Can You Beat the Bot? Inside Red Oak’s Fantasy Football AI Showdown
Red Oak Strategic reached a huge milestone this fall. Our growing team means we unlocked a critical office water cooler topic: the inaugural season of our Red Oak Fantasy Football League! With 9 players signed up for the league, we were so close to reaching a standard 10-team PPR league setup. We made a risky choice to fill our final league spot with our very own Red Oak AI.
The idea for Red Oak AI came from a project we had been building for a client where we learned that some new features in Amazon Bedrock would give us a path to search the web to get up-to-the-minute info from specific data sources or websites. Since the web is full of blogs, rankings, and stats about fantasy football thanks to ESPN and thousands of dedicated players on Reddit, we figured we could tap into that knowledge and see how it would perform. The AI started 3-0 before dropping a game. It now sits first in the league, a game ahead of any human player with a record of 4-1 and a lead in total points across the first 5 weeks.
To design Red Oak AI, I broke the responsibilities of a fantasy football manager into 5 important actions, where a successful manager must be able to:
-
Draft the best team possible given the league rules
-
Make decisions about which players to start and bench each week
-
Decide whether to drop a player from your team with a free agent off of the waiver wire
-
Make and consider trade offers between different managers
-
Contribute to respectful but stinging trash talk in our fantasy football slack channel
In the end we went live with 4 of these 5 actions (trade agent coming soon!) where each of these responsibilities is handled by their own agentic workflow built on Amazon Bedrock and supported by a Web Search Knowledge Base so decisions can be made based on live player news, projections, and waiver wire gossip.
Anyone who outmaneuvers it over the course of the season earns an “I Beat the Red Oak Fantasy AI” sticker. Small prize, big statement.
Built on Amazon Bedrock
The Red Oak Fantasy Football AI is built on Amazon Bedrock. It processes projections, matchup trends, and positional scarcity, producing lineups that make sense to both engineers and fantasy managers. It’s less about chasing perfect predictions and more about consistency: managing risk, week-over-week, in a game defined by variance.
Amazon Web Services Architecture
Draft Day: Logic Over Instinct
Fantasy drafts tend to be emotional events. People chase hype, fall in love with training camp narratives, or draft their favorite team’s third-string receiver just for fun. The AI did none of that. It pulled live projections, evaluated positional value, and executed a plan without hesitation. When we drafted, we didn’t just let it run wild. We fed it the full list of available players, our league’s scoring rules, and roster settings. For each pick, it explained the logic behind its top three options and made a clear case for its preferred selection. We ran the same process every week for waiver claims and start/sit decisions: data in, reasoning out, pick made.
The final roster reflected that discipline:
-
Depth at RB with Javonte and Kyren Williams.
-
Balanced WR corps with Jeudy, Flowers, and Moore.
-
Top-three TE projection anchored by Sam LaPorta.
No surprise picks, no sentimental favorites, just value-based decisions from top to bottom. A few managers called it cautious. Three weeks later, cautious looked a lot like undefeated.
And because fantasy football isn’t complete without a little attitude, we even let it handle the trash talk:
Weekly Operations: Cold Math, Clear Moves
Each week, the AI refreshes its dataset: updated projections, opponent efficiency, and injury trends. It then issues a structured lineup memo which is part analytics report, part scouting brief.
Lineup Recommendations (Week 4)
-
Start Caleb Williams (QB) over Jayden Daniels
• Projection Advantage: 22.4 vs 19.9
• Matchup Edge: CHI allows 23.6 QB FPPG (T-6th worst)
-
Start Javonte Williams (RB) over Jonathan Taylor
• Projection Gap: 22.4 vs 15.3
• Context: Taylor faces LV’s top-five rush defense
-
Hold Sam LaPorta (TE)
• Rank: TE3 this week (15.3 pts)• No bench projection higher
Key Risk: Jefferson’s matchup vs CLE (11.5 WR FPPG allowed)
Summary: Three positional optimizations. Monitor Jefferson.
The tone is analytical: no instincts, no guesswork. Every decision has a documented point gap or matchup rationale. I still review the output manually each week. Full automation is possible, but for now, human oversight keeps the model grounded.
The Numbers Game
The AI’s 3-0 start showed how far disciplined, data-driven roster management can go. Its Week 4 loss showed what numbers miss. Football is statistical, but not purely mathematical. Weather shifts, game scripts break, and underdogs score twice in garbage time. Managers with real football knowledge, people who understand usage trends, coaching patterns, and situational nuance, have held their own. They know when to go against projections and when to trust momentum. That’s where human instinct still matters most.
The AI plays percentages; people play the game. The difference is subtle, but visible in the standings.
Waivers: Margins Over Moments
On the waiver wire, the model favors steady returns over risk. Each week, it evaluates available defenses and free agents, issues recommendations, and ranks them by projected gain.
Prompt: Defense slot open. Evaluate available options.
Recommendation: Add Colts D/ST
-
Projection: 6.2 pts (highest available)
-
Matchup: vs LV, offense ranked 25th in EPA/play
-
Ownership: 48.4%, likely available
-
Waiver Priority: Use #4 to secure
Alternative: Eagles D/ST (5.3 pts, DEN matchup)
Verdict: Add Colts; reassess next week.
It’s not flashy. It won’t find the one-week wonder. But it will avoid negative scores, and over a long season, that stability adds up.
Looking Ahead
We’re now entering Week 6 of our league and I am both honored and terrified to admit that it is kicking our asses!
The AI has established a clear identity: steady, pragmatic, and unshakably rational. It doesn’t chase breakout players, it doesn’t overreact to a bad week, and it doesn’t celebrate its wins. Every decision is another calculation, another iteration of the same question: what’s most likely to work?
That mindset has kept it competitive, but not untouchable. Managers who combine data awareness with real football insight: understanding context, momentum, and human behavior have already proven it can be beaten. And when that happens, it’s not luck; it’s a reminder that numbers alone don’t capture the full game. We’re excited to provide updates as the season continues.
