Amazon won't let Meta's shopping AI onto its site, but builds its own instead — why the double standard?

This explores why a platform would shut out someone else's shopping agent while running its own agents elsewhere. In short, the corpus points to money, not technology. One caveat first: the collection does not cover Amazon's specific decision. What it does give you is a clear explanation of why platforms make this kind of choice, and that explanation fits the Amazon case closely.

The key idea comes from Azhar's argument about Muse Can an AI agent serve both merchant and user interests fairly?. Muse earns referral fees from merchants like Expedia. An agent paid that way can't be fully loyal both to you and to the merchants who pay it. Azhar's broader point is that these incentives decide which platforms build agents and which ones block them. Seen this way, Amazon's position makes sense. Its own agents work inside an economy Amazon controls. An outside agent paid by other merchants would be taking customers through Amazon's store while serving someone else's business. Blocking it isn't hypocrisy. It's the same logic in both directions: each company wants the agent that does the shopping to answer to it.

Another note shows why this fight matters so much Will agents compete for attention just like users do?. As people hand off shopping and booking to agents, stores stop competing for your clicks. Instead they compete to be picked by an agent. Ranking, discovery and something like ad placement all move into how the agent decides. Whoever owns the agent owns that decision, the way search engines once controlled what people saw. A platform that lets a rival's agent shop freely on its site gives that rival the power to decide what gets chosen.

There's a less obvious parallel in the corpus's notes on agent safety. Those notes argue that you can't control an agent by trusting its internal rules or filtering what it says. Real control comes from limiting what it can reach, through access tokens, restricted tools and permission layers set from outside Can agent safety rules stop destructive API calls in real time? Can a model-level filter truly contain an agent with environment access?. These notes are about safety, not business, but the logic carries over. A platform can't audit whose interests an outside agent is really serving. The one limit it can actually enforce is whether that agent gets in at all. Blocking is the commercial version of an access boundary.

The question to take away isn't really "why Amazon?" It's this: when an agent shops for you, who is paying it? The corpus suggests that answer will shape which agents you're allowed to use, and where, more than how good the technology is.