Developer API

A developer API is a defined access point through which one program can use another program. The abbreviation API stands for “Application Programming Interface,” in other words, a programming interface. The comparison with an electrical outlet fits well: the shape of the plug is standardized, and nobody needs to know what happens behind the wall. In the same way, an API precisely defines what information you send and what response comes back. Someone building a weather app pulls measurement data via the API of a weather service. Someone wanting to build in an AI feature sends their text to the API of an AI provider and gets the generated response back.

Why hardly any app runs its own AI

Training a large language model costs many millions of euros. Even just running it requires special graphics chips that individually cost tens of thousands of euros. For most companies, this is completely unrealistic. Via an API, they only pay for the requests they actually make.

That’s why today very many products depend on a few large APIs. A writing assistant, a customer chat, or a translation feature in accounting software often access the same model. For investors, this is a central point: the revenues of providers like OpenAI, Anthropic, or Google arise to a large extent through these interfaces. This is referred to as a platform business.

This convenience has a downside. Anyone who bases their product on a third-party API is dependent on that provider’s prices and rules. If the provider raises prices or discontinues an older model, the company itself has to react. Experts call this dependency platform risk.

What actually happens during a call

Technically, a call is nothing more than a message sent over the internet. The program itself sends a small data packet to a fixed internet address of the provider. It contains which model is desired, the actual text, and some settings. The server computes and sends the answer back as a data packet. This answer is neatly structured so that the program itself can process it further.

Every request contains a key, a long secret string of characters. This is how the provider recognizes who is making the request and bills for usage accordingly. If such a key becomes public, anyone can make requests at the owner’s expense. This happens regularly when keys accidentally end up in published program code.

AI APIs are usually billed by tokens. A token is a text building block, such as a short word or a syllable. The submitted text and the generated response are counted separately, with the response often being more expensive. On top of that come limits on the number of requests per minute, so that individual users cannot overload the service.

From train schedules to school projects

APIs are invisibly embedded in almost every app that displays current data. The map in a delivery app comes from a mapping service. Payment runs through the API of a payment service provider. The departure times in a schedule app come from the interface of the transit authority. You only notice this when one of these interfaces fails and part of the app suddenly stays empty.

In the news, the term usually appears in three contexts. First, with price cuts: when the price per million tokens drops, AI becomes affordable for small companies. Second, with rule changes, such as when a platform makes its API paid. Third, with security incidents involving stolen keys.

An important distinction: the API is not the same as the chat interface of an AI service. In the chat window, a human types; via the API, a program asks. A student building a small chatbot for a computer science project uses the API with just a few lines of code. That’s exactly where its appeal lies: capabilities that you could never develop yourself become usable through a narrow, clearly defined door.