Agentic Betting: When AI Chooses the Bookmaker, Who Owns the Customer?

AI agents are starting to execute real financial transactions, while sportsbooks restrict automated access. Could betting’s next big competitive battle be over which operators these agents are allowed to use?

We have already outsourced a surprising amount of our thinking to AI. Research, comparisons, planning and recommendations that once took considerable time can now begin with a conversation. Increasingly, we are allowing these systems to complete tasks as well.

Betting could be next. But the interesting question is not whether AI can predict a football result more accurately than a professional bettor. It is what happens when someone asks an AI agent to find the best available odds, and the agent decides which bookmaker to use.

For an industry that spends billions on advertising, sponsorships and customer acquisition, this could present an uncomfortable challenge. Particularly if the customer no longer needs to visit the sportsbook to make the decision.

And there are already signs of what that future might look like.

AI Agents Are Beginning to Trade, Not Just Research

On 24 September 2026, US investment platform Public announced a partnership with Kalshi, allowing customers to trade prediction-market event contracts through its platform and AI agents.

An update published on 4 October confirmed that Public’s agents could discover and trade event contracts. This is a meaningful development, although it is not quite the fully autonomous betting agent some might imagine.

Public’s system allows customers to create trading strategies using natural language. They review and approve the resulting instructions before execution. Once activated, the agent follows those rules rather than independently inventing new strategies.

It is closer to AI-assisted strategy creation combined with automated execution. Public’s initial announcement also focused on events with financial and economic significance, not unrestricted sports betting. Nevertheless, the distinction between researching a market and making a transaction is narrowing.

Elsewhere, the introduction of Grok Bot in August, followed by Meta’s Muse and OpenAI’s Dots in September, has demonstrated how general-purpose AI agents can work across websites and applications.

These systems are not authorised betting services. But they belong to a broader shift towards software acting on behalf of individuals rather than simply providing information.

The commercial implications extend well beyond gambling.

Will Sportsbooks Become Commodities?

Traditional sportsbooks have spent years encouraging customers to choose their brands. A familiar name, a generous welcome offer, a polished mobile app or a sponsorship of someone’s favourite football club can all influence where a customer registers and places bets.

An AI agent could approach the same decision differently. Imagine asking an assistant to monitor Premier League markets across bookmakers where you already hold accounts. It compares the available prices, identifies the most suitable option, and explains the difference.

Instead of opening several apps, you receive a single recommendation. That sounds convenient. But it also changes where the decision happens.

The sportsbook still provides the market, accepts the wager, holds the customer’s funds, and settles the result. What it may no longer control is the customer’s attention.

There is an interesting comparison with banks, internet providers, or eSIM cards. Most consumers want a reliable payment or connection at a competitive price. The brand matters, but much of the underlying service can appear interchangeable.

Could sportsbooks eventually face a similar problem?

Not entirely. Bookmakers differ in liquidity, market selection, reliability, withdrawal processes, and customer service. Licensing and trust also matter, particularly when real money is involved. Still, if independent AI agents increasingly select operators on behalf of customers, expensive branding may carry less weight than it does today.

A bookmaker could offer an excellent product and invest heavily in customer acquisition, only for an agent to recommend a competitor with a slightly better price. That is a potentially significant change to the economics of online betting.

Better Prices Are Valuable, Even Without Better Predictions

There is already evidence that choosing between sportsbooks can make a measurable difference. In July 2026, US odds-comparison company Iron Marker examined more than 13,000 event-markets across major sportsbooks.

Its Cost of Loyalty Index estimated that exclusively using DraftKings rather than selecting the best available regulated price involved an average expected-value disadvantage of $33.58 per $1,000 staked. The comparable figures were $23.68 for FanDuel and $23.43 for Caesars.

These were modelled differences based on quoted prices, not realised losses or guaranteed savings. The research came from a commercial odds-comparison provider and did not account for promotions, betting limits or every practical difficulty of accessing the quoted prices.

Nevertheless, it illustrates the potential value of automated comparison.

An agent does not need superior predictive abilities to recognise that decimal odds of 2.04 are more favourable than 1.95 for precisely the same outcome.

