The web just quietly passed a milestone nobody threw a party for. According to Cloudflare’s traffic radar, automated traffic recently crossed the halfway mark, which means more of the internet is now machines than people. Forecasters didn’t expect that line to be crossed until 2027. It happened early.

Sit with that for a second. When you publish something today, the odds are better than even that the first thing to read it isn’t a person. It’s a bot, an agent, a crawler, a model quietly ingesting your words to answer someone else’s question later.

Two internets, growing apart

What we’re really watching is the web split into two audiences that want opposite things.

One is the internet you already know, built for humans. It leans on visuals, story, and feel. Apple’s product pages are the classic example: gorgeous, image-heavy, emotionally tuned, and almost impossible for a machine to parse cleanly. That’s fine, because those pages aren’t trying to talk to machines. They’re trying to make you want something, and want is a human problem. This layer still rules the decisions that carry weight, the car, the house, the thing you agonize over.

The other internet is being built for agents, and it could not care less about how anything looks. It wants structured data, clean specifications, and protocols it can act on without guessing. Amazon has quietly served this audience for years with data-dense product feeds an agent can compare in milliseconds. New plumbing like Anthropic’s Model Context Protocol and Google’s Agent-to-Agent protocol pushes it further, letting agents discover, negotiate, and buy on their own.

Here’s the tension in one table:

Why this matters if you publish anything

The reflex is to pick a side. Don’t. The web isn’t choosing between people and machines, and neither should you.

An agent that can’t read your site won’t recommend it. If a shopper asks their assistant to “find the best deepfake detector for a newsroom” and your comparison lives only inside a pretty layout no model can decode, you’re invisible for that query, no matter how good your work is. At the same time, strip your site down to raw data and you lose the humans who still make the calls that matter, the ones who buy on trust, not just spec sheets.

So the job now is to run both playbooks at once. Keep the experience that earns human trust, and add the scaffolding that lets machines find and act on your content. In practice that means machine-readable signals like llms.txt, structured data, and clean, accessible taxonomies, so an agent can understand what you offer and pass it along to the person who sent it.

The trust problem underneath

There’s a catch, and it’s the one we spend our days on. A machine-majority web is also a machine-generated web. The same agents reading your content are producing an ocean of it, and a growing share of what circulates was never touched by a person at all. When bots write for bots, provenance stops being a nice-to-have. It becomes the only way to tell a real signal from a synthetic one.

That’s the quiet reason to care about who, or what, is reading. An internet where machines outnumber us isn’t automatically worse. But it is one where “is this real?” gets harder to answer by eye, and easier to answer with the right tools. Building for both audiences is step one. Being able to prove what’s authentic to either of them is step two.

We’re already living in the machine-majority web. The publishers who thrive in it will be the ones who speak to people and machines fluently, and can back up everything they say.