What is an AI artifact — and why it isn't an app

If you’ve asked Claude, ChatGPT, or Gemini to make you something visual lately, you’ve probably met an artifact — even if nobody called it that.

What an artifact is

An artifact is a self-contained piece of HTML and code that the assistant generates and renders inside the conversation, usually in a panel beside the chat. Instead of describing a chart, it draws one. Instead of explaining a form, it shows you a working one. Claude calls them Artifacts; ChatGPT has Canvas; the others have their own version of the same idea.

They’re one of the genuinely great things about building with AI. The assistant writes the code, you see the result immediately, and you change it by asking — “make the total bold,” “add a column for tax.” It’s fast, visual, and forgiving.

How the AI uses them

Behind the scenes, an artifact is the assistant’s way of showing its work as a working thing rather than as a wall of code. It keeps a live copy it can edit and re-render as you iterate, so the two of you can shape a tool together in a few turns. For sketches, prototypes, and one-off visuals, that loop is hard to beat.

Where they run out of road

The catch is in the definition: an artifact runs in your chat window, for you, right now. Which means:

• There’s no server — it isn’t hosted anywhere someone else can reach.
• There’s no database — anything you type in is gone when you close the tab.
• There’s no sharing — a screenshot or a file isn’t a link people can open and use.

A financial model with fields you fill in and numbers that update is the classic example. On your own screen it feels like software. But you can’t send it to your team, and it won’t remember a thing you entered. It’s a demo of an app, not the app.

That gap — between “my AI made a brilliant artifact” and “my team is actually using it” — is exactly what Croft closes. Here’s how an artifact becomes a real app →

More from the blog