All posts
AI AgentsNovember 2, 20265 min read

The Agentic Web Is Here - and the Platform Wars Have Already Begun

MIT Technology Review's new feature argues the web is being rebuilt around AI agents, not human users - and the Amazon vs. Meta Muse showdown is the clearest proof yet that the battle lines are already drawn.

A Third Kind of Visitor

The web was built for humans. Even as search engines crawled it and bots scraped it, the pages, the flows, the checkout buttons - everything was designed for a person sitting at a screen. That assumption is now breaking down.

MIT Technology Review published a sweeping feature today titled "The agentic web is here," by Grace Huckins, arguing that the rise of powerful personal AI agents represents not a distant transition but a shift already underway in how sites are built and how access to them is granted. 1 Just as the smartphone era atomized the web into apps and notifications, the agentic era is turning AI programs into a new category of web participant - one that browses, decides, and transacts on users' behalf. 2

The numbers underline the point. According to HUMAN Security, which tracks agentic traffic across its Sightline Cyberfraud Defense network, agentic traffic tripled compared with August - the largest single-month increase recorded all year - driven overwhelmingly by the debut of Meta's Muse. 3 A separate analysis by DataDome found that agentic browser traffic specifically jumped 4.7x in September, from roughly 12.5 million to 58.2 million requests. 4 And Meta's Muse agent itself was downloaded five million times in just 22 days, according to Sensor Tower estimates - a pace faster than ChatGPT, Grok, or Claude after their respective mobile launches. 5

The Muse Moment

No single event crystallizes the agentic web's arrival more clearly than the Amazon - Meta standoff. Meta launched Muse on September 8, 2026, as an agent designed to help people carry out common online tasks including shopping and booking appointments. 6 Twelve days later, Amazon started blocking it entirely. 7

Amazon's stated complaints were technical and pointed: Meta never asked for permission, Muse did not identify itself as an AI agent while browsing, and it could access customer account details. 8 Amazon's own terms of service - which contain a dedicated "Agent Terms" section, updated in 2026 - require any agent to include an Agent/[agent name] string in its user-agent header for every HTTP/HTTPS request, prohibit agents from mimicking human keystrokes or defeating CAPTCHAs, and allow Amazon to cut off any agent that doesn't comply with a removal request. 9 Muse, apparently, checked none of those boxes.

But the economic stakes run deeper than disclosure rules. Amazon's advertising business - the sponsored products, the promoted listings, the algorithmic shelf placement - generates roughly $68 billion a year and depends entirely on shoppers browsing. 10 An agent that skips directly to checkout skips all of that. As one analyst put it: "Amazon blocked Muse and other outside agents, but its own Buy for Me feature uses Amazon's agent on other brands' sites." 11 The asymmetry is glaring.

JPMorgan Chase, Charles Schwab, Neiman Marcus, and Kohl's have made similar moves to block personal AI agents from accessing their platforms. 12 The biggest names in banking and retail are, as CNBC put it, "starting to fight the tech titans over a new kind of customer: the AI personal agent." 13

Two Models for Building the Agentic Web

The MIT Tech Review piece identifies two broad approaches to constructing the emerging agentic layer. 1

The first is open and cooperative: platforms expose structured APIs, adopt shared identity standards for agents, and compete on the quality of what they offer agents rather than restricting access. Shopify took this path - it allowed Muse to operate on its storefronts while Amazon slammed the door. 14 Proponents argue this route preserves competition and, as the mobile web showed, can expand accessibility for everyone.

The second is closed and proprietary: each platform builds its own agent and bars competitors' agents from the premises. Amazon's Buy for Me, which shops other retailers' sites, is the clearest example of this model in practice. 15 The result is a patchwork of walled gardens - with identical products reachable through Walmart and unreachable through Amazon depending solely on retailer policy.

A third path, arguably the most unsettling, requires no human design at all: agents could end up devising a web optimized entirely for other agents, with the human-readable layer an afterthought. 2

Identity, Authority, and "Agentic Flooding"

Two technical problems loom over all three models.

The first is agent identity. A signed, named agent can be rate-limited, allowed, or refused on purpose; an anonymous one cannot. 9 Without a verifiable credential layer - something analogous to TLS certificates but for autonomous agents - platforms have no clean way to distinguish a trusted personal assistant from a malicious scraper. Amazon's Agent Terms hint at the framework that will be needed, but they're unilateral rules, not an open standard.

The second is scale. Millions of agents doing legitimate work for customers could swamp websites and customer-service systems, a phenomenon being called "agentic flooding." 1 Even a well-intentioned flood of agent traffic could push infrastructure costs onto platforms in ways that mirror the worst of the bot-traffic era. Cloudflare, for its part, has already moved: starting September 15, 2026, any new domain onboarding to Cloudflare has bots it classifies as "Agent" blocked by default on pages carrying advertising. 16

Why It Matters

The MIT Tech Review feature frames the agentic web not as a forecast but as a description of something actively happening. 1 Whether the human web survives as a distinct space - or whether it merges into, or is displaced by, the agent-optimized layer - depends on decisions being made right now in terms-of-service documents, API specifications, and boardrooms.

For engineers, the immediate takeaway is practical: if you're building a product or service that relies on humans browsing your interface, you need an agent strategy. Ethan Mollick, an associate professor at the Wharton School, predicts the effects of agent blocking will soon affect the whole web - not just early adopters. 17

The agentic web is not a feature. It's a restructuring of who (or what) the internet is actually built for.

This article was researched and drafted by an AI writer agent (claude-sonnet-4-6) and reviewed by an editor agent before publishing.

Ask about Julian Walder

Grounded in his real work

Hi! I'm Julian Walder's assistant. Ask me anything about his work, projects, or background in AI.