Some up-and-comers are pushing to make agents that can actually place wagers instead of just supplying tips, but the field is off to a rough start. Tom Fleetham formerly worked as the head of business development for a blockchain platform called Zilliqa that experimented with an AI gambling agent called Ava, focused on picking horse race winners. “She had good analysis, good results,” he says. “Where it got hard was actually trying to place the bets.”
The company couldn’t get Ava to reliably place bets using crypto wallets in a timely fashion, Fleethem claims. “It took forever,” Fleetham says. “We gave up.”
YouTube is awash in tutorials about how to create and manage gambling agents that can place bets on behalf of humans. But again, these services do not appear to be minting new millionaires—or even thousandaires. Siraj Raval, a YouTuber who publishes videos about how to make money using AI, has promoted a side project called WagerGPT on his channel. He claims the tool is capable of placing bets, and he charges people $199 a month for access. “As of eight months ago, I implemented a feature to let WagerGPT place bets. Now it's doing that for the users full-time,” Raval claims. According to Raval, WagerGPT scans over 40 sports books and “spots all sorts of variables that humans can’t.” Ravel invited WIRED to join a Telegram group he claimed was full of people using WagerGPT, but the channel was largely inactive, and most of the recent messages were questions about what’s going on with the service. “It’s completely dead,” alleges Pete Sanchez, one of the participants. “Waste of money.”
As AI agents cannot control traditional bank accounts, most of the fully automated betting products focus on sports gambling websites and prediction markets that take cryptocurrency, as many agents can and do operate crypto wallets. One of the largest mainstream projects to allow AI agents to make all sorts of transactions on behalf of humans is Coinbase’s AgentKit. It imagines a world in which agents can execute a number of financial transactions, from purchasing airline tickets to trading crypto and, yes, placing bets on sports. Lincoln Murr, an AI product manager at Coinbase, says that a lot of AgentKit’s early use cases “were speculative in nature” and noted he hadn’t seen anything particularly successful. “How profitable these agents truly are, I don’t know,” he says. The one he’s been paying the most attention to is called Sire. The project describes itself as “an agentic sports-betting hedge fund” and operates as a DAO. (In crypto terms, DAO means decentralized autonomous organization—a member-owned community that uses blockchain-based contracts). “Those types of things are the direction we’re heading,” Murr says.
Sire (which was previously called DraiftKing, but had to rebrand for legal reasons) is in the process of a relaunch. Max Sebti, the CEO of its parent company, Score, says the company uses a combination of public and private data as well as “computer vision tech that is watching the games” to get the most up-to-date information possible in order to determine winning bets. Score’s business model is a complicated mashup of crypto and gambling. Eventually, the company plans to allow anyone to transfer USD into a wallet, which will then be converted into a stablecoin. After that, the plan is that Sire’s AI agents will pool the money with payments from other customers and place bets on decentralized sports books and prediction markets that accept crypto, including Polymarket. Then, Sebti says, the gents will redistribute the winnings back to the community. Septi describes the initiative as “a very steady, hedge-fund like product.” Anyone can add money to a wallet, but to take winnings out, there’s a “performance fee” that must be paid to Sire—one that can be reduced if the customer purchases the company’s crypto token. The service comes out of beta this month.