Introduction
I have been thinking about the next generation of the web for years.
For a long time I assumed it would be an app. One place that combined everything, the tool that finally pulled the whole mess together. Part of me also expected something more cinematic, a game like universe, a VR world we would all end up living inside. We still might get there one day.
But I did not expect the AI era to arrive this fast, or to go mainstream this quickly.
What we have right now is not a metaverse. It is AI chatbots and agents. And that should be our priority, because this is not the next fad. It is real. For the first time you are not lost in the web on your own. You have an assistant that can guide you through it.
That is the opportunity. But there is a catch.
1. We cannot repeat the mistakes we made with the web
Let me be honest about the web we built.
It became a cluster. A sprawling, inconsistent, over engineered mess that, for all its scale, never quite led us anywhere. The majority of users do not even really know what the web is or how it works. Only a small group, maybe one percent of us, ever truly understood the machine underneath.
I do not want to watch that happen again with AI.
Right now AI agents are trying to use a web that was designed for human eyes. They scrape HTML. They guess at forms. They render entire pages just to find one price or one button. It works, barely, and it is already recreating the same fragmentation and waste that made the web so heavy in the first place.
If every AI company, every website and every agent invents its own way of doing this, we will end up with ten thousand incompatible standards and a decade spent catching up. Exactly like the search consoles, tracking pixels and plugin ecosystems before it.
2. The answer is simplicity
Everything I keep coming back to is one word. Simplicity.
The promise of AI is not that it can generate text. It is that it can take chaos and turn it into focus. It removes the noise and leaves you with the thing you actually wanted.
So the layer we build underneath it has to be simple too.
That means a few hard rules, and I think the whole web, and the whole community of AI and agent developers, need to follow them:
- Do not build ten thousand packages for every small thing.
- Write less code, and reuse as much as possible.
- Keep internet traffic low.
- Keep the load on data warehouses and AI farms low.
That last point matters more than people admit. AI already consumes an enormous amount of compute and energy. Making every model scrape and render a full web page, over and over, is one of the most wasteful things we could design. A shared, structured layer sends a few kilobytes of meaning instead of megabytes of markup. Less parsing, less bandwidth, less energy. It is the green argument, applied to AI.
3. So I built AI2Web
AI2Web is my attempt to give the web one simple, shared way to be understood by AI.
The idea fits in a single sentence: describe your website once, and AI2Web makes it understandable to every AI.
You publish one capability manifest that describes what your site can do. Its content, its products, its actions, its support flows. From that one description, AI2Web speaks whichever protocol an assistant understands: MCP, ACP, REST, GraphQL, OpenAPI, feeds and more. You do not rebuild for each assistant. You describe once, and it works across all of them.
It is built on a few principles that come straight from the frustration above:
- One capability model, not a portal per vendor.
- Backend and API driven, so an endpoint works in Claude and ChatGPT today.
- Safe by design. Discovery is open, but anything sensitive, a payment or a refund, always asks the user to approve first.
- Reuse over reinvention. It complements existing protocols instead of competing with them, and if a neutral standard wins, AI2Web is carried by it rather than fighting it.
There is a validator that scores how AI ready your site is out of 100, a discovery network so agents can find capable sites, and reference implementations in TypeScript, JavaScript, React, PHP, Python, Go and .NET, plus a WordPress plugin.
4. It is open, and it needs the whole web
I could have kept this as a product. I did not, because a standard only works if it belongs to everyone.
The protocol, the specification and the SDKs are open source. The whole thing lives in the open so the community can shape it, challenge it and adopt it. That is the only way this avoids becoming yet another walled garden.
If you build websites, AI agents, or the tools in between, I would genuinely love your eyes on it.
- The site and validator: ai2web.dev
- The code, spec and SDKs: github.com/ai2web-foundation
Closing
We have a rare chance here. A moment before the habits set, before the fragmentation locks in, to build the AI layer of the web the way the web itself should have been built.
Simple. Shared. Light. Focused.
We went from being lost in the web to having something that can actually guide us through it. Let us not bury that under another ten thousand packages.
Describe once. Works everywhere. That is the whole idea.