API interface

Programs constantly need to do things they can’t do themselves. A weather app knows nothing about the weather, an online shop nothing about credit cards. So they ask other programs. For this to work reliably, clear rules are needed: How do you pose the question, and in what form does the answer come back? These exact rules are called an API interface. The abbreviation API stands for the English term “Application Programming Interface,” i.e., programming interface.

Why hardly any app gets by without foreign APIs

Without interfaces, every company would have to build everything itself. Anyone developing a delivery app would have to create their own map, program their own payment system, and run their own servers for messages. That would take years. With ready-made interfaces, it takes weeks. The map comes from a map provider, the payment from a payment service provider, both via just a few lines of code.

For the providers themselves, interfaces are a business model. Many AI companies don’t sell their models as a program to download, but only as access via an API. Payment is made per request or per amount of text processed. A large share of AI companies' revenue is generated this way.

This also explains why changes to an API end up in the business news. If a provider raises prices, restricts access, or shuts down an old version, all the companies that built on it run into trouble. This dependence on a single provider is a genuine business risk.

Request, response, and the key in between

You can think of an API like placing an order at a counter. You say exactly what you want, and you get exactly that back. What happens behind the counter you don’t see and don’t need to understand. This separation is intentional: the provider can completely overhaul its technology as long as the counter stays the same.

Technically, a program usually sends a request over the internet to a specific address. The response comes back as structured text, often in the JSON format. This is a notation consisting of field names and values, such as “temperature: 21”. Humans can read it, machines can process it reliably.

This almost always includes an API key, a secret password for the requesting program. This lets the provider identify who is asking and bill accordingly. Often there’s also a limit, a so-called rate limit: for example, a maximum of 60 requests per minute. Anyone who sends more is temporarily blocked. A common misconception, by the way, is that an API is a kind of website for humans. It is intended exclusively for programs and has no user interface.

APIs in chatbots, mobile apps, and company reports

You use dozens of APIs every day without noticing. When you log into a game with your Google account, that runs through an interface. When a timetable app shows a delay, it queries an interface from the railway. Even the exchange rate in a finance app usually comes from a data provider.

In the field of AI, the term is especially prominent. Chatbots like ChatGPT have an interface for humans as well as an API for developers. Anyone who wants to build their own software with a language model uses the API. A language model, in this context, is an AI system that completes and answers text. AI agents too, meaning programs that carry out tasks independently, access calendars, search engines, or databases via APIs.

That’s why APIs show up as a metric in news about companies. Companies report how many developers use their interface or how strongly their API business is growing. When you read such reports, you now know: it’s about software that sells other software as a building block.