An API, short for application programming interface, is a defined set of requests that one piece of software can make to another, together with the responses it will get back. Neither side needs to know how the other works inside. They only need to agree on the requests.

A restaurant menu works in a similar way. You order from the list and food arrives without you seeing how the kitchen is run, and the kitchen can reorganise completely without your order changing, as long as the menu stays the same.

What do APIs do for you?

They're how separate pieces of software get combined. When a product says it "integrates with" another, an API is usually doing the work. When an agent checks a system, sends a message or looks up a record, there's usually an API underneath.

They also explain two common frustrations. If a system has no API, connecting to it is hard and sometimes impossible without manual work, so "does it have an API?" is worth asking before anyone promises an integration. And two products can claim the same integration yet behave very differently, because one uses the documented API while the other reads a web page that can change without warning.

An example

A wedding photographer takes bookings in one app and keeps her accounts in Xero. Because both offer APIs, a confirmed booking can create a draft invoice in Xero without anyone retyping the details. The two companies don't share a database; their systems exchange requests in an agreed format.

If either company changes its API without notice, the link breaks. That's why integrations need ongoing maintenance, even when nothing on your side has changed.

How are AI APIs billed?

AI model APIs are usually billed by usage, measured in tokens, with one rate for the text you send and a higher one for the text the model writes back. Chat apps for teams are more often sold per seat. Anthropic, for example, prices its API per million input and output tokens, while its Claude Team plan is charged per seat per month (Claude pricing).

The practical difference is predictability. On an API, a quiet month costs little and a month of pasting long reports costs more. A per-seat plan costs the same each month but comes with usage limits.

API or app?

An app is built for a person to use, and an API is built for software. The same service often offers both. The screen can be redesigned freely, but changing the API can break every program that relies on it, so good providers version their APIs and announce changes in advance.

Common questions

Do I need to understand APIs to use AI?

No. Most products hide them completely. They become relevant when you're connecting two systems, estimating costs or asking a provider how something works behind the scenes.

Are APIs free?

Some are, usually with limits on how many requests you can make. Many charge per request or per amount of data, and access to an AI model through an API is almost always paid.

What happens when an API changes?

Connections that depend on it can stop working, often without the end user being told. Before relying on an integration, ask the provider how much notice they give before retiring an API version.