Skip to the content
I Want AI To
17 builds on showA website

Which kind of AI assistant for coding do you need if you have never coded?

AI coding assistants come in three kinds: suggestions inside an editor, a chat you talk to, and an agent that edits files for you. If you have never coded, start with a chat: you describe what you want and get files back, with nothing to install.

Short answer: start with a chat, and leave choosing between tools until you’ve built something and know what you’d want a tool to do better. The rest of this post explains why, and the one mistake that stalls most first attempts.

What people mean by an AI assistant for coding

Broadly, it means a tool that turns plain-language requests into code, or into suggestions for code. That covers a lot. It could be something that finishes your line as you type, something you have a conversation with, or something that goes off and changes a dozen files while you make tea. They share a label but feel very different to use.

If you’ve never written a line of code, what the tool is called matters less than what you need to know before it’s any use to you. That’s how this post sorts them.

The three kinds: editor suggestions, chat, and agents that edit files

This split is our own plain-English way of sorting the category, not an official standard. Other people slice it differently.

Editor suggestions. You install a code editor, open a project folder, start typing, and the assistant offers to finish what you’re writing. It’s very useful if you already know what you’re writing and where your files live. It assumes you’re driving.

Chat. You open a browser tab or an app, describe what you want in ordinary sentences, and it replies with code or with files to download. You open them, say what’s wrong, and ask again. There’s nothing to install. In Claude, making downloadable files is a setting called “Code execution and file creation”, under Settings, then Capabilities. At the time of writing (9 October 2026), Anthropic’s help page says it’s available on every plan, Free included, on the web and in the desktop and mobile apps.

Agents. These work inside a project and change the files themselves, while you supervise and review. One example is Claude Code on the web. At the time of writing, Anthropic’s docs say it needs a GitHub repository and is available on the Pro, Max and Team plans (and on some Enterprise seats). Claude copies your repository to a machine in the cloud, makes its changes and sends them back to GitHub as a separate copy for you to review. It’s a powerful tool, but it assumes you already have a project and can judge what it changed.

So the three kinds do different jobs and expect different amounts of knowledge from you before you start.

Why the league tables are written for people who already code

Search for an AI assistant for coding and the top results are mostly ranked lists. They’re written for working developers choosing between tools that all assume you can already find your way around a codebase. So they compare things like which editors a tool supports and how well it fits a way of working you already have.

Those are good questions if you already work that way. If you don’t, it’s like being handed a comparison of professional kitchen knives when you haven’t yet boiled an egg.

We don’t rank anything here, and we don’t think a beginner should start from a league table at all. You’d be choosing a tool to solve problems you haven’t met yet, judged by someone else’s needs. A better first question is which kind of assistant suits someone starting from nothing.

Which kind a complete beginner should start with

A chat. It’s the only kind where the skill you already have is the one you need: describing what you want. You’ve done that all your life, to friends, to colleagues, to the person at the paint shop. There’s no new screen to learn first.

The loop is simple. Say what you want, get files back, open them, say what’s wrong. That loop is most of the skill.

It also makes real things. Each build on this site started from five plain-English prompts, and Claude’s files are published exactly as it wrote them, with a list of what went wrong. A website is the broadest example: a three-page site made from a description and a screenshot we liked. There’s also a landing page with an “I’m interested” form that opens an email, an app that keeps a shared job list in a database, and a Chrome extension that counts words, made of three files.

One thing to know before you start. Anthropic’s help page talks about Word, Excel, PowerPoint and PDF files and doesn’t mention web pages. If Claude shows your code in the chat instead of offering a download, ask it to give you each file to download. The set-up page says what to do if that still doesn’t work.

The mistake to avoid: installing an editor before you have a project

It usually goes like this. Someone hears that developers use an editor with an assistant built in, downloads one and opens it. The first thing it asks for is a folder to open: a project folder. If you don’t know what that is or what should go in it, you’re stuck before you’ve started.

The AI isn’t the problem. The editor is built around a project, and you haven’t got one yet. The assistant inside it is waiting for something to help with.

You end up learning the editor instead of building the thing. That’s a fine hobby, but it isn’t what you came for.

The fix is boring and it works. Start in a chat, get a set of files you can see in a folder on your computer, and only then will an editor make sense, because now you have something to open in it. Chat first, project second, editor third.

What you still have to do yourself: check it on a phone, put it live

The chat writes the files. It doesn’t check them for you and it doesn’t publish them, so two jobs are yours.

First, check the result on a phone. A page that looks tidy on a laptop can fall apart on a small screen. When we tested the website build at laptop and phone size after its second prompt, 78 of 94 checks passed: two pages had no main heading and some text was as small as 11.5px. Open your pages on a phone and scroll through them. If something’s off, go back to the chat and say so in plain words.

Second, put it live. The route on this site needs no command line. On GitHub, you use Add file, then Upload files, to put your files in a repository. GitHub’s docs say you can upload up to 100 files at a time that way. In Cloudflare, you connect that repository under Workers & Pages, then Create application, then Pages, then Connect to Git. Cloudflare’s docs say it then builds and publishes your site every time something changes in the repository. Its pricing page says requests to static files are “free and unlimited”, so a site made of plain files costs nothing to host. The set-up page walks through every click.

When to move to a heavier tool

That point does come. You’ve built a few things, you know what a project folder is, you can find your files without panicking, and downloading files from a chat and uploading them again has started to feel clumsy. That’s the moment to switch. A league table or a friend’s favourite tool isn’t a reason, but a chat that has started to slow you down is.

By then an editor assistant, or an agent that saves changes to your repository for you, will feel like a relief instead of a wall. The set-up page shows the agent route we tested, Claude Code on the web on the Pro plan, and how it fits with the same GitHub and Cloudflare set-up. If you’re weighing up which AI to use for a first site, this comparison for beginners covers that question.

It’s a bit like learning to drive in a quiet car park before you meet the motorway. Nobody skips the car park, and nobody there will sneer at you for being in it.

Questions

Questions people ask.

What is an AI assistant for coding?

A tool that turns plain-language requests into code or suggestions for code. It can sit inside a code editor and finish what you type, work as a chat you describe things to, or act as an agent that changes the files in a project for you.

Do I need to know how to code to use one?

Not with the chat kind. You describe what you want in ordinary words, open what comes back and say what's wrong. You'll pick up words like file and folder as you go, but you don't need them first. The build pages on this site show plain-English prompts and the files Claude made from them.

Is ChatGPT good for coding?

We haven't tested it, so we can't say. Every build on this site was made with Claude. For a beginner, the kind of assistant matters more than the brand: you want a chat you can describe things to that hands you back files.

Do I need to install anything to start?

Not for a chat. Claude works in a web browser, and Anthropic's help page says file creation is available on every plan, including Free, at the time of writing (9 October 2026). Later you'll want free GitHub and Cloudflare accounts to put your files live, and the set-up page walks through both.

When should a beginner switch to an editor-based assistant or an agent?

Once you have a project worth keeping, you know where its files live, and downloading and uploading files from a chat has started to slow you down. Until then, stay in the chat.