Just days after Meta’s Muse was sent out into the wild to scour the web for your next coffee maker or new pair of shoes, the cuddly agent known as Jolly got stopped in its tracks: Amazon, the world’s biggest digital marketplace, had locked its doors and thrown away the key.
At the time, Amazon had a blunt response, writing: "We think it's fairly straightforward that third-party applications that offer to make purchases on behalf of customers from other businesses should operate openly and respect service provider decisions about whether or not to participate.”
The company said it had not received prior warning for the agent crawling its site, and Muse didn’t identify itself when it did.
The block led many proponents of agentic commerce – where an agent autonomously makes purchases based on a users’ parameters after doing its own research and price comparisons – to ask an uncomfortable question: With the biggest tech companies building their own agents, will industry fragmentation bring down the whole idea of an end-all, be-all, AI assistant that can buy things for you anywhere across the web?
The answer might not be so black-and-white. Peyman Shahidi, a PhD student at MIT focused on AI agents and labor economics, said the fragmentation issue may not have been anticipated before the introduction of mainstream AI assistants this year.
“It was expected that agents are going to start shopping on our behalf, but the extent to which they are able to do so, how they are going to operate on these marketplaces, this is something that was unknown,” Shahidi told TWIF. “I could have imagined that Amazon could have let Muse operate on its platform, and things would have gone very differently from how we are observing it is rolling out today. So it's not something that is baked into the technology. It is up to our collective decision making how it goes.”
When Muse was blocked by Amazon, the company said that it was doing so for security reasons, and that other sites have similar responsibilities. The e-tailer said it was all about ensuring a “safe, secure, and reliable” experience for customers.
Meta did not immediately respond to TWIF for comment on the block.
Meta’s Muse, which dropped in September, has taken the agentic space by storm, becoming one of the most popular free apps on the market.
‘Move Fast and Break Things’
“This is classic Meta. Like, move fast and break things,” Georg Zoeller, a former business engineering director at Meta, told This Week in Fintech. “These things cannot be secured.”
And although security is one piece of the puzzle of allowing AI agents onto their sites, many experts noted that advertising probably has the biggest impact on whether merchants will allow the virtual assistants through their digital gates.
Amazon shoppers know that sponsored product ads are everywhere on the marketplace; however, an agent shopping for one specific product never sees these. Throw in the customer data that would disappear if agents did all the buying and you have a recipe for retailers like Amazon to wall off their sites.
“Any website that is financed by ads loses money if you send an AI agent,” Zoeller said. “They cost the same money to serve as a human, right? Traffic bandwidth, whatever it is. But they, at the same time, could also be a scraper for another service.”
Marketplaces like Amazon rely on people browsing, clicking ads and making impulse buys. For an independent brand, it doesn't matter whether a human or a bot clicks "buy," a sale is a sale.
That difference is why Meta has leaned heavily into formal partnerships with Shopify – which makes money off of subscriptions and merchant tech – rather than trying to force its way into closed marketplaces, and why the commerce platform has opened itself up to other AI agents as well.
“Platforms whose business models are very dependent on humans crawling on their website are the first point of impact,” Shahidi said. “It is up to them whether they would want to change their business model, allow agents to operate on their platform, and they shift their monetization toward other sections of the e-commerce economy of the world.”
Not All Partnerships Work Out
Still, merchant collaborations with tech are anything but a sure thing. Last year, OpenAI partnered with Walmart. The initial plan allowed consumers to make purchases directly through ChatGPT. That eventually fell apart just months later after sales weren’t as high as anticipated, and Walmart instead installed Sparky, its own chatbot, into Google Gemini and ChatGPT, WIRED reported at the time.
Walmart did not immediately respond to a request for comment on where their partnership stands now, but OpenAI confirmed it was still ongoing.
“It creates a dichotomy for the merchants, saying, ‘OK, where do I enable? Where do I stop? It's hard for them to figure out,” said Kumar N Senthil, the CEO of Firmly.AI. His company helps connect merchants to AI agents.
“Because the liabilities would be [on] the merchant, if something goes wrong, they have to carry the bag,” Senthil said. “Which means they need the guardrails in place.”
For now, Amazon’s door slam makes one thing clear: the vision of a single, universal AI agent that can buy anything for anyone across the open web isn't happening for now, at least. Instead, agentic commerce is splitting into two camps—an open ecosystem of independent merchants welcoming bots to close the sale, and walled-garden giants determined to keep their checkout counters entirely to themselves.
