Last winter the fastest-growing social network almost nobody joined was Moltbook. Within weeks it had around 1.5 million accounts posting, replying, and upvoting each other around the clock. Almost none of them were people. Moltbook was built for AI agents to talk to other AI agents, and the humans were there to watch.
That 1.5 million is worth pausing on. It is a registration count, not a count of people. One person can spin up hundreds of agents before breakfast, and by the closest reckonings the overwhelming majority of those accounts were automated, with only a few thousand humans behind the glass. Keep that pairing in mind. We will need it later.
Moltbook is a novelty. The web was built on one quiet assumption: that the reader is a person. That assumption has broken, and breaking it breaks two things at once: how the web makes money, and how it knows who it is dealing with. This series is about those two problems. The web's defining change this decade is demographic. The audience it was built for is no longer mostly human.
The audience changed species
Imperva, which sits in front of a large slice of the world's web traffic, reported that automated traffic reached 51% of all requests in 2024, the first time in a decade that machines outweighed people on the internet. A year and a half later Cloudflare, measuring its own enormous network, said the same thing had happened across the wider web: bots had passed human traffic for the first time, somewhere near 57% of requests by early 2026. Cloudflare's own chief executive noted he had not expected that line to be crossed until 2027. It came early.
The web is majority machine, and the share is still climbing.
The asymmetry is not only in headcount but in appetite. A person shopping for a camera might visit five sites before buying. An agent running the same errand visits a magnitude more, because software doesn't share a person's limits of attention and patience. One human intent, handed to an agent, becomes thousands of requests against dozens of sites, none of which the person will ever see.
Every traffic number in this series comes from a company that also sells traffic security or delivery, and each measures its own slice of the internet rather than the whole thing. Treat them as directional, not as a census. The direction, at least, is not in dispute. The disagreements are about how far past the halfway mark we are, not whether we crossed it.
The web was built for a reader who is gone
Almost every way the web pays for itself assumes a human on the other end, and each assumption fails quietly when the reader is software.
Advertising assumes eyes: a display ad is a bet that a person will see it and maybe click, but an agent fetching a page to answer a question renders no ad and clicks nothing. Subscriptions and logins assume someone willing to create an account and come back, and an agent signs up for nothing and remembers nothing between calls. Search optimization assumes a crawler that indexes your page and sends a human back to it, but the newer crawlers index the page and keep the reader, answering the question themselves. CAPTCHAs, the web's oldest machine filter, assume that telling humans and bots apart is possible at all, and that assumption is the one this series will spend its second half taking apart, because it has failed more completely than the rest.
None of this means the web is doomed. It means the machinery that funded it was built for a visitor who is now in the minority, and the replacements are being improvised in real time. The rest of this series is largely about those replacements: what they are, whether they work, and who they leave out.
Two problems wearing one face
When the reader stops being human, two questions open up.
The first is economic. If the visitor sees no ads, subscribes to nothing, and is often an answer engine that never sends a human back, the ad-and-subscription model that paid for two decades of the open web stops working. Someone still has to be paid, and nobody has agreed how. That is the subject of the first half of this series: the collapse of the old model, the payment rail the web reserved in 1997 and never used, and the scramble to price content and commerce for buyers made of software.
The second is about trust. If anyone on the other end might be a machine, every mechanism that assumed a human (a login, a review, a vote, a "prove you're not a robot" checkbox) becomes unreliable. You can no longer take for granted that there is a person there at all, let alone which person. That is the second half: the failure of the web's human filters, and the emerging attempts to prove a real person is present and to hold someone accountable for the agents acting on their behalf.
To pay the right party, you have to know who the party is. To let an agent spend money, you increasingly have to prove a human stands behind it. The same infrastructure that lets a machine pay is now being wired to prove a person authorized it. The series ends where those two threads join, because that is where the machine web actually starts to work.
The easy version of this story is that some new technology fixes everything. It doesn't. Some of the fixes ahead are boring and load-bearing. Others are elegant and smart.
The next part starts with the money. If the reader is a machine that sees no ads and pays for no subscription, the first question is unavoidable: who pays?
This is Part 1 of The Machine Web, a series on what changes when the web's main character stops being human.
AKENA is a blockchain engineering studio. We build the infrastructure AI agents run on: agent-facing RPC and MCP endpoints, on-chain data pipelines, and the products on top of them. If you're building for a web whose readers are increasingly machines, we should talk.