Meta's personal AI agent app, Muse, was downloaded 5 million times in its first month, and agentic web traffic tripled between August and September. That surge is forcing websites into a choice: block the bots, or build infrastructure for them. Shopify is already doing the latter, giving AI agents their own checkout lanes that skip the pages designed for humans.
The split isn't academic. Amazon has blocked Muse's agents and is in a court battle with Perplexity over its agents. A judge recently ruled in favor of Perplexity. Amazon has appealed.
The legal question buried in the bot fight
The Amazon-Perplexity litigation isn't really about whether bots can scrape. It's about whether an agent acting on behalf of a logged-in user inherits that user's authorization. The judge's pro-Perplexity ruling suggests the court saw it that way. Amazon's appeal will test it.
đź“Š Market Data Snapshot
That matters beyond e-commerce. Every agent-payment standard — on-chain or otherwise — has to answer the same question: how do you prove an autonomous agent is authorized to act for a human who isn't present at the transaction? If courts say agents are legally the user, decentralized delegation models get a legal foundation. If they don't, every agent-payment protocol inherits a liability layer that centralized platforms like Shopify can absorb and open protocols can't.
Why the blocking isn't about security
Yelp doesn't allow AI to book restaurant reservations. Amazon blocks Muse's agents outright. Neither company is technophobic. They're defending revenue models.
Ad-supported sites have an existential reason to keep agents out: agents don't view ads. Transaction-revenue sites, like Shopify, have the opposite incentive — an agent that completes a checkout is a sale, not a lost impression. The "vast majority of agentic web traffic is to e-commerce sites" stat likely understates the real volume, because Shopify's dedicated agent lanes bypass the human-facing pages where analytics normally count visitors.
That reframes the whole debate. The internet isn't splitting into agent-friendly and agent-hostile. It's splitting into ad-model sites and transaction-model sites, and the two are making opposite bets on the same traffic.
The identity problem nobody's priced
When Shopify gives an agent its own checkout lane, that agent needs two things a human doesn't: a persistent, verifiable non-human identity, and a payment credential that isn't a stolen credit card. That's decentralized identity, delegated spending wallets, and agent-scoped payment tokens — infrastructure that exists but has no merchant integrations at scale.
The first adversarial move in an agent economy is agents impersonating other agents or humans. The only scalable defense is cryptographic attestation. Ethan Mollick, an associate professor at the Wharton School of the University of Pennsylvania, has been cited in coverage framing this as an organizational-behavior shift, but the more concrete question is what happens when machine-identity fraud becomes a real liability for merchants.
Decentralized identity registries and verifiable credentials are the picks-and-shovels play. Agent tokens are the narrative trade. They're not the same thing.
What to watch
Amazon's appeal is the next concrete marker. The docket will determine whether agents are treated as users or trespassers, and that binary will shape which agent-payment standards get merchant adoption over the next six to twelve months. Shopify's model works if courts say agents can transact on behalf of users. If the appeal flips the ruling, the checkout lanes become a liability.
The 5 million Muse downloads and the tripling traffic number are almost certainly inflated by app-store bundling and OpenAI's Dots rollout. Download-to-active-agent conversion is typically in the single digits. The real signal isn't installs. It's retention and transaction completion rate — and nobody is reporting those yet.
