Interface

An interface is the point where two systems meet and hand over data. It specifies exactly what may be requested and in what form the response arrives. The inner workings of the other system remain hidden. An everyday example: the power outlet is an interface to the electrical grid. You don’t need to know whether the electricity comes from wind or coal. You just need the right plug, and that’s shaped the same everywhere. In software, this kind of hand-off point is usually called an API, short for Application Programming Interface, or in German 'Programmierschnittstelle'.

Why software today is built from foreign parts

Hardly any program is built entirely from scratch anymore. A weather app doesn’t measure temperatures itself, it queries them from the weather service. An online shop doesn’t verify credit cards itself, a payment provider does that. Every one of these requests runs through an interface. Developers save months of work as a result.

For companies, interfaces are also a business model. Whoever offers an in-demand capability sells it per request. Providers of AI models do exactly that: access to the model happens via an interface, billed by the amount of text. This is one of the industry’s most important revenue channels.

The downside is dependency. Whoever builds on someone else’s interface is at the mercy of its rules. If the provider changes the prices, your own costs rise. If it shuts down the interface, your own product grinds to a halt. In the tech industry there are numerous companies that lost their foundation overnight in this exact way.

Request, response, and the contract behind them

Technically, an interface on the web is usually an internet address to which a program sends a request. The request contains the desired operation and the necessary details, such as a place name. What comes back isn’t a colorful web page, but a plain data structure. Another program can process it directly.

This almost always includes a key, a kind of personal password for programs. This lets the provider recognize who is making the request and bill accordingly. It also limits how many requests are allowed per minute. This limit is called a rate limit, and it prevents a single user from overloading the service.

The documentation is important: the written description of all permitted requests. It functions like a contract between provider and user. As long as the provider adheres to it, all programs keep working. That’s why changes to interfaces are so delicate that old versions often keep running in parallel for years.

From banking apps to AI assistants

You use interfaces constantly without seeing them. The map snippet in a restaurant app comes from a mapping service. Logging in with your Google account on a third-party site runs through an interface. Third-party banking apps also access account data this way, regulated in Europe by law through the Payment Services Directive.

In business news, interfaces usually show up amid disputes. When Twitter drastically raised the price of its interface, many add-on apps and research projects disappeared. Such reports show how much power lies in access to data.

With AI systems, a new direction is emerging. Language models now call interfaces themselves, for instance to look up a current exchange rate or schedule an appointment. This is called tool use. A program that merely generates text thereby becomes one that actually acts on the web. This is precisely what the industry is currently working on especially intensively.