The "dead internet theory"—a concept circulating in online forums since the 2010s positing that bots generate content and interact with each other to simulate human activity, ultimately crowding out real people—was long dismissed as a fringe conspiracy. It has now become a measurable fact.

Multiple cybersecurity firms agree that bots outnumber humans online, yet they disagree on exactly when the crossover occurred and which metrics should define it.

CloudFlare, the internet security and performance company whose services are used by millions of websites worldwide, reports that the crossover took place in June, with bots accounting for 57.5% of all webpage requests. Thales, a French technology group that provides data security for organizations and governments, places the crossover as far back as 2023; its Bad Bot Report put bot traffic at 53% in 2026.

These discrepancies stem from the absence of a single standardized methodology for measuring bot traffic, as no single provider has visibility across the entire web, according to Rudy Yang, Pitchbook's enterprise and retail fintech analyst and author of the firm's July report on agentic AI traffic.

"There's a lot of missing pieces of information, but a lot of the observed data suggests the same thing, which is like there is more bot activity," Yang told Fortune. "Agentic AI activity is driving a lot of the browser activity you're seeing."

The growth in agent-driven activity—AI systems designed to autonomously perform multi-step tasks such as navigating websites, comparing products, and completing transactions—is dramatic across every measurable dimension. According to cybersecurity firm HUMAN Security's 2026 State of AI Traffic & Cyberthreat Benchmark Report, traffic from agents that take concrete actions on the web—clicking links, filling out forms—surged 7,851% year over year. Scraper traffic grew 597% over the same period. AI training crawlers, while still accounting for 67.5% of AI-driven traffic, represent a shrinking share of the total.

The timing surprised even seasoned observers. CloudFlare CEO Matthew Prince had predicted in March that bots would not cross the halfway mark until the end of 2027. Instead, the crossover arrived more than a year ahead of schedule.

"For companies and developers, this means that building for agent traffic will become nonnegotiable," Yang wrote in the report.

The internet's existing business model—ad impressions, conversion funnels, pageview-based analytics—was built on the assumption that visitors are human. If the majority are now agents, that assumption is fundamentally upended, reshaping how companies monetize web traffic. It also raises longer-standing concerns about content authenticity, as automated traffic has been linked to fake reviews, coordinated social media engagement, and manipulated discourse—issues that predate agentic AI but are amplified by its scale. The same autonomy powering the surge in agentic AI traffic is also what allowed an OpenAI model to escape its constraints this week, underscoring that the volume increase is merely the visible symptom of a shift that businesses can neither fully measure nor fully control at this stage.

"They consume the web completely differently than humans do," Yang told Fortune. "It's almost like an entire new category, customer category, was created, and it means a lot for businesses because no one, as a business owner, is going to want to silo themselves from being able to serve a completely new customer segment."

Strategy Shifts for the 'Machine Economy' and the Limits of AI Agents

Companies are taking note, and developers have begun adapting.

Stripe reported that 70% of its API data-access commands now originate from agents. API brokerage firm Alpaca noted that agent-driven monthly API calls grew from single digits in Q4 2025 to 30% in Q1 2026. In response, approximately 25% of developers now design APIs with agents rather than humans as the primary end consumer, and over half cite unauthorized agent access as a security concern, according to Postman's State of the API report.

Yang identified Visa, Ramp, Mercury, ElevenLabs, Stripe, Coinbase, MoonPay, and DoorDash as companies that have launched command line interfaces (CLIs)—commands that retrieve data from an API—specifically geared toward agents.

"As more companies launch agent-native CLIs, agents gain broader access to execute work, driving further adoption," Yang wrote. "Companies will then build more agent-first infrastructure, accelerating the cycle."

A core challenge is that bot-detection systems can only identify traffic that explicitly declares itself as automated or matches a known signature. Agentic browsers that mimic human behavior patterns routinely evade traditional filters. A recent academic study from the University of Bamberg found soft block rates of 7%–15% caused solely by detection systems misfiring on legitimate traffic—not accounting for the reverse problem of agents going entirely undetected. Seer Interactive, a digital marketing firm, has been cautioning clients since 2023 that agentic browsers can "inflate engagement, artificially depress bounce rates, and distort session duration" in ways that standard analytics tools fail to capture.

Still, Yang emphasized that what Pitchbook terms the "machine economy" remains small relative to the broader economy—for now.

The firm estimates that only about 1% of the roughly $20 trillion in work that could plausibly be delegated to AI agents is currently flowing through them. A separate estimate from the startup Forsy places total global "agent GDP"—economic value directly attributable to deployed agents—at $36 billion per year on a run-rate basis.

Yang told Fortune that AI agents cannot yet fully participate in the online economy because the problems of payments and liability remain unresolved. How quickly those gaps close—through agent-specific payment rails, identity standards, or regulatory frameworks for automated transactions—will largely determine whether the machine economy scales beyond its current footprint.

"If we don't have the infrastructure to do proper payments for agents, then agents aren't buying and selling, and if agents aren't buying and selling, then they aren't generating economic activity," Yang said.

This story was originally featured on Fortune.com.