What is an API?

An API is a doorway between programs. Instead of a person clicking buttons in an app, software sends a structured request — “here’s a prompt, give me a reply” or “create this invoice” — and gets a structured response back. Nearly every modern app is held together by APIs: payments, maps, email, login, and now AI.

How AI APIs work

1. Get an API key from a provider (OpenAI, Anthropic, Google and others) and add billing.
2. Send a request from your code or automation tool: which model, your instructions (system prompt), the user’s message, and settings like maximum length or temperature.
3. Receive a response: the model’s output, plus how many tokens you used.
4. Pay per use: priced per million input and output tokens, with different rates for different models.

APIs also offer features consumer apps may not expose: tool calling, structured JSON output, batch processing, prompt caching and fine-tuning.

An analogy for the meatbags

An API is a restaurant’s kitchen window. You don’t wander into the kitchen and cook; you pass an order slip through the window in the format the kitchen expects, and food comes back. The chat app is the dining room with a waiter. The API is for people ordering in bulk.

Why it matters for the Chosen

The API is where AI turns from a personal assistant into infrastructure you can build on:

• Automation: plug models into workflows — summarise every new support ticket, draft replies to every lead, tag every product — via Zapier, Make or n8n. See workflow automation.
• Products: most AI apps are wrappers calling model APIs behind the scenes.
• Cost control: for high-volume tasks, API usage can be far cheaper than seats, and you choose cheaper models for simple steps.
• Client work: building API-driven tools and automations is a core service for AI freelancers and agencies.

Related: the Model Context Protocol is a standard that lets AI apps use other software’s capabilities, often wrapping APIs so agents can use them.

Things to watch

• Costs scale with usage. Set spending limits and monitor — a loop or a viral app can burn through credits fast.
• Key security. Leaked keys get abused; store them in environment variables or a secrets manager.
• Rate limits and outages. Build retries and fallbacks.
• Data policies. Check how each provider handles API data, retention and training, especially for client information.
• Model changes. Providers deprecate older models; test before switching.