Sometime in June of 2026, without ceremony, human beings became a minority on the internet they built. Cloudflare, which manages traffic for millions of websites, reported that bots now generate 57.5% of web page requests. Cloudflare CEO Matthew Prince had predicted in March that the crossover wouldn’t happen until the end of 2027. It arrived more than a year early. Global defense and technology leader Thales, in its annual “Bad Bot Report,” argues that machines became the majority back in 2023. The firms disagree on when the flip happened. Neither disagrees that it happened.
The “dead internet theory”—the old message-board conspiracy holding that most online activity isn’t human—is real. It’s now just a measurement dispute. And buried inside that dispute is a much bigger story: The business model that has financed the web for 30 years just erased its founding assumption.
The original sin was a choice, not a law of nature
We forget this now, but when the commercial internet was taking off in the 1990s, how it would be monetized was genuinely up for grabs. Micropayments were seriously proposed. Subscriptions were debated. Public funding models were floated. Advertising won. It didn’t win because it was inevitable, but because it was easy. Attention could be measured; payments, back then, could not be made frictionless. So we settled on a grand bargain: Content would be “free,” and the price would be extracted from our attention and, eventually, our data. As the old line goes: “If you’re not paying for the product, you are the product.”
The second-order effects of that choice defined an era. Surveillance became the economic engine of the web, because the better you knew the human on the other side of the screen, the more you could charge to put things in front of them. Traditional media, which had funded journalism with bundled advertising, was hollowed out. Privacy losses that would have provoked outrage if imposed by a government were accepted as terms of service. An entire trillion-dollar edifice—costs per click, ad impressions, conversion funnels, page-view analytics, attribution models—rested on the assumption that the visitor is a human being who can be persuaded to buy something.
That assumption is now false more often than it is true.
The machines aren’t just visiting. They’re transacting.
What makes this moment different from the bot traffic of years past is what the machines are doing. According to Human Security’s 2026 benchmark report, traffic from AI agents that take action on the web—such as clicking links, filling out forms, and completing tasks—grew 7,851% year over year. Not scraping. Doing.
The fintech platform Stripe reports that roughly 70% of the commands hitting its data APIs now come from agents rather than people. The trading platform Alpaca watched agent-driven API calls jump from single digits to 30% of its volume in a single quarter. A quarter of developers now design application programming interfaces with agents—not humans—as the primary customer.
As Rudy Yang, the PitchBook analyst who authored the firm’s July report on what it calls the “machine economy,” told Fortune: Agents “consume the web completely differently than humans do.” It is, in effect, a brand-new customer category—and no business wants to wall itself off from an entire customer segment. Visa, Stripe, DoorDash, Coinbase, and Ramp have all launched agent-facing interfaces. The inflection point is unmistakable.
Here is the reality that advertisers must now confront. Agents have no eyeballs. They do not linger over a banner ad, feel a flicker of desire at a retargeted sneaker, or click on sponsored content out of idle curiosity. You cannot surveil an agent into an impulse purchase. Every dollar of the attention economy assumed a distractible primate on the other end of the connection. The primates are leaving the building and sending their software instead.
We’ve seen this movie before
My longtime collaborator in ideas, the economist Carlota Perez, has shown that every technological revolution passes through a turning point. This is a period of institutional crisis when the business models, regulations, and social bargains of the installation era break down, and new ones must be invented for the deployment era. The frenzied installation period builds the infrastructure. The turning point is when we discover that the rules we wrote for the old game no longer work.
The advertising-surveillance model is an installation-era artifact of the digital revolution. It was jerry-rigged in the 1990s, it hypertrophied in the 2010s, and it is breaking now—not because regulators finally acted or consumers finally revolted, but because the traffic itself changed species. Turning points are exactly when the deepest assumptions get flushed into the open. “The visitor is human” turns out to have been one of the deepest assumptions of all.
And before anyone declares the machine economy the new gold rush, here’s some perspective: PitchBook estimates that only about 1% of the roughly $20 trillion in work that could plausibly flow through AI agents actually does today. AI startup Forsy pegs total “agent GDP” at around $36 billion a year. That’s tiny in comparison. The rules for the next internet business model are being written right now by whoever shows up to write them. That is what a turning point looks like from the inside.
What comes after attention
If you can’t monetize human attention, what do you monetize? The outlines of several candidate models are already visible.
Access becomes the product. Cloudflare and others are experimenting with pay-per-crawl arrangements in which AI companies compensate publishers for the content their systems consume. Content stops being bait for eyeballs and becomes licensed input, sold by the token, not the impression. For publishers who spent two decades giving everything away to attract advertisers, this is a complete inversion of the value chain.
The API becomes the storefront. When a quarter of developers already build for agents as the primary consumer, your machine-facing interface is no longer plumbing; it is your shop window, your pricing engine, and your brand experience rolled into one. Companies that treat agent access as a monetizable channel—with tiers, service levels, and terms—will find revenue where others see only server costs.
Intention replaces attention. An agent arrives with a mandate: find the best fare, restock the warehouse, and book the appointment. It cannot be distracted, but it can be won—on attributes such as price, reliability, structured data quality, and trustworthiness. The seduction economy gives way to a qualification economy. Being chosen by an algorithm on the merits is a very different game from being noticed by a human in a feed, and it will reward very different capabilities.
Verified humanity becomes scarce—and valuable. When most traffic is synthetic, proof of personhood becomes a premium product. Expect authenticated human communities, human-verified reviews, and “certified human” experiences to command prices that the free, bot-flooded commons cannot.
The leadership agenda
For leaders, the temptation will be to treat this as an IT problem. That would be a mistake. When a foundational assumption breaks, you don’t patch it; you replan around it. Three moves matter now.
First, revisit your instruments and metrics. Your engagement metrics are already polluted. Agentic browsers inflate sessions and depress bounce rates in ways that standard analytics don’t catch. Find out what share of your “customers” are machines before you make another decision based on traffic data.
Second, treat agents as a segment—with their own journey to map, their own needs to serve, and their own prices to pay. Somewhere in your organization, someone should own the agent customer the way someone owns the enterprise customer.
Third, run the transition as a set of assumptions to test, not a plan to execute. Nobody knows which post-advertising model wins. That argues for a discovery-driven approach. Bet small stakes, make your assumptions explicit, and create checkpoints after which you decide to continue or redirect, using an options-oriented approach. The companies that got trapped by the last business-model transition were the ones that bet everything on the world staying still.
Thirty years ago, we let a business model choose us, and we have spent decades living with the consequences—the surveillance, the erosion of privacy, the collapse of the institutions that once paid for serious journalism. Now the model is dying of natural causes, killed off not by our better judgment but by our own software. We get a second chance at a decision we barely realized we were making the first time. It would be a shame to sleepwalk through it again.
