Cloudflare's 2026 Annual Founders' Letter was published on September 28, with founder Matthew Prince and co-founder Michelle Zatlyn using the company's 16th birthday to release network data measured first-hand: the moment when automated traffic would overtake human traffic, originally forecast for the second half of 2027, already arrived in May 2026, pulled forward by AI agents and AI crawlers. If current trends continue, automated traffic will be 1,000 times human traffic in five years.
The number matters for more than arriving "over a year early." According to Cloudflare, human traffic isn't declining — what's exploding is agent traffic. Ask an agent to find you a lunch spot and it may survey 1,000 restaurant menus in the area within minutes, only to recommend one. The one restaurant gets your business; the other 999 shoulder the server costs of being crawled over and over. Who benefits and who pays are completely disconnected — the letter calls this the internet's "tragedy of the commons."
Automated Traffic Overtook Human Traffic a Year Ahead of Schedule
At the end of 2025, Cloudflare's internal estimate still placed the crossover in the second half of 2027. Then the spread of AI agents steepened the curve: by May 2026, the crossover had already happened. Another counterintuitive fact from the letter: from 2012 through 2025, the web itself had largely plateaued or even shrunk; the real explosion came in mid-2025, when masses of people with no coding skills started launching websites with "vibe coding" tools — most of which use Cloudflare's developer platform as their preferred deployment target. More than 7 million developers are now building on Cloudflare.
Agents crawl the web the way search engines always have: everything, over and over, whether or not anything changed. Cloudflare's data shows that more than half of what well-behaved bots fetch hasn't changed since their last visit. In other words, a huge share of that traffic is pure repeated work. For site operators, bandwidth and compute bills are climbing without a matching rise in revenue.
The Agents' "Tragedy of the Commons": Who Pays the Bill
Last year's founders' letter was about AI's impact on publishers: content taken for training, traffic siphoned away. This year, the same logic extends to much smaller businesses. The letter uses the corner deli as an example: small shops win customers through emotional ties — the clerk who remembers your name — but agents don't care who remembers your name. They recommend whichever place has the most information, usually the oldest, most established player. New entrants get harder to see, and the market consolidates around incumbents.
This is the flip side of the same wave behind the recent string of agent incidents — the review of OpenAI agents' internet behavior documented agents accessing external websites at scale with no human watching. Once agents running errands for humans become the internet's dominant traffic, the rules have to be rewritten.
The Cloud Playbook: Crawl Less, Get Paid for Use
The letter doesn't stop at warnings. Cloudflare says it's doing two things: technically, making crawlers "see more while fetching less" — only pulling content that actually changed, cutting the load of repeated crawls; and commercially, giving content creators a way to get paid when agents use their sites' content. This extends its earlier "Pay Per Crawl" idea and is evolving toward pay-per-use.
Cloudflare will also announce a batch of partnerships this week, aiming at a future of "not five AI companies, but 500,000" — one where newcomers, creators, and small businesses still have room in the agent era. For site operators and developers, the signal is concrete: the agent-dominated internet is already here, and server costs, bot defenses, and content monetization all need to be redesigned for visitors that are mostly agents.
