Making the internet better for AI agents might make it worse for everyone else

Personal AI agents that book, shop and negotiate on someone's behalf have gone mainstream fast. MIT Technology Review reports Meta's Muse hit 5 million downloads in its first month and agentic web traffic tripled from August to September. Most of that traffic heads to e-commerce.

The web's response so far is messy. Some sites block agents outright, Amazon is fighting Perplexity's agents in court and has blocked Muse, and Yelp won't let AI book reservations. Others, like Shopify, are building dedicated checkout lanes for agents. Either way, the internet is starting to reorganize around a visitor that isn't human.

We're optimistic about what agents can do for ordinary people. An assistant that contests a bogus medical bill or shops for cheaper car insurance is exactly the kind of useful AI we want more of. The trouble is that the web wasn't built for a visitor that is neither a person nor a classic bot, and the businesses on the other end are now deciding how to treat it.

The worrying scenario isn't agents getting blocked. It's agents getting welcomed on terms that quietly favor the platform. Picture a web where prices differ by whose agent is asking, where recommendations carry undisclosed commissions, and where the human-facing site withers because the money moved to the agent lane. Add constant prove-you're-human checks for the people still browsing directly. That's a worse internet for everyone, including people who never use an agent.

There's a capacity problem too. Wharton's Ethan Mollick expects that once it's common knowledge agents can chase refunds or haggle, customer service lines could be flooded to the point that real people can't get through. That's a plausible near-term harm, and it lands hardest on people least likely to have an agent of their own.

Amazon's appeal in the Perplexity case shows what's really at stake for incumbents. It argues agents threaten its advertising business because ads served to bots are hard to tell apart from ads seen by humans. That's an honest admission that the ad-funded web depends on human eyeballs, and agents break that bargain. Expect business-model defense to show up dressed as security.

Our read: some rules here are easy wins, like keeping card details from being waved around by agents. The contested ones matter more. Can a business block a legitimate agent acting for a paying customer? Can it stop that agent from negotiating? Those answers will decide whether agents become leverage for consumers or another channel for extracting from them. I'm watching the courts and the platforms, because policymakers haven't really shown up yet.

The article draws on recent reporting that people struggled to use agents on sites from Pizza Hut to Yelp, and notes a judge recently ruled for Perplexity against Amazon, which has appealed. OpenAI's Dots and Meta's Muse are among the consumer agent apps credited with driving the surge.