But better odds do not automatically make a bet profitable.

In fact, a 2026 study involving four leading AI models found that none outperformed betting-market probabilities on its forecasting measure across 104 World Cup matches.

That single-tournament preprint does not settle the wider question of AI’s predictive ability. It does, however, challenge the assumption that sophisticated language models automatically possess a betting advantage.

Agents could help customers find better prices while simultaneously making genuinely attractive prices harder to obtain. If thousands of systems identify the same discrepancy, operators may adjust their markets more quickly.

AI might improve price discovery without making gamblers more profitable.

Sportsbooks Can Block AI Agents, but There May Be a Commercial Cost

There is an obvious obstacle to agents freely selecting bookmakers.

Many sportsbooks do not permit automated access or wagering.

DraftKings’ Pennsylvania terms explicitly prohibit automated interaction with its platform, including artificial intelligence used to gather information or place wagers.

Bet365’s UK terms also restrict automated data extraction and third-party assistance software, including AI tools designed to provide an unfair advantage.

These policies have legitimate purposes. Operators need to protect customer accounts, prevent unauthorised access, manage suspicious activity and maintain market integrity.

We should not assume they exist simply to protect bookmaker margins.

Nevertheless, restrictions can have commercial consequences.

Imagine an AI assistant that can execute transactions through several approved betting platforms, while another sportsbook prohibits it from accessing customer accounts.

The agent might still compare that sportsbook’s prices using legitimately available information and recommend that the customer place a bet manually.

But it cannot necessarily complete the transaction itself.

If customers become accustomed to agents handling online tasks, operators offering approved integrations could gain an advantage.

This is not entirely new territory. Betfair has long operated an Exchange API, enabling authorised software to access markets and place bets. Its software vendor programme provides a formal route for approved third-party applications.

What AI potentially changes is accessibility. Someone might eventually describe a permitted strategy in ordinary language rather than developing specialist trading software.

This leaves traditional bookmakers with several options.

They could continue restricting independent agents, develop their own assistants or establish controlled partnerships with selected AI platforms.

The commercial question is whether protecting direct customer relationships today could make it harder to reach customers tomorrow.

Of course, there is a competing possibility. Approved integrations might reduce acquisition costs and bring customers directly to relevant markets without requiring expensive advertising campaigns.

The operators that collaborate with agents might not simply be surrendering control. They could be developing a new distribution channel.

Regulation Is a Bigger Barrier Than Technology

Whether an AI agent can technically place a wager is only part of the question. It must also be permitted under the relevant gambling rules.

In Pennsylvania, for example, restrictions on automated sports wagering are not merely a commercial decision by DraftKings.

Under 58 Pa. Code § 1408a.4, sports wagering operators must include a prohibition on automated computerized software used for wagering in their customer terms.

The regulation also addresses account access, identity verification and the requirement that bettors be physically located in Pennsylvania.

An operator cannot simply change its terms to authorise independent agents while disregarding those obligations.

New Jersey’s internet gaming regulations similarly require restrictions on automated computerized software used for play. However, those provisions concern internet gaming and should not be treated as an identical rule governing every form of sports betting.

The British gambling framework is different again.

The UK Gambling Commission’s Remote Technical Standard 16 addresses third-party software in peer-to-peer gambling. Relevant operators must explain whether such software is permitted, which types are allowed and what measures they take against prohibited systems.

It does not grant general permission for AI agents to place fixed-odds sportsbook bets. Even where an operator is willing to collaborate, practical compliance issues remain.

An agent operating through a cloud computer raises questions about how the customer’s location is verified, who authorised the transaction and how account protections are enforced.

A remote computer’s location does not automatically determine the legal location of a wager. But the underlying customer and transaction still need to meet the applicable requirements.

Approved integrations may eventually resolve some of these issues through verified accounts, explicit permissions and controlled execution.

That is considerably different from allowing an independent agent to log in and start betting.

Prediction Markets Are Not a Regulatory Shortcut

Kalshi offers another route into agentic trading, but its regulatory position is not straightforward.

