API

An API (Application Programming Interface) is a contract between two pieces of software that defines how they can interact. It specifies what requests a client can make, what data it needs to provide, and what responses it will receive. APIs allow developers to use functionality built by others without needing to understand or access the underlying implementation.

In AI, APIs are the primary way most developers interact with large language models. Services like OpenAI, Anthropic, and Google expose their models through REST APIs where developers send prompts as HTTP requests and receive model outputs as JSON responses. This abstraction means you can build applications powered by frontier models without running any ML infrastructure yourself. API pricing is typically based on token usage, with separate rates for input and output tokens.

APIs also play a central role in AI agent architectures. Tool calling allows models to invoke external APIs during a conversation, enabling them to fetch live data, interact with databases, send messages, or trigger actions in other systems. The quality of an API's design - its documentation, error handling, rate limiting, and versioning - directly affects how reliably AI-powered applications can be built on top of it. As AI moves from research demos to production systems, API design has become one of the most consequential decisions in the stack.