How to pick and set up an AI coding tool

Before you install anything, one piece of vocabulary, because it is the word that scares people off for no reason.

IDE

Short for integrated development environment. A text editor with a lot of tools built around it. Developers keep it open all day to write and run their work. That is the room your agent lives in.

There is more to that room than typing. Developers spent decades turning it into a place to build, understand, organize, and design, and it does a lot:

• run Python to crunch a month of analytics into a clear chart or summary
• pull records straight out of a large database with SQL
• build a landing page or a custom HTML file from scratch
• reach into the huge library of free code other people have already written

Every one of those used to mean asking someone else, a data analyst, a front-end developer, a designer, and waiting on their queue. An agent working inside this environment can now take on a good part of that work with you. The specialists still matter for the hard problems, and now you save them for exactly those, while you handle the routine work yourself.

You will not touch most of those tools by hand, and that is fine. What the editor gives the agent is three things a chat window cannot. Your actual files to work on, a place to run them, and somewhere to keep what it builds. The agent is the part you talk to. The editor is where it works.

The pairings

You have a few sensible combinations, and they all give you roughly the same experience. Pick by what you already have and what you are allowed to use.

Nothing is lost if you start with an app and outgrow it. The folder is the project, and an editor can pick it up later. What you give up is the neutral room. An editor can run either vendor’s agent on the same folder, while an app ties the room to one vendor.

Which tool should you pick?

Do not overthink this. Three questions settle it.

1. What are you allowed to use? Plenty of companies restrict which AI vendors touch their data. A friend of mine can only use Gemini at work, so Antigravity was the obvious call. Your IT policy decides more here than your preference does.
2. What do you already pay for? If you have a Claude or ChatGPT subscription, start with the matching agent so you are not paying twice. We cover what these cost in a later lesson.
3. What feels least fiddly? If installing two things sounds like one too many, Cursor bundles them into one, and the desktop apps skip the editor entirely. Starting there costs you nothing later, because the folder you work in moves with you.

Whatever you land on, the moves in the rest of the course are the same, and I will point out where a tool differs. Now set it up. The install clicks change often, so each tool below has its official guide and a short walkthrough, plus the quick steps in text. Nothing here is permanent; a wrong step just means you uninstall and try again.

Set up Claude Code in VS Code

1. Download and install VS Code from its official site.
2. In VS Code, open the Extensions panel, search for Claude Code, and install it.
3. Open the Claude Code panel and sign in with your Claude account.

The official guide has the exact clicks if you get stuck, or watch a short walkthrough.

Set up Cursor

1. Download and install Cursor.
2. Open it and sign in.
3. Choose your model when it asks; Claude is the usual pick.

The official guide covers it, or watch a short walkthrough.

Set up Google Antigravity

1. Download and install Antigravity from Google.
2. Open it and sign in with your Google account.

The official guide has the steps, or watch a short walkthrough.

Set up the Claude Code desktop app

1. Download the Claude Code app.
2. Open it and sign in with your Claude account.
3. Point it at a folder on your machine.

The official guide has the exact steps, or watch a short walkthrough.

Set up the Codex desktop app

1. Download the Codex app.
2. Open it and sign in with your ChatGPT account.
3. Point it at a folder on your machine.

The official guide has the install steps, or watch a short walkthrough.

How much you let the agent do

Every one of these tools has the same dial, only the labels change. At one end the agent just reads and plans and changes nothing. In the middle it does the work but checks in with you. At the full-auto end it edits and runs on its own. While you are learning, you will live in the middle.

You switch on the fly: Shift+Tab in Claude Code, the run mode in Cursor, /permissions in Codex, the execution policy in Antigravity. The middle is where you build the instinct for what the agent will do. Once you have it, you will reach for auto more and the checkpoints less.

I run mostly in full auto now, only because so many hours in the middle taught me what to expect. Drop to read and plan for anything you cannot afford to get wrong. A later lesson goes deep on when to use which.

Open a real folder

Here is the step that makes it click. An agent with no folder open is a chat. An agent with a folder open is working on something.

Make an empty folder anywhere, call it something like agent-test, and open it in your editor, or point your desktop app at it. That folder is now the agent’s workspace. With the usual permissions it works on what is inside that folder and leaves the rest of your machine alone, unless you grant it more. That is exactly the boundary you want while you are learning.

Your first task

End here, with something small and real. In your empty folder, ask the agent in plain English to create a file called hello.md. Have it write three sentences about your product in the file.

Watch what happens. It tells you it is going to create the file, asks permission, you approve, and the file appears in your folder. That is the whole loop, and every lesson after this is a bigger version of it. You just ran an agent.