On 12 November 2026, OpenAI published a customer story about Holiday Extras and Codex. The story says Holiday Extras is using Codex to turn customer expertise into better apps, higher booking conversion and new possibilities for growth.

I couldn’t load the full article, so I’m working from OpenAI’s summary of it. I won’t quote figures or details I haven’t seen.

What the story claims

The headline reads as a tool story. A company adopts a coding agent, its apps improve, and more people book.

That reading is the easy one. The summary itself says Codex is turning customer expertise into better apps. Expertise is the input, and the tool is the thing that carries it into software. I’d put the credit for the conversion gain there.

What Codex changed

Is there a real need? Yes. Any operator with an app or website has a list of improvements they know would help customers and never get built. The list is long because building is slow and expensive.

Codex shortens the distance between an idea and a working app. That is a real change, and I use tools like it myself. But it is the only distance it shortens.

What Codex didn’t change

Codex doesn’t know why a customer abandons a booking. It doesn’t know which question people ask just before they buy, or which worry stops them. It can’t test an app with real customers or collect their feedback. It doesn’t bring in the customers either.

Those things decide whether a faster build turns into more bookings. A team that already understands its customers has them. A team that doesn’t can now ship a wrong guess in a day instead of a month.

So when conversion goes up after a tool is adopted, I ask what the team put into the tool. The tool gives every team the same speed. The context is the part that differs, and the context is what decides the result.

What this looks like in a normal business

Take a clinic that wants an app for booking appointments. The front desk staff know things no developer or AI does. They know which treatments patients ask about by name and which ones they confuse. They know which question comes up right before someone books, and which one comes up right before someone drops off.

If the owner writes none of that down, the builder gets a one-line prompt: “build a booking app for a clinic.” The result will be a generic booking app. It will work and it will look fine. It won’t convert any better than the generic apps already on the market.

If the owner writes it all down, the builder gets a brief with real content. It lists the questions patients ask in the order they ask them, the wording they use, and the points where they hesitate. The same tool then produces something fitted to those patients.

The same applies to a shop, a tuition centre or a service business. The staff who answer customers every day hold the knowledge that makes an app good. Most of it is in their heads or scattered across old chats.

Where this meets my own work

This is how I work with AI in general. I direct, it produces, I check and decide. Then I fix the brief. The quality of what comes out follows the quality of what goes in.

It’s also why AeroChat asks the operator to describe their workflows in their own words. The business’s knowledge of its customers has to get into the system somehow, and writing it down is the first step.

What a business should do

  1. Write down what your staff know about customers. Ask the people who answer them: the top questions, the objections, the moments someone gives up, the words customers use. Pull from your chat and email history too.
  2. Brief the builder with that material. Give Codex or any similar tool the real content, not a one-line request.
  3. Pick one outcome before you build. For an app that takes bookings, it is bookings. Count them before and after the change.
  4. Check the result against real customers and fix the brief. When the number doesn’t move, the brief is usually what needs changing.

Source: OpenAI News: Holiday Extras is rethinking what it can build with Codex