HTML was built for eyes. The next web will be built for artificial brains.
In 2024, a threshold was crossed in silence. For the first time in a decade, automated traffic surpassed human traffic on the internet: 51% of web requests now come from machines, according to the Imperva Bad Bot Report 2025 (Thales Group). The inversion is complete.
At the same time, Gartner predicts a 25% drop in traditional search engine volume by end of 2026, replaced by conversational AI agents. This is not a marginal projection—this is the firm advising Fortune 500 strategy teams.
The numbers keep accumulating. Google searches ending in zero clicks—where the user gets their answer without ever visiting a site—reached 69% in May 2025 according to Similarweb, up from 56% a year earlier. When a query triggers a Google AI Overview, the zero-click rate climbs to 83% according to multiple converging studies (Bain & Company, Ahrefs, aggregated Similarweb data). The Seer Interactive study from September 2025, covering 25 million organic impressions, measured a 61% drop in organic click-through rate on those same queries. Google traffic to content publishers fell by a third in one year according to Press Gazette and Chartbeat—up to 38% in the United States. Most publishers surveyed by the Reuters Institute anticipate further declines and plan to reduce their SEO efforts in 2026.
The web browser—Chrome, Firefox, Safari—is a visual interpreter. It receives HTML, applies CSS, executes JavaScript, and produces pixels on a screen. This process exists for one reason only: humans read with eyes.
An AI agent has no eyes. When Claude, GPT, or Gemini need to read a web page, they download the HTML, ignore the CSS, ignore the JavaScript, ignore decorative images, ignore navigation menus, ignore cookie banners—and painstakingly extract useful text buried in thousands of <div> tags. On a typical 120 KB page, actually usable content represents only a fraction of the total weight—often less than 10% once markup, scripts, and decorative elements are removed. The agent pays for 100% and uses a fraction.
This waste has a concrete cost. A typical HTML page consumes about 15,000 to 20,000 tokens at an AI provider. The same useful content, extracted and structured in binary format, would fit in a few hundred tokens. For an agent processing thousands of pages per day, the difference is not an optimization—it is an order of magnitude change in cost and performance.
In January 2026, at the NRF trade show, Google launched the Universal Commerce Protocol (UCP)—an open protocol co-developed with Shopify, Etsy, Wayfair, Target, and Walmart that enables AI agents to discover products, authenticate users, and complete purchases without a browser. OpenAI, Anthropic, and Perplexity have entered what Gartner calls the “AI browser war”—not to improve Chrome, but to make it obsolete.
The Model Context Protocol (MCP), backed by Anthropic, OpenAI, Google, and Microsoft, has become the de facto standard for connecting AI agents to enterprise systems. This is no longer HTTP to HTML—it is structured machine-to-machine. Gartner predicts that 40% of enterprise applications will integrate specialized AI agents by end of 2026, up from less than 5% in early 2025.
The future is not a better browser. It is the absence of a browser.
A human wanting to book a flight won’t search “Paris Madrid ticket” on Google to compare 15 tabs. They’ll tell their agent: “Book me a Paris-Madrid next Tuesday, under 200 euros.” The agent will query airline systems directly via structured protocols—not their websites. The transaction will happen in milliseconds, without a single pixel being rendered.
For this future to work at scale, one piece is still missing: a standard format allowing any website to expose its content in structured format to machines, alongside its HTML for humans. Binary formats like CBOR (Concise Binary Object Representation, RFC 8949) already exist and are used in IoT and web authentication protocols. One can imagine tomorrow a single file at the site root, natively readable by agents, weighing 50 KB where the HTML weighs 500. This is no longer science fiction—it is the logical next step.
HTML won’t disappear tomorrow. But it will become what the fax became after email: a legacy format, maintained by inertia, used by a shrinking fraction of traffic. Sites will continue serving HTML for human visitors—as long as there are enough to justify the cost. Meanwhile, the structured channel will become the primary one, carrying sales, transactions, recommendations, and decisions.
The web of 2030 will probably no longer have “pages.” It will have structured data streams consumed by agents acting on behalf of humans. The browser will have joined the fax machine in computing museums. And the question will no longer be “which site to visit?” but “which agent to send?”
Sources:
We help SMBs integrate operational AI agents. Free audit, no commitment.
Request a free audit →