AivexaNewsSearch
AI news for builders and product teamsChecked every hour

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

Collected Oct 19, 2026

Personal AI agents that act on a person's behalf online have arrived at scale, and the web is already reacting badly to them. According to MIT Technology Review, the shift follows the launch of consumer apps such as OpenAI's Dots and Meta's Muse. Muse was downloaded 5 million times in its first month, and agentic web traffic tripled between August and September.

Until now the internet sorted visitors into two buckets: bots and people. Bots were allowed to do automated work like scraping pages, but under rules about what they could reach and how they behaved, and the web itself was built primarily for humans, who occasionally solve captchas to prove they aren't machines. Agents break that model. They are not human and they operate at machine speed, but they are not ordinary bots either, because they carry out legitimate tasks for an individual at far greater pace than that person could manage alone.

Because agents don't fit existing bot-or-human checks, security-conscious platforms have responded by banning them. TechCrunch reported users struggling to send agents to sites from Pizza Hut to Yelp, which does not allow AI to book restaurant reservations. Amazon has blocked Muse's agents and is litigating against the startup Perplexity to block its agents as well. A judge recently ruled for Perplexity, and Amazon has appealed, arguing that agents threaten its advertising business by blurring the line between ads served to humans and those served to bots.

Ethan Mollick, an associate professor at the Wharton School of the University of Pennsylvania, expects the friction to spread across the whole web. In his view, once people widely understand that agents can request refunds or negotiate prices, agent traffic will overwhelm customer service channels until real people can no longer use them. He wrote on LinkedIn that humans will eventually become very annoyed when their agents are barred from acting on their behalf.

Economics points the other way. Most agentic traffic goes to e-commerce sites, so platforms stand to profit by admitting agents that shop for real customers. Shopify is already doing so, giving agents dedicated checkout lanes that skip the pages built for people.

The open question is what this means for everyone else. The web has increasingly been designed to capture human attention, but humans have at least remained its intended users. That could give way to a web tuned for agents that may eventually outnumber people, and the article sketches how that might feel. Sites could be neglected as destinations for humans, leaving poorer interfaces for anyone who wants to browse and compare products directly. Human-verification systems are imperfect, so people may face constant proof that they are not agents, possibly extending to iris scans. Pricing could fragment, with the same sweater shown at different prices to different people's agents, and it may become impossible to tell whether an agent's recommendation was shaped by advertising or platform commissions.

That outcome contrasts with the future promised by Google, Anthropic and OpenAI, in which agents give consumers more power and bargaining leverage. The article argues the promise isn't far-fetched: agents can plausibly contest bogus medical bills or hunt for cheaper car insurance. Whether they get to do so depends on rules governing how agents interact with sites and what they are permitted to do, and those rules are still being written. Some will clearly help everyone, such as preventing agents from spreading credit card details across the web and other security measures. The contested ones, like whether a business may block agents or stop them from negotiating better deals, will decide whether the web becomes a channel for agents working for users or against them.

For developers and product teams, the practical effect is that agent access becomes a product decision rather than a technical default. Sites will increasingly need to identify agent traffic, decide which tasks to permit, and expose machine-friendly paths such as Shopify's checkout lane, while keeping fraud and ad-integrity controls intact. This suggests teams building consumer agents should expect uneven access and design for graceful failure when a site refuses, and that businesses may face pressure to publish clear agent policies before courts or competitors set the terms for them.

Read at MIT Technology Review · AI

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

This story originally appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first, sign up here. Until recently, you could divide internet users into just two categories: bots and people. Websites have long allowed bots to perform automated tasks, like capturing the content of web pages, but rules…