Code

Cursor - review

Name: Cursor - review
Item: Cursor
Rating: 7.9
Author: Piotr Olszewski

7.9/10

Cursor works on a subscription model, with a free variant with limitations and paid plans, and on top of that comes usage depending on how heavily you load the more expensive models, especially in agent mode. In practice, this means two layers of cost: a fixed fee and a variable part depending on work intensity, which is worth watching if you don't want to get a surprising bill. The provider changes the details of thresholds and billing fairly regularly, so I'm not giving specific amounts here - check the current pricing directly with the provider before deciding, because what I write today may be outdated in a month.

Last updated: Nov 6, 2026

Try Cursor ↗

Rating: 7.9/10.

Cursor is a code editor built from the ground up around AI, technically a fork of Visual Studio Code, developed by Anysphere. Put simply: it's VS Code where artificial intelligence isn't an add-on plugin but the core of the entire experience. It doesn't host its own model - under the hood it uses leading models from providers like Anthropic and OpenAI, and builds its value on the layer that understands these models and feeds them context from the entire repository. It matters because it was the first to truly solve the problem that was previously scattered across dozens of plugins: editing aware of the whole project, agent mode, and completion that predicts not one word but your next move. My thesis is simple: this is today's best AI-coding UX on the market, and if you write code professionally, sooner or later you'll test it anyway. You pay for this with a subscription, usage costs, and the obligation to read every change the agent suggests.

Strengths

Best AI-coding UX on the market. The entire app is designed so AI is at hand at every moment, not hidden behind a shortcut and separate panel. Chat knows open files, suggestions appear where you're looking, and getting into the habit of working with the model takes fifteen minutes, not a week.
Agent mode that actually executes tasks. Instead of pasting you a code snippet to copy, the agent itself edits files, creates new ones, runs commands, and iterates on the task until completion. That's the difference between an assistant giving hints and a collaborator you delegate concrete work to and check the result.
Edits aware of the entire repository. Cursor can pull into context the project structure, related files, and dependencies, so its proposals fit what you already have, not a generic example from the internet. In larger, mature code, it's precisely this context awareness that decides whether a suggestion is useful or trash.
Tab completion that predicts your next move. This isn't classic single-line autocompletion - the model proposes entire blocks, subsequent edits, and jumps to other parts of the file, often accurately guessing what you want to do next. During refactors and repetitive changes, it saves genuinely a lot of keystrokes.
Multi-file work in a single request. You can ask for a change that touches several files at once - renaming, moving logic, adding a function with all consequences - and Cursor will handle it consistently across the entire set. This is something classic editor plugins usually failed at.

Weaknesses

Costs can surprise you. You pay a subscription, and on top of that comes usage depending on how intensively you use the more powerful models in agent mode. With heavy work, the bill can be higher than you expect, and the billing model changes often enough that you need to watch it.
The agent can change too much. You request a small fix, and you come back to a diff that touches five files, refactors things you didn't ask for, and breaks something along the way. The bigger the task, the easier it is for overzealousness that you won't notice without careful review.
Requires discipline in reviewing changes. The convenience is so great that it's tempting to click accept without reading, and that's the easiest way to let in a bug or silent regression. Cursor doesn't relieve you of being an engineer - it just shifts the weight from writing to reviewing, and that reviewing needs to be done seriously.

What forReally shines in everyday programmer work on an existing project: refactors, adding features that touch many files, fixing bugs, writing tests, handling unfamiliar code you just jumped into. Agent mode is great for well-defined tasks - change this API across the entire repo, add edge case handling, rewrite this module to a new pattern - because it sees consequences across the entire file set. Tab completion gives the most with repetitive, mechanical work, where you know what you want to write and the model just gets ahead of you. The better you describe the task and the narrower the scope you set, the better result you get.

Who is this forYes: for professional programmers and people who code a lot and regularly, especially in medium and large projects, where awareness of the entire repository makes a difference. Also yes for people who want to squeeze the maximum out of AI and aren't afraid to review someone else's work, including machine work. Caution: if you're just starting to program, it's easy to fall into the trap of accepting code you don't understand - Cursor will speed you up, but won't teach you the basics for you. Also not necessarily for someone who writes sporadically or has a strictly limited budget and doesn't want to watch usage, because then lighter and cheaper solutions will suffice.

AlternativesIf you don't want to change editors, consider GitHub Copilot plugged into your own IDE - weaker in agent mode and working on the entire repo, but simpler and often cheaper for just completion. If you work mainly in the terminal and like the agent approach without an editor layer, reach for tools that work in the command line, such as Claude Code or Aider - they give great control and plug conveniently into existing workflow. Windsurf is the closest direct competitor in the same league of AI-first editors and is worth comparing on your own code. VS Code itself with well-chosen plugins will also close part of the needs cheaper, if you care mainly about suggestions, not an agent aware of the entire project. Best to test two or three of them on a real task - differences are fastest seen in your own repo, not in a description.

VerdictRating: 7.9/10. Cursor is today's reference point when it comes to working comfortably with AI in code - agent mode, awareness of the entire repository, and Tab completion add up to an experience that classic plugins don't touch. It's neither free nor innocent for your budget, and the agent can be overzealous, so without the habit of reading every diff it's easy to let an error slip in. Still, for someone who codes professionally, the speed boost is real enough that the cost usually pays for itself. I recommend it without hesitation to professionals and people who code a lot - just treat it like a very fast junior whose work you always review. Before purchasing, check the current pricing with the provider, as the billing model likes to change.