Kalshi operates a federally regulated designated contract market under the Commodity Futures Trading Commission. However, its sports-related event contracts have been the subject of ongoing disputes about whether federal commodities law prevents states from applying their gambling regulations.

In April 2026, the US Third Circuit upheld preliminary protection for Kalshi against enforcement by New Jersey regulators.

The Ninth Circuit reached a different conclusion in August regarding Nevada. On 25 September, the Sixth Circuit also ruled against Kalshi’s federal preemption arguments involving Ohio and Tennessee.

These decisions concern ongoing litigation, and the national legal position remains unsettled.

For AI platforms and betting operators, the lesson is important: an approved technical integration does not automatically make every transaction lawful in every jurisdiction.

The relevant market, regulatory regime, customer location and scope of authorisation all matter.

Could AI Platforms Become the Next Gambling Gatekeepers?

Suppose customers increasingly use personal AI agents to compare products, manage subscriptions and complete transactions.

Some betting operators collaborate with those agents. Others remain closed.

At first glance, participating operators might appear to benefit.

But who decides which bookmakers the agent recommends?

A system that can transact with only a limited group of partners might recommend the best price among those partners rather than the best available price overall.

And who pays for the recommendation?

This is where the traditional affiliate industry becomes relevant.

Affiliates already help operators acquire customers through comparisons, reviews and promotional partnerships. Their commercial relationships can influence which brands they feature.

An AI assistant could eventually perform much of the same function, except the recommendation would arrive through a personal conversation rather than a comparison website.

That creates potential conflicts of interest.

An agent promising to find the most suitable sportsbook should not quietly prioritise an operator because it pays a higher referral commission.

For affiliates, this could change the value of search traffic and comparison content. For operators, access to AI-controlled distribution might become an important acquisition expense.

The customer relationship could move from the sportsbook to another intermediary.

And the intermediary might be considerably larger than any gambling affiliate.

What Happens When the Agent Gets It Wrong?

There is one reason betting deserves more caution than ordinary agent-driven shopping.

Finding a cheaper flight can save money. Finding better odds does not remove the financial risk of gambling.

The objectives assigned to an agent are therefore critical.

An assistant instructed to identify every supposedly favourable betting opportunity might generate far more recommendations than its user originally anticipated. Its estimates could be wrong, and losses could accumulate quickly.

Automation also removes some of the friction between considering a bet and spending money.

That could create problems for safer-gambling protections, particularly if an agent interacts with several operators.

Individual sportsbooks may see only their own account activity, while the AI system could potentially have a broader view of the customer’s exposure.

Used responsibly, that might become an advantage. An agent could help users track their spending, respect financial limits and recognise when further betting is inappropriate.

But it also introduces questions about privacy, consent and liability.

Who is responsible if an agent misunderstands an instruction, places an unintended transaction or continues operating after a customer has attempted to stop it?

And how can operators ensure that automation never bypasses self-exclusion or other mandatory safeguards?

The technology will need answers to these questions, not simply faster execution.

The Next Competitive Battle May Be Over Access

I suspect agentic betting will become one of the industry’s next important competitive battlegrounds.

Not because AI will suddenly make everyone a profitable bettor, but because it could change who decides where a customer places a bet.

There may eventually be two different experiences operating alongside each other.

One remains centred on the sportsbook brand, its application, promotions and entertainment. The other is more transactional, with AI agents comparing markets and accessing approved providers on the customer’s behalf.

The two could coexist. Many recreational bettors may continue to prefer familiar brands and making their own decisions, even when automated alternatives are available.

Traditional sportsbooks also have reasons to resist unrestricted third-party access, and regulators may impose limits that prevent fully autonomous wagering from becoming widespread.

But the broader direction of digital commerce is difficult to ignore.

For most of the internet’s history, companies competed to persuade us to visit their websites and applications. Increasingly, we are asking AI to handle parts of those interactions instead.

If that continues, established UK bookmakers and new betting sites may find themselves competing not only for customers, but also for access to the agents representing them.

And the companies controlling those agents could become just as important to the future of betting as the bookmakers offering the odds.