What's an API? The underground pipes your AI can use

When I explain APIs to friends, I use a picture. Imagine a busy street lined with shops. Each shop is a website: a calendar, a store, a weather service, your bank. Now imagine that under the street there's a network of pipes running straight into each shop's back room, where the actual stuff is kept.

You and I use the sidewalk. We walk up to a window, look in, scroll, click, squint and walk on to the next one. An API is the pipe underneath. Instead of browsing, a program turns a valve and gets exactly what it asked for, in a second, without looking at a single page.

That's the whole idea. API stands for application programming interface, which is a mouthful, so I just say pipes.

The pipe is faster, it costs less and it breaks less.

Browsing vs. pipes

Say you want your AI to tell you how many orders came in this week. On the sidewalk, it would open the store's dashboard, wait for it to load, find the right menu, read the number off the screen and hope the layout didn't change since last time. Through the pipe, it asks one plain question and gets back one plain answer: 37 orders, here they are.

The pipe is faster, it costs less and it breaks less. Websites change their buttons and menus all the time, because they're built for human eyes. Pipes are built for programs, so the shop keeps them steady. When your AI can use a pipe, it does the same job the same way every day.

The key to the valve

Pipes aren't open to everyone. To turn a valve, you need a key, and that's what an API key is: a long string of letters and numbers that proves you're allowed to draw from that shop's pipe, usually only your own account's water.

Pipes have rules

Every pipe has a limit on how much water you can draw per minute. In tech-speak that's a rate limit. One shop might allow a hundred requests a minute, another a thousand an hour. Draw too fast and the valve closes for a bit.

This matters more than it sounds. A good AI setup paces itself: it asks for what it needs, waits when it's told to wait, and doesn't hammer the same pipe in a loop. When something you built suddenly stops working at the busiest hour, a rate limit is one of the first places to look.

Not every building has pipes

Some shops never built pipes, or keep them for big partners only. When that happens, your AI can still do the job the human way: it drives a browser, walks down the sidewalk and clicks through the pages like you would.

That works, and we use it every week for the handful of things that have no other way in. But it's the slow lane. Pages load, layouts change, a pop-up gets in the way, and a person sometimes has to step in. Our AI teammates use pipes wherever they exist and the sidewalk only where they have to.

What this means for you

You don't need to learn to code to use any of this. You need one new habit when you pick tools: check whether the tool has an API or a ready-made connector for your AI. A connector is a pipe someone already plumbed for you, so you can hook it up with a few clicks instead of a key and a manual.

• When you compare two tools that look alike, pick the one with an API or a connector. Your future self will thank you.

• When your AI says it "can't see" something, ask whether there's a pipe for it before you settle for screenshots and copy-paste.

• Keep your keys out of chats, give each tool only the access it needs, and throw a key away the moment you're unsure where it's been.

• When something works fine by hand but breaks when the AI does it at volume, think water per minute.

Once you see the pipes, you start noticing them everywhere. The calendar invite that showed up on its own, the receipt that landed in your inbox, the weather on your lock screen: those are pipes doing quiet work under the street. Giving your AI access to a few of them is often the difference between a chatbot that talks about your work and a teammate that does some of it.