I can simply describe the boundaries of what I want to achieve, and the AI takes care of the rest. Review collected by and hosted on G2.com.

The main downside is that my own coding skills drops because I no longer have to think long and hard about how to solve a coding problem. Cursor's AI is able to come up with a good solution in 99% of cases.

Also I find the price to be quite expensive, and there should be more tokens included in the subscription. I often find that my spending per month is 2x-3x the subscription price because of purchasing extra tokens. Review collected by and hosted on G2.com.

What I like best about Cursor is that it lets me work with different AI models for different types of tasks instead of relying on one model for everything.

For example, I can use stronger models for architecture, backend logic, or complex product decisions, and faster/cheaper models for smaller UI changes, documentation, or simple fixes. This makes the workflow feel much more flexible and practical.

The multi-agent style workflow is also very useful. I can separate tasks like implementation, review, testing, and documentation, which helps me move faster while still keeping control over the quality of the project. Review collected by and hosted on G2.com.

The main downside for me is that heavy users can run into limits quite quickly, especially when working on larger projects with many files, long context, and multiple AI-assisted workflows.

I would like to see more generous usage limits or clearer options for power users who rely on Cursor every day. For example, larger context windows, more included model usage, and better visibility into remaining usage would make the experience even better.

Overall, Cursor is very useful, but for intensive development work the limits can sometimes become the main friction point. Review collected by and hosted on G2.com.

It’s a smooth experience having AI and IDE workflows merged together. The Composer model has become my go-to for coding, and I use it for almost 90% of my tasks. Composer usage feels very generous, and the premium API allowance is also solid, giving me enough room to use state-of-the-art models when I’m working on more complicated problems. Review collected by and hosted on G2.com.

Performance can be sometimes a bit dodgy but nothing to really be concerned about. For non-programmers, the onboarding experience could be better. The UI is a bit cluttered and needs some getting used to Review collected by and hosted on G2.com.

I use Cursor to make all my programming dreams come true. It solves my every programming need, allowing me to track individual line changes and run terminal commands beautifully. Cursor has stopped glitching a long time ago and runs flawlessly every time. It generates plantuml and mermaid diagrams, k8s yaml and plan files, making my workspace look like a giant notion-looking tree with filebases for each sub-project. When I type 'redis' it offers me a Redis plugin and tons of other integrations.

The Cursor team responds to my needs incredibly fast, literally shipping features in 48 hours of me simply thinking of the missing feature. Two days later it's there. I don't know how they read my thoughts, but these updates feel like magic to me. I fully switched from using vscode to Cursor because it fits my dev mind like a glove, and I convince everyone to switch. The initial setup was very easy, with auth flow working like in Figma.

The most important part to me is the generosity of the company. The literally allowed me work for free in Auto and fully get accustomed to agentic coding before subscribing. The 'Auto' mode was impressive a year ago, well before Composer 2.5. So now 'Auto' is Composer2, which is plenty of IQ for your daily tasks. I totally recommend Cursor; it's a 10 out of 10 from me! Review collected by and hosted on G2.com.

I don't like how file inclusion '@' action works: I want to be able to configure what's in my inclusion list. I don't want to @insert .ts or .tsx files, there are tens of thousands in hundreds of Ring Platform clones, so I simply drag-and-drop them into window whenever need to mention in chat. I would like the files, when I drag them into chat area, be included as relative-path-to-file text string values instead of widget-looking filename-only elements, because frankly, they sometimes take a while to render in the input area of chat (my cpu is m4max). Sometimes Cursor perceives these paths as ones in project's main codebase dir (though I specifically dragged its clone counterpart).

Another meh is the skills itself that people still use everywhere. They are linear, not structured and fill up context windows fast.

I use structured data in my own nested json. Anysphere can buy my entire LegioX library of skillsets. It would tremendously save tokens on natural data processing for all users. And THAT I would like to include with @. A skillset. Review collected by and hosted on G2.com.

Agent mode is what hooked me. I describe what I need — a new screen, a database migration, a full component — and it writes the whole file, not just a snippet. What makes it actually useful is that it reads the rest of my codebase first. It already knows my folder structure, my naming, my patterns. I don't explain context every time.

I'm a civil engineer and surveyor, not a professional dev. I had ideas for years but couldn't execute them alone. Cursor changed that. It's the first tool that made those ideas real — actual working software, deployed, used by people. That's not something I could say before. Review collected by and hosted on G2.com.

