Amazon Is Not aMused

Written by Brook Schaaf

This isn’t the first time Amazon has slammed its door in an agent’s face. As The Information noted, “The shopping giant has previously blocked shopping bots from OpenAI and Google from browsing its site.” Meanwhile, Amazon’s lawsuit against Perplexity is ongoing.

Yet Amazon’s decision to block Facebook’s Muse generated more headlines and controversy than before. One commenter argued that “many people are overlooking the digital knife fight that’s about to occur” because “Nobody wants to get commoditized or layered.” No less a figure than Elon Musk replied that “Amazon won’t be able to tell whether the buyer is a human or an AI acting on their behalf if access is via the user’s IP address & cookies”—which, I suppose, takes us back to the aforementioned lawsuit.

Nikesh Arora, CEO of Palo Alto Networks, argued that users will want a single agent to do everything. Consequently, “[s]maller players have no choice,” while “[c]ommoditized back ends will need to worry, insurance, tickets, hotels, services—if they don’t adapt new players will.” That may reflect Arora’s own personal preferences; still others said they would rather use different agents for different purposes.

The deepest cut may have come from Eric Seufert, who previously argued in a series titled “Agentic commerce is a mirage” that the “knife fight already took place; it ended a year ago.” The winner, in his telling, was not the independent shopping agent but a merchant’s own on-site assistant, such as Amazon’s Rufus or Walmart’s Sparky.

Who is right? My guess is that the market will segment according to user preferences, merchant tolerance, and agent capabilities. At one extreme, a friend of mine wants to record the contents of his pantry so AI can reorder everything for him. At the other, my wife surveyed subscribers to her meal-planning website, PrepDish, and found that, by and large, they neither use AI nor plan to. The sample may be biased: people inclined to use AI may be less likely to subscribe in the first place. Her next move, incidentally, is to have Claude turn her recipe library into Markdown files.

Somewhere between those extremes is this account from a mother who used an AI agent for back-to-school shopping and came away feeling a little sad. Reading between the lines, the experience also seems less effortless than the headline suggests. My own limited experiments have been disappointing. Rufus, for example, recommended completely incompatible parts for my lawn tractor even with its make and model.

As I have suggested before, if some customers want a shopping agent, affiliates may be best positioned to provide one—especially if the underlying technology becomes commoditized. Some contend that Muse is basically just OpenClaw. Affiliates already have or at least might have structured data about merchants, coupons, discounts, brands, products, and reviews. They may also know a user’s shopping habits, sometimes in considerable detail, allowing them to recommend particular products or competing merchants. In short, they are well positioned to build and maintain a superior agentic offering.

Affiliates could also act as trust agents for merchants, providing some assurance that an agent represents a genuine shopper. Unless, of course, Amazon slams the door in their faces too—which, to Musk’s point, might be difficult if not impossible. Here, the fragmentation of the affiliate industry could actually be an advantage: its agents would be harder to identify, and even Amazon might tolerate some data exfiltration if it were widely dispersed.

Then again, this is the same company that years ago removed product and pricing information from its emails, as ad tech veteran Ari Paparo reminded us this week. More recently, Amazon removed much of the same data from affiliates, wrecking a longstanding arbitrage play. Smaller merchants, however, may not have a sturdy door or the luxury of locking it.

Still, I have yet to see a notable affiliate-built shopping agent emerge. Perhaps Seufert is right—at least about this. He has made some other claims about affiliate marketing that are dead wrong, but those will be the subject of a future blurb.