Agentic commerce is a mirage

In late July, The Information reported that Amazon had begun blocking external chatbots and shopping agents from accessing product listings on its website. From the piece:

In recent weeks, Amazon updated the code underpinning its website to add language warding off new AI agents from Google. These changes follow restrictions Amazon added earlier this year calling out bots from Perplexity, Anthropic’s Claude and OpenAI’s ChatGPT … Amazon’s changes come in the form of updates to instructions in files known as robots.txt pages, which aren’t legally binding but signal the rules the sites expect AI firms to play by. Amazon’s robots.txt file currently disallows Google’s Project Mariner agent as well as another Google shopping agent, and it also disallows crawlers from OpenAI, Perplexity and Claude.

A few weeks ago, ModernRetail noted that Amazon had expanded this effort by increasing the number of agents blocked in its robots.txt file. This contrasts with the approach implemented by Shopify, which has developed an MCP server specifically to support agentic commerce across its retailers. But, as noted by The Information, this comes with a catch: agents cannot complete transactions without human review, and they must integrate Shop Pay to make purchases.

These divergent approaches highlight each company’s strategic priorities. Amazon generated $15.7BN in advertising revenue in Q2 2025; its advertising business generated $56.2BN for the full year 2024. While Shopify does operate several advertising products (see my podcast episode with the company’s Director of Advertising Products), it doesn’t currently generate meaningful revenue from advertising. So while both companies impose restrictions on agentic access to their underlying merchant data, Amazon’s restrictions begin at the point of discovery (because it wants to retain advertising primacy) and Shopify’s start at the point of purchase (because it wants to retain transaction primacy).

But, critically, neither company allows for unfettered access to its platform by agents. There are several reasons why they would prefer not to:

• Amazon derives as much, if not more, value from the user-level data artifacts created by a transaction than from the transaction itself. That data is used for advertising (both on-platform, including on Amazon Prime Video, and on third-party sites, through its DSP), for recommendations, and for pricing. User-level behavioral data is mission-critical for Amazon; maintaining exclusive access to it is a strategic imperative.
• Shopify is the retail payments platform. And, again, the data generated by any given Shop Pay transaction is arguably more valuable to Shopify than the revenue generated from that transaction. And while Shopify might see less value from that data than Amazon for advertising purposes, I think that’s probably only a temporary reality; ultimately, I do believe Shopify will pursue advertising in earnest as a scaled revenue opportunity, which will diminish the appeal of agent-sourced transactions.

While Amazon and Shopify have already limited (in Amazon’s case, completely) the scope of agentic commerce on their platforms, I’ve seen two counter-arguments be proposed to suggest that their policy is a strategic mistake:

1. Both platforms are depriving themselves of incremental transactions by not giving agents full purchase access to their catalogs.
2. That consumers will embrace agentic commerce to an extent that sidelines both of these companies.

While the first counter-argument is true, the additional transaction opportunity is likely marginal at best, and this completely discounts the value of the data generated by owning the relationship and the transaction.

The second counter-argument misses a broader point: if agentic commerce becomes a well-worn consumer behavior, why would it not be catalyzed and captured by the platforms themselves? The digital advertising ecosystem is instructive: the open web has given way to walled gardens and content fortresses for a multitude of reasons, resulting in the Everything is an Ad Network phenomenon. Walled garden economics are more compelling than those of open ecosystems; why would Amazon or Shopify yield the agentic commerce experience to a third-party product that disintermediates them from their users? Amazon already offers its users shopping agents: Rufus, a chatbot, and “Buy For Me,” which allows users to search for products on third-party websites. What incentive does Amazon have to make its products purchasable from third-party agents?

The counterargument I’ve seen submitted for this is that the entire advertising market will be disrupted by independent commerce agents when platforms can advertise to the agents who make purchases on behalf of their users, as articulated by the CEO of Perplexity in an interview with India’s Economic Times. I view this outcome as far-fetched, if not preposterous. I outlined my skepticism toward this dynamic in this X thread, but briefly:

• No coherent incentive structure supports advertising to agents. Who is getting paid to show the ad? If the agent receives money (either through an advertising fee or an affiliate-style commission), their objectivity is compromised.
• How does the agent budget purchases on behalf of their user? Is it given a recurring, fixed amount of money each month to eg., buy staples like groceries? If so, the agent would likely only respond to discounts and promotions, which could lead to hoarding and front-loading by consumer agents. This would distort the pricing of the products being advertised and result in those companies scaling back on advertising altogether.
• Any ad filtration performed by the agent effectively amounts to ad blocking, which would disincentivize participation from advertisers.

But perhaps the most compelling and convincing argument against the notion of agents as recipients of advertising is that the largest retail platforms simply won’t participate in or support it. These platforms will seek to preserve their own advertising businesses. They prefer to layer their own agents into the retail experience, where they control the surface area and can monetize it with advertising.

None of this is to say that I don’t view agentic commerce as unlikely or impractical. Quite the opposite: I see it as inevitable. But agentic commerce as a standalone, independent enterprise strikes me as unworkable and untenable, given that it runs counter to the incentive structure of the largest retail platforms. For a commerce agent to achieve scale, it wouldn’t merely need to provide a shopping experience that matches what retail platforms — which have and will continue to launch their own agents — can provide, but also to build the retail backend against which the agent would operate.

In other words: siphoning consumer attention away from Amazon isn’t a function of building a better search experience but of building a better Amazon. In this way, independent agentic commerce is mostly a mirage: it’s not a standalone category but rather a means of concentrating engagement on the largest retail platforms as an extension of the existing shopping experience. As such, it might be viewed as analogous to the AI enhancements implemented by social media platforms in pursuit of increased engagement.