How operators can leverage agentic AI in sports betting
In June 1914, Lawrence Sperry flew over the Seine near Paris with both hands raised while his mechanic climbed onto the wing. A gyroscopic stabiliser kept the aircraft level. Sperry was demonstrating one of the first autopilots.
More than a century later, autopilot handles much of a commercial flight, but few passengers would board without a pilot. The crew sets the flight path, monitors the systems and takes over when needed.
Betting is heading towards its own version. A sports bettor could give an AI agent a budget and instructions to follow a cricket model, back an NFL tipster’s selections and monitor golfers throughout a tournament. The software could research selections and, where the platform permits, place bets within agreed limits, leaving the customer to get on with their day.
We expect this convenience to draw more customers into wagering and make specialist tools and betting expertise more useful and accessible to a wider audience. Prediction markets such as Polymarket already support automated trading, offering an early home for these products. Sportsbooks will face pressure to offer similar convenience within their own platforms.
A bot of your own
Automated betting is well established. Professional racing syndicates use computer-assisted wagering, combining probability models with software that places bets into tote pools at scale. Betfair also supports automated betting through its Exchange API.
AI agents could widen access to these capabilities. Customers who would never build a betting bot could describe a strategy in ordinary language and have software carry it out.
A conventional bot might take a tipster’s selections from a feed and submit bets within preset price and stake limits. An agent can work from a broader instruction, deciding which information and tools to use. For cricket, that might mean gathering team news and weather forecasts, consulting a specialist model and comparing its predictions with available odds before placing a bet.
AI agents can also follow familiar strategies, such as betting on price momentum or a reversal after a sharp move. They can make markets by quoting buying and selling prices to try to earn the spread. Exchanges have reason to welcome this activity. Market-making bots can improve liquidity, making it easier for other customers to trade, while additional fee-paying trades increase revenue.
Consumer products are already live. AI agent platform Olas launched Polystrat in February. Users fund an agent and choose or describe a strategy; the software then evaluates markets and trades on Polymarket. Forkast’s ATLAS lets customers request and approve trades through Telegram.
Forkast’s ATLAS proposes a trade and asks for the customer’s approval through Telegram. Source: Forkast.
Hands off the controls
The “trade while you sleep” pitch is already visible on X. Providers and promoters have reason to publicise profitable trades: screenshots can sell subscriptions. As with Sperry, the demonstration helps sell the technology. However, a profitable trade says little about lasting returns.
Polymarket’s public trading records let curious bettors investigate which accounts are making money, what they trade and whether the approach could be followed. As the tools get easier to use, bettors can copy accounts, adapt strategies and test ideas that once required a developer. Some may build tools for themselves and find others willing to pay.
Earning trust
An immediate benefit for bettors is consistent execution. A recreational bettor who likes a selection might treat a $25 bet at $2.20 much like one at $2.50. Yet if the selection has a 43% chance of winning, $2.20 implies an expected loss of 5.4 cents per dollar staked; $2.50 implies an expected profit of 7.5 cents, before fees.
Given a probability estimate and staking rules, software can calculate stakes as prices change and decline bets below a minimum price. It can also enforce agreed staking and spending limits, helping customers manage their bankroll. That consistency and the time saved may be worth paying for, but automation cannot make a poor strategy profitable.
The customer still sets the flight path. Letting an agent choose bets requires more confidence than asking it to follow instructions, and customers may want to approve individual trades before granting more discretion.
One contact in our network uses a Telegram bot to identify bets and calculate stakes under Australian racing’s minimum bet rules. These require bookmakers to accept qualifying bets up to a specified potential win. The bot calculates the stake quickly, helping him act before the price moves. He still approves each bet before it is placed.
Professional betting teams will scrutinise execution, error handling and the protection of their strategies. Any tool they adopt must preserve the advantage they have spent years building.
A crowded trade
If many AI agents pursue the same bets, their orders can move prices against later customers. Bill Benter described a similar problem in horse racing: independently developed models could favour the same horses, reducing payouts for everyone backing them.
Copying a successful account’s selections is also easier than reproducing its returns. Positions spread across several accounts can leave followers with an incomplete picture, while a worse entry price can erase the original edge. An agent still needs to judge whether a bet offers value at the price its customer can obtain.
The opportunity
As more AI agents compete for the same opportunities, we believe proprietary data, original analysis and specialist betting expertise will become more valuable. Agents could also expand the audience for tipsters by acting on selections that customers would otherwise miss, provided they can still obtain prices that preserve the advantage.
As customers expect more useful, personalised tools, wagering operators will need to decide what to build and what to buy.
FanDuel’s AceAI lets customers research markets and build bets through conversation. Customers still place their own bets. Source: FanDuel.
FanDuel developed AceAI internally, then shared code and infrastructure with fellow Flutter brand Sportsbet for its equivalent assistant. Large groups can spread that investment across their own brands.
Specialist suppliers can spread development and maintenance costs across multiple operators. Combined with proprietary data or betting expertise that is difficult to reproduce, this can give operators a compelling reason to buy.
For Waterhouse VC, the opportunity is to back suppliers that deliver better products at a lower cost than operators could achieve in-house.