Skip to content

Course · Cursor · Getting started

Getting started

You’ve inherited a repo. A friend built budgetcli over a few weekends - a small self-hosted personal-finance API for tracking accounts, importing bank statements, and flagging overspend - and now it’s yours, with your own money running through it. It mostly works. It’s lightly documented. And there are places where you already suspect the original author got something subtly wrong. That’s the situation this course uses as its running scenario - real code, real stakes, imperfect understanding - and you can map every move in this chapter onto a repo of your own. This chapter is the first hour: from nothing installed to a first change you actually reviewed and shipped.

The reader this chapter assumes is the engineer who already lives in VS Code - a handful of extensions, a typed codebase open in the gutter, inline completions already humming - and is eyeing an AI-first editor with reasonable suspicion. (Coming from another editor? Everything here still applies; you just skip the VS Code import step.) The good news lands in the first five minutes: you don’t have to give up the editor you know, because Cursor is the editor you know. Cursor is a fork of VS Code, built by Anysphere and shipped as its own desktop application. Your muscle memory survives the move; what’s new is everything bolted onto the editor, and that’s what this hour is about.

What “a VS Code fork” buys you (and what it costs)

Section titled “What “a VS Code fork” buys you (and what it costs)”

Because Cursor is a fork, it inherits VS Code’s whole configuration model. Your keybindings carry over. Your user settings.json is honored, and so is a project’s .vscode/settings.json. Cursor offers a one-click import of your Extensions, Themes, Settings, and Keybindings from an existing VS Code install - surfaced as a first-launch onboarding prompt (if you don’t see it, the Settings route always works) and, any time after, from the Settings menu (General → Account → VS Code Import, or Cmd/Ctrl+Shift+J, as of September 2026; migration guide) - so the editor you open on minute one already looks like the one you closed.

There’s one inheritance the fork can’t carry, and it’s the part most likely to trip you up on day one: extensions don’t come from Microsoft’s marketplace - its terms prohibit third-party clients - so Cursor’s extension panel is backed by Open VSX instead, and some extensions never ship there. The escape hatch when one you need is missing is side-loading a .vsix. The install lesson owns the full story and walks the .vsix path step by step.

Set the inheritance aside for a second, because the reason you switched editors at all is the AI, and it shows up in two distinct places from the moment you open the app:

  • The Agent sidebar - a chat panel where you give the agent a task in plain English and it reads files, proposes edits, and runs the read-propose-review loop the whole course leans on. This is the surface that does the real work in this chapter.
  • Tab - inline completion that predicts your next edit as you type, accepted with the Tab key. It’s the descendant of the ghost-text autocomplete you already know, and it’s a different tool for a different moment: Tab is for the keystrokes you were already going to make; the Agent is for the work you’d rather describe than type.

Cursor has a third surface - a CLI that runs the same agent in your terminal and in CI. Its primary command is agent, the entrypoint Cursor has documented since January 2026 (CLI changelog, which also notes that cursor-agent remains as a backward-compatible alias for older scripts). We deliberately leave the CLI for its own chapter later in the course. For your first hour, two surfaces is plenty.

Signing in is a step in the next lesson, but it’s worth knowing the shape of the plans before you pick one. Cursor’s individual tiers run a free tier, then paid Plus, Premium, Super, and Ultra (as of October 2026, cursor.com/pricing); the older Hobby, Pro, and Pro+ names continue only on legacy plans for existing customers (models and pricing), and Teams and Enterprise layer admin controls on top. Usage is metered against model-usage pools rather than a flat seat fee, which matters once you start running the agent heavily - but for shipping one reviewed change this hour, the free tier is enough to follow along, and you can decide on an upgrade later with real usage in front of you instead of a guess.

budgetcli imports transactions from bank CSV exports, and it mangles some dates: a statement row that plainly reads one day lands in the database as another. You can check that yourself - no sample CSV ships with this site, so write a one-row CSV with a day-first date such as 03/02/2026, run it through the importer, and compare the date it returns with the date in the row. That’s the chapter’s ticket: small, real, and verifiable. We’ll go from zero to that fix shipped, in five moves:

  • Install Cursor - the desktop app, the VS Code migration prompt, and a first run-in with the Open VSX extension source (plus one .vsix side-load) - Install Cursor and migrate your VS Code setup
  • Sign in and pick a plan - understand the tiers before you commit, and confirm the agent is alive - Sign in and choose a plan
  • Meet the two surfaces - the Agent sidebar and Tab are different tools for different moments, and knowing which is which is half the skill - Agent sidebar vs Tab
  • Watch one turn of the loop - how the agent searches and reads before it proposes, and where your review lands - Understand one turn of the loop
  • Hand it the date bug - and watch that loop close on a real fix you review in the diff and commit when it’s green - Ship your first reviewed change

None of it is hard. The point of doing it as one continuous task - on a repo you’ve genuinely just inherited - is that the habits you pick up here are the same ones every later chapter leans on: read before you edit, review the diff before you accept it, and commit the moment a change is green and reviewed. Cursor gets faster and more autonomous as the course goes on - by the time you reach the parallel-agent chapters and Cloud Agents (Cursor agents that run on remote machines instead of your editor), it can fan a single prompt out across isolated git worktrees and run those agents in parallel - but the review never goes away. On code you didn’t write and don’t fully trust yet, those habits aren’t optional polish. They’re the whole game.