Cursor review: The AI code editor that's made us forget about GitHub Copilot

The first time Cursor finished a refactor I'd half-started, I sat there for a second and just looked at it. Three files changed. New tests passing. Names consistent with the rest of the codebase. The thing read my git history and understood what I was trying to do before I'd really committed to a direction.

That moment was about six months ago. I've been a Cursor subscriber ever since, and I've slowly stopped opening VS Code.

This isn't really a review in the traditional sense, because Cursor doesn't lend itself to one. It's not a feature you turn on; it's an editor that has fundamentally rewired how I write software. The question worth answering is whether that rewiring is worth $20/month, and whether it's the right pick versus the competition that's caught up considerably.

The codebase context thing

Most "AI in the IDE" tools start with a chat window or autocomplete on one file. Cursor's bet is different. It indexes your entire codebase, and that index is what makes its suggestions feel less like autocomplete and more like having a senior engineer who's already read your code.

In practice this means a few things. When I ask Cursor to "add error handling here," it doesn't suggest a generic try/catch. It uses my project's existing error handling pattern, the same logger I'm already importing, the same response shape my other endpoints return. When I'm renaming a function, Agent mode finds every call site and updates them, including the ones in tests, including the ones in obscure adjacent files I'd forgotten about. None of this requires me to manually attach files to a chat. The model sees what it needs.

GitHub Copilot has been racing to add similar features. It got "edits across files" recently, and it works fine for small changes. The difference is that Cursor was built around this capability from day one, and you can feel it everywhere. It's the difference between a tool with codebase awareness bolted on and a tool that's been designed assuming it has codebase awareness.

Agent mode is where it earns its rate

Cursor's Agent mode lets you describe a multi-step change in plain English and watch the editor execute it. "Add a dark mode toggle that persists to localStorage and respects the prefers-color-scheme media query." Cursor reads your existing code, identifies the relevant components, writes the new ones, updates the existing ones, and shows you a diff. You review and accept, or you push back and ask for changes, and it iterates.

The first time this works on a real task, you remember why you got into programming. The fifth time, you start to wonder how you ever shipped anything without it.

There are real limits. Agent mode is bad at strongly architectural decisions. It will happily implement what you ask for, even if what you asked for is the wrong abstraction. It can also get into loops on tricky bugs, generating fix after fix while you watch your tokens burn. Disciplined developers will get more out of it than vibe coders, because you need to know what good looks like to direct it well.

What costs and what's slower

Cursor is $20/month for the Pro plan, the same as ChatGPT Plus and Claude Pro. There's a Business plan at $40/month per user with SSO, audit logs, and zero data retention. The Pro tier gives you 500 fast premium-model requests, then drops you to slower variants for the rest of the month. I've never hit the cap on light days; I have hit it during heavy refactor sessions.

The premium-model dependency is the biggest practical limitation. When OpenAI or Anthropic is slow, Cursor is slow. When you're offline, Cursor's smartest features don't work. Tab completion still functions, but Agent mode and Composer require the cloud. This isn't unique to Cursor (Copilot has the same issue), but it's worth knowing if you work on flights or want truly local AI.

Where it's not yet the right call

Cursor is built on a fork of VS Code, which means most VS Code extensions work. Most. Not all. If your daily driver is JetBrains, Vim, or any non-VS Code editor, Cursor either won't work or feels like a clean rebuild that you might not want. JetBrains has its own Junie AI agent and Copilot integration. Vim users have plugins. None of them are quite as fluid as Cursor's experience, but the friction of switching editors is real.

The other group I'd hesitate to recommend Cursor to: people who don't yet know what they want. AI coding tools are highest-leverage in the hands of developers who can read code carefully and recognize when the suggestion is subtly wrong. For someone learning to code, the AI's confident wrong answers can be actively damaging. Use it as a tutor occasionally; don't use it as a teacher.

The competition picture

GitHub Copilot ($10/month) is cheaper and now bundles GPT-5 and Claude. For pure autocomplete and one-off code suggestions, the gap to Cursor is smaller than it used to be. Where Cursor still wins is the editor experience, the agent capabilities, and the speed of new feature rollouts. Cursor ships fast. Copilot, being inside Microsoft, ships at GitHub's pace.

Windsurf (formerly Codeium) is a more direct competitor and has been catching up on agentic features. The pricing and capability are close. I've tried both and prefer Cursor's interface, but I wouldn't call it a clear win.

Claude Code in the terminal is a different beast, agent-native and command-line-first. Some developers will prefer it. If you live in a terminal, try it before committing to Cursor.

Verdict

Cursor is the AI coding tool I'd recommend to a working developer without much hesitation. It's not perfect, the dependency on cloud models is real, and the agent capabilities still need careful supervision. But it's the only AI editor I've used that consistently makes me faster at non-trivial work, not just at typing.

If you write code professionally, the $20/month math is easy. Saving an hour per week pays for it ten times over. If you're a hobbyist or just starting out, Copilot at $10 is the safer floor. And if you've already built habits around a different editor, the switching cost is real and the marginal upside is shrinking. The lead Cursor had in 2024 has narrowed; in 2025 it's still ahead, but the field has caught up enough that "best AI code editor" is no longer obvious.

For now, I'm still here. I'll let you know when I switch.