AI Artifacts & Generated Sites

AI Code Generation

Definition

AI code generation means asking a model for a program and getting back source: a component, a script, or a complete page of HTML with its styling and behaviour included. Assistants are strongest on self-contained front-end work, where the whole result fits in one file and can be judged by looking at it. They are weakest on anything depending on context they cannot see, such as your database schema or the quirks of an internal service. What comes out is ordinary code — no runtime of its own, no licence oddity, no tie to the tool that wrote it — so once you have the file the model is out of the loop. Results vary with how specific the request was, which is why people iterate rather than prompt once and accept it.

Why It Matters

For a lot of small web work the bottleneck was never the idea, it was two days of setup. Generating a page cuts that to about ten minutes, which changes what is worth attempting at all: an internal calculator, a one-off landing page, a chart for Thursday's board meeting. The risk moves as well. Generated code can be confidently wrong, can pull a library from a source you have not checked, and can quietly embed a key you pasted into the chat earlier. Read it before you publish it, exactly as you would read a colleague's pull request.

How It Works

You describe what you want, the model writes a file, you run it and say what was wrong. Most assistants render the result straight away in a preview frame, so the loop takes seconds rather than builds. Since the model cannot install packages while writing, it reaches for a CDN — Tailwind or React arriving through a script tag — instead of a dependency tree, which happens to be exactly what a static host can serve. When the page is right you export it: one HTML document, or a folder with an index.html and its assets. Publishing after that is a file upload, and the code is yours to edit by hand from then on.

Real-World Example

An operations manager needs a shift-swap form for twenty staff and has no developer time to ask for. She describes the rules in a chat, gets a single HTML file back, corrects two things about how overtime is counted, and publishes it at shift-swaps.99helpers.site. Submissions post to a form endpoint her team already had; everything else runs in the browser. The work took an afternoon, and hosting it cost nothing.

Common Mistakes

  • Publishing generated code without reading it — a confident-looking page can hide a broken calculation or a credential you did not mean to share
  • Asking for a full application with a database and expecting a static file to deliver one; the generated front end is real, the back end it assumes is not
  • Pinning nothing, so a library loaded at a floating version can change behaviour under a page you have not touched in months

Related Terms

Put a file online in seconds

Drop in a document, an image, a page or a whole static website and share the link — free, with no build step and no server to set up.

Host a file free →