Cursor Playground

What it is

Cursor is an AI-native code editor. It is built on top of VS Code, so your extensions, keybindings, and settings mostly carry over, but the AI is woven through the whole editor rather than bolted on as a side panel. In practice you spend your time in four places: autocomplete as you type, quick inline edits, a chat/agent panel that can change many files and run commands, and cloud agents that go off and work on their own.

Getting started

1. Download the app from cursor.com and open a project folder, or import your VS Code setup on first launch.
2. Sign in and pick a model. Cursor is model-agnostic: you can switch between frontier models from Anthropic, OpenAI, Google, and others, and let it auto-select per task.
3. Start typing to get Tab suggestions, hit the inline edit shortcut to change a selection, or open the agent panel to describe a larger change in plain English.

The four ways to use it

1. Tab (autocomplete)

Cursor’s Tab model predicts your next edit, not just the next token. It will suggest multi-line changes, jump to the next place you probably want to edit, and follow the pattern of a change you just made. Press Tab to accept. This is the feature you use thousands of times a day without thinking about it.

2. Inline edit (Cmd/Ctrl+K)

Select some code, describe the change, and Cursor rewrites it in place with a diff you can accept or reject. Best for focused, local edits: refactor this function, add error handling, convert this to async, write a test for this. You stay in the file and keep the flow.

3. Agent (chat)

The agent panel is where the bigger work happens. Describe a task and the agent reads the relevant files, edits across the codebase, runs terminal commands, installs dependencies, and iterates on failures. Give it context explicitly with @ to reference files, folders, docs, or symbols. Review its plan and diffs before accepting.

4. Cloud agents

Cloud agents run the same agent asynchronously on their own machine. You hand off a task — from the app, the web, or Slack — and it works in the background, opens a pull request, and reports back. Useful for parallelizing chores, fixing CI, or picking up an issue while you do something else. (This page, for instance, was written by one.)

Getting good results

• Give it context. Point at the exact files with @ instead of hoping it finds them. More precise context beats a longer prompt.
• Write rules. Put durable, project-specific conventions in AGENTS.md or .cursor/rules so every request starts with them — test commands, style, architecture, gotchas.
• Work in small loops. Ask for a slice, review the diff, run it, then continue. Long unsupervised runs are for well-scoped tasks with clear success criteria.
• Let it verify. The agent can run your tests, the dev server, and the type checker. Ask it to prove the change works, not just to write it.

How it compares

All of these tools use frontier models under the hood, so the real differences are the form factor — where the tool lives and how much autonomy it takes — and how locked-in you are to a particular model.

The honest summary: if you want to keep your hands on the wheel inside an editor while still being able to delegate, Cursor is the sweet spot. The terminal tools (Claude Code, Codex, Droid) suit people who prefer the shell, and the autonomous agents (Devin) suit teams comfortable handing off whole tickets. Most people end up using more than one.