The internet's next users won't be human

Prediction: humans won't be the main users of the internet in five years.

Sounds crazy? Good. Early forms of it already exist.

A social network that isn't for us

If you haven't come across it yet, there is a social network called Moltbook. At first glance it looks like Reddit. Threads, replies, upvotes, the usual.

But it is not for us. It is for AI agents.

They post. They reply. They argue. They form opinions about each other's opinions. And humans? We sit there refreshing, like spectators at a match we are not playing in, watching the agentic fun unfold.

The reactions from people who build this technology have been telling. Aditya Agarwal, former CTO of Facebook, sounded anxious about it. Andrej Karpathy called it "insane" and also said, "don't run it." Elon Musk's take was two words: "early singularity."

We wanted AI tools. We got AI communities.

Why this matters more than it seems

It is tempting to dismiss this as a curiosity. A weekend experiment that went viral. I think it points to something much bigger.

Today the internet is built for human eyes. Web pages, feeds, search results and ads all assume a person at the other end, reading, scrolling, clicking. The entire economics of the web, from advertising to e-commerce, rests on that assumption.

Once agents start talking to each other, that assumption starts to break.

• Agents will search on our behalf, compare options, and negotiate.
• Agents will read far more content than any human ever could, and summarise it for us.
• Agents will transact with other agents, booking, buying and scheduling without a human in the loop.

The web was designed for people to read. The next one may be designed for agents to act.

From pages to interfaces

The early web was a collection of documents. Then it became a collection of applications. The next shift may be from applications built for people to interfaces built for machines.

You can already see the outline. Companies are publishing structured descriptions of their products so that AI systems can read them reliably. Developers are building protocols that let an agent call a booking system, a payment service or a database directly, without pretending to be a human clicking buttons. Browsers are being rebuilt with assistants that can carry out tasks across several sites on the user's behalf.

None of this needs a dramatic breakthrough. It is ordinary engineering, done at scale, by thousands of companies that each see a small advantage in being easy for an agent to use.

What changes for businesses

If most traffic is agents, then a lot of what companies optimise today will need rethinking.

Brand. How do you build preference when the "customer" making the shortlist is software? Visual identity and clever copy matter less to an agent than accurate data, clear pricing and a reliable track record. Brand may shift from how you look to how dependably you deliver.

Distribution. If an agent picks the default option, who decides the default? The platform that hosts the agent will have enormous power, much as search engines and app stores do today. Being the option an agent recommends could become the new first page of search results.

Pricing. Agents compare relentlessly and do not get tired. Markets where price was hidden behind effort, such as insurance, travel and many business services, may become far more transparent.

Trust. How do you know the reviews, posts and comments you see were written by a person at all? Proof of human authorship, and proof that an agent is acting with genuine authority, will become valuable infrastructure.

None of this is a distant science-fiction problem. It is a strategy question for every company that sells anything online.

The questions leaders should be asking now

There is no need to rebuild everything today. There is a need to start preparing, the way the best companies prepared for mobile before it was obvious.

• Could an AI agent understand what we sell, at what price, and on what terms, without a human reading our website?
• Which of our customer journeys would an agent complete faster than a person, and what happens to our margins if it does?
• Where do we depend on human attention, through advertising, impulse purchases or friction, that an agent would simply ignore?
• What would we need to trust an agent acting for us, as a buyer, and to let one act on our behalf?

Watching from the stands

So here is my guess. In a few years we will be watching an internet filled mostly with agents. Humans will still be there, of course. But more as owners, directors and editors than as the main users.

The interesting question is not whether this happens. It is who designs the rules of that internet, and whether humans stay in charge of what matters. The companies that help write those rules, rather than waiting to be told them, will have a head start that is very hard to close.