Cursor: Beginner

Beginner

Cursor is an AI code editor (a fork of VS Code) built around two things: inline Tab autocomplete that finishes your code as you type, and an in-editor Agent that reads your project, makes changes, and shows you a diff to review. This guide gets you from install to your first real Agent task.

What you need

• A computer running macOS, Windows, or Linux
• A project (a folder of code) to open
• A Cursor account (you sign in on first launch)

Install

1. Download Cursor from cursor.com and install it like any desktop app.

2. Open Cursor and sign in when prompted.

3. Open your project folder with File → Open Folder, then point Cursor at the root of the project.

Tip

Because Cursor is a VS Code fork, on first launch it can import your existing VS Code extensions, themes, and keybindings, so the editor feels familiar from minute one.

Tab: autocomplete that thinks ahead

As you type, Cursor suggests the next chunk of code in grey. Press Tab to accept it.

The part people miss: Tab isn’t just line completion. After you accept a suggestion, press Tab again and Cursor jumps to the next edit it predicts you’ll make, often somewhere else in the file. Accept, jump, accept, jump. You’re steering, not typing every character.

Key	What it does
Tab	Accept the suggestion; press again to jump to the next predicted edit
Esc	Dismiss the current suggestion
Cmd/Ctrl + K	Inline Edit: select code and describe the change in place

Your first Agent task

The Agent is the chat panel in the sidebar. Open it, describe what you want in plain English, and let it work. It runs a loop: it gathers context from your codebase, makes edits, runs commands when needed, and then surfaces the result as a diff for you to approve.

A good first prompt is small and concrete:

Add a /health endpoint to the API that returns { status: "ok" }. Show me the diff.

By default the Agent uses Auto model selection, which picks a model that balances capability and cost, often Composer, Cursor’s fast in-house model. It’s a sensible default while you’re starting out. The Agent will:

1. Search your code to find the right file
2. Propose the edit as a diff
3. Wait for you to Accept or Reject

Caution

The Agent asks before running terminal commands. Read what it proposes before approving, especially anything that installs packages, deletes files, or touches git.

When something goes wrong: Checkpoints

Before a significant edit, Cursor saves a Checkpoint: a local snapshot of your codebase, separate from Git. If a change goes sideways, revert to the checkpoint and try a clearer prompt. You don’t have to get the wording perfect the first time.

This is the agentic loop

Gather context → act → show a diff → you verify. That cycle is the same one behind every AI coding tool, not just Cursor. Understanding it makes every tool here easier to use.

How AI coding tools work The agentic loop behind every AI tool, and why the harness, not just the model, decides what it can do.

Three habits that make it work

1. Be specific. “Make it better” gives the Agent nothing. “The date shows in US format. Switch it to DD.MM.YYYY” gives it a target.
2. Work in small steps. One change, review the diff, then continue. Don’t ask for a whole feature in one prompt yet.
3. Read the diff before you accept. The diff is your safety gate. Skimming it is how you stay in control.

You’re done when…

• ✅ Cursor is installed and you’ve opened a project
• ✅ You’ve accepted a Tab suggestion and used the jump-to-next-edit
• ✅ You’ve run one Agent task and reviewed its diff before accepting

Next step

Ready to feed Cursor the right context and give it standing instructions?

Cursor: Intermediate Feed Cursor the right context with @-symbols and steer it with .cursor/rules.