Latest
AI Affairs, home

Thursday 19 November 2026

Opinion

Nadella is right about AI agents, but the web faces more than a reshuffle

If agents replace visits to shops and other sites, charging for human attention becomes a poor way to earn from the transactions those agents arrange.

Microsoft CEO Satya Nadella smiling in glasses and a dark shirt
Photo: Brian Smale and Microsoft, CC BY-SA 4.0, via Wikimedia Commons (cropped)

An internet organised around human attention cannot simply change its price tag when software starts doing the looking. That is the difficulty beneath Microsoft chief executive Satya Nadella’s warning at the New York Times DealBook Summit on 18 November that AI agents could alter the economics of web traffic, TechCrunch reported. A business that earns money when a person visits its site faces a different bargain when an agent visits on that person’s behalf.

Nadella is right about the pressure on advertising-dependent businesses, but “reshuffle” is too mild a word for the transactions in which an agent replaces the visit. A site can sell an advertiser a person’s attention only when that person is there to give it. If the customer delegates discovery and purchase to software, the site may still have something valuable to sell: access to its products, a referral or a completed sale. It cannot count on selling the same view of a page.

Nadella’s move from views to transactions

Nadella told the summit that businesses making money from advertising shown to human visitors might move towards cost-per-click or transactions. He compared the possible change in who pays whom for traffic with the earlier move from web portals to search engines. The comparison captures the contest for a place between a customer and a seller. It also risks making this transition sound like another contest over where people will look. An agent’s value to its user may lie precisely in reducing the number of places that person has to look.

Cost-per-click could remain useful when an agent sends a customer to a site. A payment tied to a transaction could serve a merchant whose product an agent finds and buys. Neither is a straightforward replacement for advertising sold around visits. The buyer of that placement wants the chance to influence a person. The buyer of an agent referral would be paying for a route into a decision made partly through software. Those routes need different terms, and the party controlling the agent would have considerable say over them.

The strongest objection is that people will continue to browse, compare products and respond to adverts themselves. Agents could become another source of traffic rather than the route through which customers do everything. That is a serious possibility, and it makes any prediction about the end of web advertising premature. But a business need not lose every human visitor for the economics of its agent-handled sales to change. For those sales, the question is how a customer finds and pays a merchant when the customer never sees the merchant’s advertising space.

Amazon’s block and Shopify’s possible opening

Amazon blocks AI agents from browsing its retail catalogue and making purchases for users. Nadella argued that a shopping platform such as Shopify might gain an advantage by allowing agents in if Amazon continued to refuse them. That is his proposed outcome, not a report that Shopify has taken such a step or won customers from Amazon. It identifies a choice retailers face: keeping agents away can preserve control of access to a shop, but it also prevents customers from using those agents to shop there.

The case for a block deserves more than dismissal as resistance to change. A retailer accepting instructions from software acting for a customer has to decide whose instructions count and who deals with a purchase that goes wrong. AI Affairs reported that Citi puts customers on the hook for third-party AI agents and account hacks. That example concerns account access rather than shopping, but it puts responsibility where the agent debate too often puts convenience. Access terms can protect customers and merchants. They can also protect an incumbent’s position. Customer protection alone cannot settle how far a block should go.

Nor would accepting agents automatically make a retailer more open. A platform could admit them on terms that favour its own commercial relationships or give merchants little choice over how their products appear. Nadella’s challenge to Amazon is persuasive as a warning about refusing a route customers may want to use. It is less persuasive as a complete account of what replaces the blocked route. For merchants, the price of access and the ability to contest an agent’s treatment of their products would be as consequential as the permission to enter.

Microsoft’s stake in agent access

Nadella speaks for a company with its own interest in how much work users delegate to AI. AI Affairs reported that Microsoft put three Office apps inside Copilot. His argument about web access should therefore be treated as a proposal from a prospective participant in the new arrangement, not as a neutral account of its likely terms. If agents become an important path to customers, the companies operating them may gain bargaining power over the retailers and publishers whose pages they traverse.

That does not make Nadella wrong about the weakness of a model built on visits an agent can bypass. It makes the choice of replacement more important. Moving from impressions to clicks or transactions changes who can charge whom, and for what. A retailer’s refusal to admit agents may prove costly if customers prefer to use them. An agent provider’s demand for unrestricted entry would still leave merchants and customers asking who sets the terms. Neither side earns authority over the other merely by claiming to know where the market is headed.

Microsoft and retailers considering agent access should work towards terms that let customers authorise purchases, let merchants understand how agents find them and assign responsibility when a transaction fails. They should publish evidence of whether agent visits produce sales that would otherwise have come through human browsing, and of what merchants pay for that route. If customers keep visiting sites themselves and advertising continues to support those transactions even where agents are accepted, the case for a fundamental break in web economics would need revising.

Sources

Topics: Agents, Enterprise adoption

What should we cover? Tell us

The weekly briefing

The AI Affairs weekly briefing: the week's 10 most important AI stories in business, government and markets, every Monday. We'll email you a link to confirm. Every email has a one-click unsubscribe. Privacy policy