Cursor Review

What Cursor actually does

Cursor is a fork of VS Code with AI baked into every interaction. The differentiators over Copilot are the multi-file edit ("Composer"/Agent mode) that can refactor across an entire codebase, the chat sidebar with full repo context, and the Cmd-K inline edit that replaces selected code with whatever you describe in natural language. Under the hood, Cursor lets you choose the model — Claude Opus and Sonnet, GPT models, Gemini, and a fast in-house model for autocomplete.

The honest framing: Cursor took the right bet that the editor itself, not the suggestion sidebar, is the surface where AI in coding has to live. It remains the category leader among AI-native editors, though the field has gotten more crowded — Windsurf (formerly Codeium) has built a genuine following with a similar agentic-editor pitch, and Anthropic's own Claude Code CLI has pulled some terminal-first engineers away from IDE-centric workflows entirely. GitHub Copilot has also added its own multi-file "agent mode," narrowing the gap Cursor opened up.

What works well

The multi-file Composer/Agent is the feature that changed the workflow. You can describe a refactor — "rename this prop across all consumers, update the storybook stories, and adjust the tests" — and Cursor produces a diff across every affected file that you review before applying. For real engineering work, this is still one of the most reliable AI features that operates at the right unit of work.

Tab completion is excellent. The custom model is fast and contextually aware enough that pressing Tab feels like a continuation of the keystroke, not an interruption. Compared to Copilot, the latency and accuracy difference is noticeable in daily use, though Copilot has closed some of the gap.

Model choice matters. Claude is generally stronger for refactoring and reasoning across larger contexts; GPT models are stronger for some specific patterns. Being able to switch per-task without leaving the editor is a real workflow advantage.

Where it falls short

Cursor is a fork of VS Code, which means it lags upstream VS Code by a release or two on extension compatibility and security patches. For most users this is invisible; for users with strict supply-chain requirements (regulated industries, large enterprises), it is a real concern.

Cursor moved from a fixed "fast requests" allowance to a usage-based credit system, which is more transparent about what heavy model use actually costs but also means power users can burn through their monthly allocation faster than before, especially when running Agent mode against large repos. The Business/Teams tier pricing is still steep relative to Copilot Business once you account for model spend, and the new Ultra tier exists specifically because Pro's allocation isn't enough for the heaviest users.

The Composer/Agent can be overconfident on large refactors. It's right enough that you trust it; it's wrong enough that you must always review the diff before accepting. Teams that adopt it as if it were correct by default produce regressions.

Who should use it

Cursor is the right tool for any full-time engineer or solo founder shipping production code. The productivity gain over a non-AI editor is large, and the gain over Copilot is smaller but real, especially for refactoring-heavy work. Engineers who prefer working outside a full IDE should also look at Claude Code or Windsurf before committing, since the AI-native-editor category is no longer a one-horse race.

Casual coders, students, and engineers who write very little code outside of established patterns get less marginal value than they expect. Copilot inside their existing editor (often free for students or open-source maintainers) is a better starting point.

Pricing notes

Cursor now runs four tiers: Hobby (free, limited usage), Pro ($20/mo, usage-based credit allocation with model choice), Teams/Business ($40/seat/mo, team features and centralized billing), and Ultra ($200/mo, roughly 20x the usage of Pro for the heaviest users). The shift from fixed "fast request" counts to usage-based credits means actual monthly cost can vary more than it used to depending on which models and features you lean on. Pro remains the right starting choice for serious individual users, with Ultra worth it only if you're running Agent mode constantly against large codebases.

The product-management gap

Cursor accelerates the act of writing code. It does not decide what code to write — what feature to ship next, what tech debt to address, what bug to prioritize, what user feedback to act on. Those decisions still require either a human PM or a product agent that reads support tickets, sales calls, and analytics, then files prioritized engineering tasks. The editor automates the typing; the program management still requires a layer above.