Are You Willing to Become the Backend?

In our last piece, we argued that every financial firm needs to decide on purpose whether to serve, block or limit the AI agents that now act for its customers. What large retailers and banks have done over the past few weeks makes clear that this decision belongs in the strategy conversation, because the real question underneath it is whether your firm is willing to become the backend for someone else's front door, and what value you still provide if it does.

What the last few weeks showed

Large companies have started to choose, and their choices fall into three groups. Shopify and Walmart have chosen to work with agents on their own terms. In late September, Shopify opened checkout on its eligible merchants' stores to browser-based agents acting with the buyer's authorization. On October 21, Walmart launched an Agent Gateway that lets approved agents search its inventory and buy for customers. Walmart set its own terms: agents can't see payment details, can only use the data a task needs, and get a separate profile in the customer's account, where the customer can approve each purchase or set a spending limit, and agents that break the rules are blocked.

Yelp has chosen to charge for access, and says it does not permit non-human traffic unless the agent has paid for access through its data licensing program.

Amazon and several large banks have gone the other way and are keeping outside agents out. In September, Amazon blocked Meta's Muse from shopping on its site. In October, large US banks pushed back after reports that unvetted personal agents were operating inside customer accounts, in some cases taking actions the account holder had not asked for. JPMorgan's CEO said the bank won't let "just any outside agents" operate inside the company, and Schwab's CEO said the firm has been blocking unauthorized agents.

The banks' concerns are real, because an agent that holds a customer's login can be compromised like any other credential, and an agent with full access to an account is a much bigger risk than one that can only read balances. Some have suggested read-only access as a middle ground, which shows that the choice is rarely all or nothing.

Why Amazon and Walmart chose differently

The two retailers made opposite choices, and the reasons have more to do with strategy than with security. Amazon has long described itself as "Earth's most customer-centric company, where customers can find and discover anything they might want to buy online." Its value to shoppers is being the one place to search, compare and buy. If a customer's own agent does the searching and comparing across every store, the agent takes over that role, and sellers could start dealing with the agent platforms directly. Amazon's advertising business, which passed $68 billion in 2025, depends on shoppers seeing its search results and sponsored listings. When an agent compares prices and goes straight to checkout, the shopper never sees those listings, and Amazon loses both the ad revenue and the chance to suggest something else.

Amazon is not against agents as such, since it offers its own: Alexa for Shopping, which replaced Rufus in May, and Buy for Me, which can purchase products from other retailers' websites on a customer's behalf. Its own agents keep the customer's decisions inside Amazon, and that is the position the block protects.

Walmart has made the opposite bet, because its strengths are its prices, its inventory and its store network, and those still matter when an agent does the shopping. If agents become a large share of how people buy, Walmart would rather be the backend they prefer, on terms it sets, than a site they can't reach. Each company is acting on what it believes customers actually pay it for, which is why neither choice would suit every firm.

The question for financial firms

Financial services has the same split, and most firms have not yet asked the question directly. If more of your customers deal with you through their own agent, what do you still offer that the agent can't replace?

For insurers, the answer is fairly reassuring, because much of their value sits in underwriting, claims handling and financial strength, and an agent can't supply any of those. Many insurers also already sell through brokers and comparison sites and have long weighed what they give up in brand and customer data against what they gain in reach, and that experience should help them judge an agent, which is in effect a new kind of intermediary. Becoming the backend need not be a loss, and for firms whose real strengths are underwriting skill, claims capability or balance-sheet strength, it can be a deliberate and advantageous position, as long as they set the terms on which agents interact with them.

For insurers, the underwriting and claims core is relatively protected. Across financial services, the more exposed part is the revenue that depends on the customer's attention or inertia: the money that comes in because the customer is still looking at your site or app, or because shopping around feels like too much work.

• Idle balances. Research from Harvard Business School found that about 94% of depositors leave their money at the same bank each year even when moving it would earn more, and estimated that about 60% of bank value comes from these inattentive depositors. An agent that checks rates and moves idle cash to the best-paying account every week removes the inertia that value rests on, and the deposit analytics firm Curinos already expects agents to shop for rates on consumers' behalf.
• Renewal inertia. Many customers renew a policy or keep an account because shopping around is tedious, and an agent that re-shops at every renewal takes that friction away.
• Cross-sell and upsell. Offers shown inside your app or website only work if the customer is there to see them, and an agent that logs in, completes the task and leaves is unlikely to pass them on.
• Advice and relationship. Firms whose value lies in guidance, such as wealth managers and advisors, need to show that value in a way a customer's agent can recognize and pass on. If the firm publishes a clear summary of its planning approach, its fees and what sets it apart, an agent can pass that on, and the relationship is more likely to survive than if the agent only sees product rates.

Before deciding how to treat agents, it helps to know which of these exposures you are actually trying to protect, and whether blocking would protect it for long when customers can move to firms that welcome their agents.

Four postures, chosen task by task

The market is settling into four postures, and most firms will end up mixing them, using a different one for different tasks.

A regional bank, for example, might partner with approved agents on rate quotes and balance inquiries, require step-up authentication for any money movement, and block any agent that cannot show a verifiable authorization from the customer. An insurer might welcome agents on quotes and claim status checks while keeping coverage changes with a person. Choices like these can be agreed in a single leadership discussion and then turned into policy and technical controls, as long as each one follows from a clear view of where the firm's value lies.

Questions for your leadership team

1. What do customers come to us for, and how much of it survives if they never see our website or app?
2. Which of our revenue depends on the customer's attention or inertia, and how much of it would an agent remove?
3. Which of our strengths does an agent need and can't replace? Those are the areas where being the backend is a good position.
4. For each major task, do we partner, charge, block or build our own, and who signs off on that choice?
5. If a competitor opens its doors to agents on terms that customers prefer and we do not, how quickly do we lose the customers who already use agents for everything else?

Where this leaves the front door

In most firms today, no single executive owns the question of how agents should be treated, so the choice gets made by default across digital, risk, legal and product teams. Our last piece ended with the advice to name one owner for how your firm handles AI agents, and this piece raises the level of that decision. The operational work still sits across marketing, customer experience, compliance and IT, but the choice about whether to become a backend, and on what terms, belongs with the executive team and, in many firms, the board. Once leadership has made that choice, the agent experience described in our last piece can be designed to match it. If nobody makes it, the choice ends up being made by whichever agents happen to get through.