AX Design

AX Design

AX Design means designing websites, programs, and services so that not only humans can use them well, but also autonomously operating computer programs. The term is a response to the fact that increasingly software rather than humans is clicking through pages, filling out forms, and triggering orders.

Until now, websites and apps were built for humans. One would consider where a button needs to sit, which color is easy to read, and how many steps an order requires. This kind of thinking about usability has been called user experience for decades, often abbreviated with the English acronym UX. By now, however, it’s not only humans sitting in front of screens anymore. Increasingly, programs take over the work, autonomously completing tasks: they search for a flight, compare prices, or fill out a form. AX Design is the attempt to make digital offerings usable for these programs as well.

Why software is suddenly becoming a target audience

The trigger is a new type of program called an agent. An agent receives a goal in plain language, such as “book me the cheapest train to Hamburg,” and then works independently from there. It calls up pages, reads them, clicks, and checks the result. Major providers like OpenAI, Google, and Anthropic have been building such agents into their products since 2024. This creates a second kind of visitor that nobody had planned for.

For companies, real money is at stake here. If an agent fails during a shop’s checkout process because a pop-up window appears or an image without a caption serves as a button, the order is lost. The customer only notices that “it didn’t work.” They then try a competitor whose site the agent does understand. Visibility to machines is thus becoming a topic similar to what visibility in search engines has been for the past twenty years.

There is also the opposite position. Some operators specifically do not want agents on their sites, because their business depends on advertising that a machine doesn’t look at. That, too, is a decision in the spirit of AX Design, just with the opposite sign. So the question isn’t whether one serves agents, but how deliberately one does so.

Clear interfaces instead of beautiful surfaces

The first and most important building block is an interface, that is, a technical access point through which a program can query data directly. An agent then doesn’t have to laboriously take apart a screen page, but instead receives prices and availability in neatly sorted form. This is faster, cheaper, and considerably less prone to errors. Since 2024, the Model Context Protocol, abbreviated MCP, has spread as the shared standard for this purpose. It specifies how a service offers its functions to an AI program.

The second building block concerns the ordinary website, since not every provider has an interface. Here, clean structure helps: unambiguous labels for buttons, sensible headings, machine-readable additional information about products and prices. Much of this overlaps with accessibility, that is, design for people with disabilities. A screen reader for blind users and an agent fail at exactly the same points.

In addition, there are rules that build trust. The operator wants to know which agent is acting on behalf of which person and what that agent is allowed to do. This is why procedures for registering and authorizing agents are currently emerging. A common misconception is that AX Design is just a new word for automation. The difference: automation follows a fixed script, whereas an agent makes decisions on its own along the way.

How to recognize AX Design online

In everyday life, this topic mostly appears invisibly. When an AI assistant reliably reserves a table or tracks a package, there is often a deliberately built interface behind it. If it breaks off mid-process instead, that’s rarely due to the language model itself. Usually, the target page is simply not comprehensible to machines.

In business news, AX Design comes up in connection with online retail, travel portals, and banks. There, discussions revolve around whether agents will trigger a relevant share of orders in the future and who retains the customer relationship in that case. Disputes are part of this too: some publishers and shops block agents or demand payment for access. Anyone who knows the term understands these reports as a single question — who is the web actually still being built for.

Subscribe free. Unsubscribe the second it sucks.

High-signal news across AI, business, UX, and tech. Every morning.