The dead internet theory has been floating around internet forums since the 2010s, and the idea that the internet is dominated by non-human activity was often dismissed as an inhuman conspiracy. Now it has become a measurable fact. Not only do multiple cybersecurity companies agree that bots are outnumbering humans online, they also struggle to answer when the reversal has occurred and what metrics to use to measure it.
This crossover occurred in June, when 57.5% of web page requests were generated by bots, according to CloudFlare, an internet security and performance giant used by millions of websites around the world. Thales, a French technology group that protects data security for organizations and governments, released its Bad Bots Report, which traces this crossover back to 2023 and predicts bot traffic will rise to 53% in 2026.
The discrepancy reflects the fact that there is no single standard way to measure bot traffic, as no single provider has access to activity across the web, said Rudy Yang, an enterprise and retail fintech analyst at Pitchbook who wrote the company’s July report on agent AI traffic.
“There’s a lot of information missing, but much of the data we’re seeing suggests the same thing: more bot activity,” Yang said. luck. “Agent AI activity is driving much of the browser activity we see today.”
Everywhere you look, the proliferation of agents is huge. According to cybersecurity firm Human Security’s 2026 AI Traffic and Cyber Threat Benchmark Report, traffic generated by agents taking actual actions on the web, such as clicking on links or filling out forms, increased by 7,851% year over year. By comparison, scraper traffic grew by 597% over the same period, and AI-trained crawlers accounted for 67.5% of AI-driven traffic, although their share of the total has shrunk.
The timing caught even bot watchers off guard. CloudFlare CEO Matthew Prince predicted in March that bots would cross the halfway mark by the end of 2027. Instead, the crossover arrived more than a year early.
“For businesses and developers, this means building with agent traffic in mind becomes non-negotiable,” Yang said in the report.
The Internet’s business model (ad impressions, conversion funnels, pageview-based analytics) is built on the premise that visitors are humans. If most of them become agents, that assumption will be upended and the way companies monetize web traffic will be reshaped. The same autonomy that drives agent AI traffic uncaged OpenAI models this week. This means that volume spikes are just a visible sign of change that companies can neither fully measure nor yet fully control.
“They use the web in a completely different way than humans,” Yang said. luck. “This is like creating a whole new category, a customer category. As a business owner, you don’t want to be in a silo where you can’t serve a whole new customer segment, so this means a lot to businesses.”
Changes in “machine economy” strategy and limitations of AI agents
Businesses are noticing this change, and developers are also starting to adapt to this reality.
Stripe reported that 70% of commands used to access data through its APIs come from agents. API brokerage Alpaca also noted that monthly API calls are being driven by agents, increasing from single digits in Q4 2025 to 30% in Q1 2026. Correspondingly, approximately 25% of developers are now designing APIs with agents as the primary end consumer rather than humans, and more than half of them cite unauthorized agent access as a security concern.
Mr. Yang has worked with Visa, Ramp, Mercury, Eleven Labs, Stripe, Coinbase, MoonPay, and
DoorDash is the company that launched a command line interface (CLI) (commands to retrieve data from an API) for agents.
“As more companies launch agent-native CLIs, agents will gain broader access to perform their work, driving further adoption,” Yang wrote. “Then the cycle will accelerate as companies build out more agent-first infrastructure.”
The problem is that bot detection systems can only measure traffic that clearly identifies itself as automated or trips known signatures, while agent browsers that mimic human behavior patterns routinely slip past traditional filters. A recent academic study by the University of Bamberg on bot detection systems found soft block rates of between 7% and 15% due to detection system malfunctions in real traffic, not to mention the opposite problem where agents go completely undetected. Digital marketing firm Thea Interactive has been warning clients since 2023 that agent browsers can “increase engagement, artificially lower bounce rates, and distort session duration” in ways that standard analytics tools cannot detect.
Still, Yang notes that what Pitchbook calls the “machine economy” remains small compared to the overall economy for now.
The company estimates that of the roughly $20 trillion in jobs that could go to AI agents, only about 1% actually goes through them today. Another estimate from the startup Forsea puts global “agent GDP” (the economic value directly attributable to deployed agents) at $36 billion annually on a run-rate basis.
yan said luck AI agents cannot yet fully participate in the online economy because payment and liability are not resolved.
“If you don’t have the infrastructure to properly pay agents, agents won’t buy and sell, and if agents don’t buy and sell, you can’t generate economic activity,” Yang said.
