The Invisible Economy

A small experiment

Open ChatGPT. Type: "Book me a table for two at an Italian place near me tomorrow at 8pm."

It does it. Quietly, in the background, it finds a restaurant, checks availability, confirms the reservation. You get a name, an address, a time. Done.

Now try this: "Book me a haircut at any salon near me tomorrow morning."

It can't.

Not because the AI is dumb. Not because the salon doesn't exist. The AI knows the salon is there. It can see its Instagram. It can read its Google reviews. It just can't do anything with it.

This is not a small gap. This is the shape of what's coming.

The web you can talk to, and the web you can't

Two internets are forming, and most people haven't noticed.

The first internet is the one ChatGPT, Claude, Gemini, and Perplexity can actually use. It includes restaurants — through OpenTable, Resy, Tock. It includes hotels — through Booking.com and Marriott's new agent integrations. It includes Uber, DoorDash, Amazon, concert tickets, flights.

When you ask your AI to do something in that internet, it does it.

The second internet is everywhere else. Your hairdresser. Your dentist. The padel club. The yoga studio. The driving school where your teenager is learning. The little shop that sells the only coffee you actually like.

These places exist. They have websites — sometimes. They have Instagram. They have Google Maps pins. But ask your AI to book any of them and you'll get the same answer your grandmother would give: call them.

There are around 230 million small businesses like these in the world. Soon, the AI in your phone is going to be your front door to everything. And right now, almost none of them are on the other side of it.

Why this should feel familiar

In 1998, the businesses that built websites captured the people shopping online. The ones that didn't, lost them. Not overnight. Slowly. And then suddenly.

In 2010, the businesses that ended up on Google Maps captured the people searching with phones. The ones that didn't, became invisible to anyone under thirty looking for something nearby.

This is the same pattern, again. Faster this time.

A restaurant that can't be booked through OpenTable is invisible to whoever is planning dinner with ChatGPT. A salon that can't be booked through AI is invisible to whoever is planning their week with Claude. A shop that can't be browsed by an assistant is invisible to whoever just doesn't have time to open seven apps to find the thing they want.

If you can't be operated by AI, you don't disappear. You're just never offered.

What it actually means to be "AI-ready"

Some companies say they're "AI-powered" or "AI-ready." Most of them mean they put a chatbot on their website. That's not what we're talking about.

What we mean is much simpler, and much harder.

A business that AI agents can really work with has four things:

It can be found. Not just by Google. By any AI in the world that's looking for a haircut in your town tonight.

It can be read. Not last month's homepage. Today, right now: who's available, what costs what, what's actually on the menu.

It can be operated. Not by a human filling out a form. By an AI that says "reserve this slot for Joan, here's his email," and gets a confirmation back.

It remembers you. Not by asking you to make a new account at every place. One identity, every business in the network, no twentieth credit card form.

Almost no small business has any of these four things. And almost none of them are going to build it themselves. Not because they don't want to. Because nobody can be expected to. The standards are still forming. The protocols are alphabet soup. The work is technical and constant.

So it has to come from somewhere else. It has to come from infrastructure.

Where we come in

We've spent the last few years quietly running the day-to-day software of about five hundred small businesses. Hairdressers. Padel clubs. Driving schools. Clinics. Sports clubs. Local shops. Eleven city halls.

We are their calendars. Their payment system. Their customer database. Their booking pages. Their event tickets. Their text-message reminders. The boring software that runs the things they don't have time to run themselves.

So we already have the thing AI agents need. We know which slots are open at the salon tomorrow. We know which courts are free at the club this weekend. We know what's in stock at the shop. We have the live, real, accurate picture of hundreds of businesses, in real time.

What's missing is a door. A door any AI in the world can walk through to ask these businesses questions and book things on behalf of real people.

We're building that door.

And because we're building it once, in the middle of the platform, it opens for all five hundred businesses on the day we ship it. Then for the next five thousand. Then for the next fifty thousand. None of them have to do anything. They just keep running their business. The door opens itself.

That's the entire idea.

The piece that surprises people

When we explain this, the question we get most is: "OK but which AI are you betting on? ChatGPT? Claude? Google? Some new one nobody knows yet?"

The honest answer is: all of them. And whichever comes next.

There are several emerging standards for how AI agents talk to websites — alphabet soup names like MCP, WebMCP, agent.json, Schema.org. Each one is championed by a different combination of Google, Microsoft, Anthropic, OpenAI, and open-source communities. Nobody knows which will win. Maybe none of them. Maybe all of them, badly.

We don't have to pick. We implement every one of them, on every business, automatically. Whichever AI shows up — today's ChatGPT, tomorrow's whatever — at least one of the standards we already speak will be the one it knows. The door opens regardless.

This isn't us being clever. It's just the only sensible position. Startups don't write global protocols — the W3C and Google and Anthropic do, over years. What startups can do is be the layer of universal translation between thousands of businesses and whoever ends up speaking which language. So that's what we are.

Why now, and not in two years

People sometimes ask whether this is too early. It is not.

Google AI Mode launched agentic restaurant booking in January 2026. It's now in nine countries. ChatGPT Agent browses websites by itself and completes purchases. WebMCP became a web standard in February. OpenClaw — a free, open-source AI agent on WhatsApp — passed a hundred thousand users in a few weeks. McKinsey estimates that by 2030, AI agents will be moving three to five trillion dollars a year through the economy on behalf of humans.

None of this is hypothetical. None of it requires a leap of faith. It is happening now, in production, with real money.

What is uncertain is which businesses will be on the other side of all that activity. Right now: restaurants, hotels, big retailers. The long tail of everyday service businesses — your hairdresser, your dentist, your gym — is wide open. Whoever builds infrastructure for it now, before the question is decided, owns the layer for decades. The same way AWS owns cloud, Stripe owns online payments, Shopify owns small ecommerce.

The window for that is short. Maybe eighteen months. After that, the answer will already be set, and everyone else builds on top of whoever got there first.

What we believe

We don't believe AI will get magically smarter and solve this on its own. We don't believe one big AI will win and consolidate everything. We don't believe small businesses will quietly learn to write technical specs in their spare time.

We believe something simpler.

That sooner than feels comfortable, your phone is going to be your front door to everything.

When you want a haircut on Saturday, you'll tell your phone, and your phone will handle it. When you want a padel court at six, same thing. When you want to enroll your kid in driving school, when you want to buy something local, when you want to find a therapist — same thing.

The businesses that are on the other side of that door will keep their customers. The ones that aren't will lose them slowly, then suddenly. And the difference between the two groups won't be how good they are. It won't be how much they care. It won't be the quality of their service.

It will be whether someone, somewhere, built them the door.

We have five hundred of them. We know how to build the door. And there are two hundred and thirty million more waiting.