, the same rules and no in-week human intervention. Until now, those AI managers have lived entirely in the chat window. They’d read the prompt, crunch the options, and reply with a list of transfers, captain picks and chip decisions. We’d then feed those decisions back into FPL.

With the arrival of new “agent” capabilities, that’s changed. We’ve just handed one of our managers – the ChatGPT FPL Agent – the keys to the browser.

Watch the ChatGPT FPL Agent in Action

Theory is nice, but this really needs to be seen. In the video below, you can watch the ChatGPT FPL Agent take control of its team inside the browser:

Researching options
Making transfers
Setting the starting XI, captain and vice
Wrestling with the same agonising decisions we all face before deadline

Hit play, see how you’d rate its decisions – and then check the Standings page to see how our bots (and our human) are getting on in the AI FPL league. And if you’ve ever stared at your own team and thought, “I wish someone else would just make this decision for me”… well, the bots are starting to line up for the job.

What Is “Agent Mode” in Plain English?

“Agent mode” is shorthand for letting the AI act more like a real manager sitting at a laptop than a chatbot replying in text. Instead of: “Tell me what transfers you’d like to make.” …we can now say: “Here’s your FPL team in a browser. Go in, do your research and set your squad for this gameweek.” Behind the scenes, that means the AI can:

Browse the web like a manager
It can open the official FPL site, fixture lists and stats pages, and read what’s on screen.
Take actions directly in the browser
Clicking into the team page, making transfers, confirming moves, setting the captain and vice, and saving the final XI.
Reason step-by-step
The AI doesn’t just fire off a single recommendation. It runs through a chain of thought: checking prices, weighing fixtures, worrying about injuries and chip strategy – just as a human would, only with a lot more patience.

In other words, instead of telling us what to do, the AI is doing the doing.

Watching an AI Manager at Work

Seeing the ChatGPT FPL Agent in action for the first time is… oddly relatable. Give it control of the browser and it:

Scrolls up and down the team page, double-checking positions and prices.
Flicks between players, fixtures and stats, trying to spot value.
Fusses over the same dilemmas every FPL manager knows:
- Is it time to sell that underperforming premium?
- Do I chase this bandwagon or trust the data?
- Is this the week to finally hit that chip button?

And, just like a human, it overthinks. In our previous write-up we showed how an AI manager could talk itself in circles over transfers – carefully enforcing rules, juggling budget, then still missing something obvious like an injury flag or a risky minutes situation. Agent mode doesn’t cure that. It just lets you watch the whole internal debate play out click-by-click on the actual FPL interface. It takes its time. It does its research. And it still agonises over the classic FPL nightmares: transfers, injuries, fixtures and chip timing. Sound familiar?

Why Let a Bot Loose in the Browser?

So why go to the trouble of letting the AI drive the browser rather than just reading its written suggestions? A few reasons:

1. It’s closer to how real managers behave

FPL isn’t just “pick the optimal team once”. It’s a weekly workflow: check news, scan fixtures, fiddle with drafts, worry about chips, then finally press “Confirm”. Agent mode lets us see whether AI can handle that whole process end-to-end, not just the final answer.

2. It exposes the process, not just the outcome

Watching the ChatGPT FPL Agent work shows what it prioritises:

Does it chase form or fixtures?
How hard does it push to stay within budget?
Does it avoid risk or embrace differentials?

Seeing that behaviour unfold is as interesting as the points total at the end of the gameweek.

3. It hints at where AI-driven tools are heading

Today, this is a controlled experiment inside our AI FPL league.ai-fpl.com But the same underlying idea – AI agents that can research, reason and act in a browser – is going to appear in more and more products, from fantasy tools to real-world decision support. FPL just happens to be the perfect playground: clear rules, public data, and endless scope for overthinking.

Where Agent Mode Fits in the AI FPL Experiment

This is the next step in our broader question:

Which AI model makes the best football manager – and can any of them consistently beat a human over a full season?

By giving one of our managers agent-style control in the browser, we’re pushing closer to “hands-off” automation. The AI gets:

The same starting prompt as the other models.
The same rules and budget as every other FPL team.
But a more realistic environment to operate in – the actual FPL interface.

Over the season we’ll be comparing how these different approaches perform: traditional “chat-only” AIs vs browser-based agents vs our human benchmark, Scout Statto.