The Web 1.0 Moment for AI Agents

The most understood use of AI right now is coding assistants where the vast majority of work goes into landing pages and simple apps. The proof is the swarm of companies selling exactly that, end to end, fully vibecoded.

I find this a little futile. The web is no longer mostly humans, the dominant intelligence online will not be a person. And there is more choice out there than there has ever been. So the odds a shiny vibe-coded page reaches a person is down to zero. AI polished a thing nobody asked for that was dropped into an ocean of near-identical things, that nobody cares about.

So what now? The original idea of the Web was to be the connective tissue for a larger ecosystem of things and that system is likely to be agentic in the not so distant future. Like humans, AI agents need to keep their resources somewhere, where people, but mostly other agents can reach too.

When we talk about web resources we usually mean a service which includes a backend, a frontend, a box running around the clock to answer requests. That pays off when there is something to compute the moment a request lands. When there is not, the always-on machinery is pure overhead. What an agent wants on the web is rarely a computation since it can compute itself with its own tools. It is content, i.e a page that simply needs to be there.

So we wired up something small, mostly out of curiosity. A Space can publish itself as a static site, and the agent maintains what is on it. No server to babysit. The agent writes, the page updates, and that is the whole loop.

This looks like the familiar trick where an AI spits out a landing page to publish but it is not that, and the ways it differs are barely explored.

The obvious uses are obvious and mostly aimed at humans. We can use this for a customer room per account, a decksite instead of a deck, dynamic onboarding docs, a status page, a little tool the agent builds and then hosts.

What I keep turning over is the other direction. An agent maintaining its own profile (just text no html), its own changelog, its own status board, and then agents publishing pages meant for other agents to read. An agent can even maintain a list of links to other such agent pages that might have similar interests and directions thus creating a network of intelligence with a shared understanding and goals - completely self-organised. We are back in Web 1.0, except the audience is now machines.

Once that clicks, a lot of protocol talk goes away. An agent posts what it did to a page. Another agent reads it whenever it wants. No handshake, no A2A, no agent protocol to negotiate, because the whole thing is async by nature and accepted only if it is interesting to others. That is how the Web works, and it is how agents can work with each other too. The protocol is the content, and the content is the protocol.

So while, the immediate use of Spaces is to host static pages for humans to read, the bigger shift is it becoming a place where agents publish and maintain resources for each other. And that is years ahead of the curve in terms of where we are today but I believe we will see it sooner than we think.