The cost adds up fast when you're in a heavy build session. I've hit my Pro limits on intense days and had to slow down or wait. For a solo developer or someone doing this as a side project, that's a real friction point. Not a dealbreaker, but it's something you feel. Review collected by and hosted on G2.com.

Cursor has been my “AI Command Center” since late 2024.

I can do pretty much everything in one place: pick whichever AI tools fit my needs, orchestrate my workflow, and integrate other useful applications into it. I mostly use the interface to visualize the code, and when I have too many agents running, I just switch views. I’m not stuck with a single layout; I can alternate depending on what I need at the moment.

The software never feels slow. It’s consistently smooth and has a premium feel. The price is competitive too: the yearly plan is $16, and the “Auto” mode is generous—coding 3 hours a day still isn’t enough to burn through all the quotas.

Onboarding is straightforward: install it, add your project, and you’re done. Auto-tab was my first real “wow” moment, because it saves me from copy-pasting the same thing N times throughout my code. Overall, the AI integration is the smartest I’ve used so far, especially compared with other well-known “AI agent coding” tools. Review collected by and hosted on G2.com.

Some of the newer updates can be annoying. Before, Cursor would receive around two updates per day on average, sometimes even more. It’s better now that they’ve reduced the frequency, but some updates still feel too big, and the UI/UX keeps changing. As a result, my shortcuts no longer work the same way, and my workflow gets disrupted for a while.

I don’t think the product needs more enhancements right now; it mainly needs more stability. Review collected by and hosted on G2.com.

AI is integrated into every aspect of Cursor, and I like how all those integrations make sense and make development easier. Switching between plan/debug/agent mode is intuitive. I appreciate how models from OpenAI, Anthropic, Google and Cursor are available for use, allowing me to choose the best model for each task. Composer 2.5 is extremely affordable and very comparable to other frontier models.

The way that getting assistance from agents is integrated into every part of the tool is super helpful - from the terminal, separate agent interface, and even the integrated browser. It's transformed the way that I write software, letting me focus on the interesting parts (architecture and system design) instead of the more mechanical parts of implementation.

New features are constantly being released and improved upon, and it does a good job of onboarding. Documentation is extensive and helpful, and new models and features often come with promotional periods so you can try them out. Review collected by and hosted on G2.com.

There are a couple of edge cases that make some things frustrating, especially with regards to multi-repo and teams features. Specifically, internal plugins and local plugin development have some issues where custom rules, commands, and agents aren't loaded as they should be.

It also isn't always clear which rules and skills are being applied. They've recently greatly improved the interface to show where the tokens in the context window are coming from, so something similar to see information about skills and rules would be great.

Finally, every time you open a new window, MCP servers are automatically enabled. If you have a lot and you only want to conditionally use some, you have to remember to go turn them on/off. Some way of remembering that or a quicker way of toggling that perhaps.

Performance is generally great, but the interface for entering a git commit message sometimes is extremely delayed - but only there. Review collected by and hosted on G2.com.

I really enjoy how fast and straightforward it is to pick from a wide range of AI models. I also appreciate that the subscription includes more API usage than I expected for what I paid, and that there’s built-in review functionality that lets me have another AI agent double-check any code changes made.

As someone who used to use VSCode, I love that its built on it and didn't take much work to move over to it. The UI is similar because of that and any extensions I had also work for it too. Review collected by and hosted on G2.com.

The ability to integrate your own AI models or even self-hosted ones is lacking and could be simplified. I saw that deepseek had lowered their costs so wanted to use it in cursor but found that they had removed it as an option and I had to add it myself by entering a specific name. Review collected by and hosted on G2.com.

The project's understanding capabilities that Cursor can display are very good, and the interface is simple and well-made, almost like a wrapper for VS Code. The plugins, skills, MCPs, and rules work well on a daily basis. However, performance is a somewhat delicate point: it consumes a lot of memory. Regarding the price, I believe the PRO plan suits me well for 20 dollars. I never needed support, and overall, the features, available models, and integrations work perfectly. Review collected by and hosted on G2.com.

This new feature where, every time I open the Cursor, it opens in the agents window, I find it terrible. I wish there was an option to choose which default I want to use, instead of it always opening this way. Review collected by and hosted on G2.com.

Switching from a classic IDE like VS Code to Cursor feels very natural. The Agent interface integrates seamlessly and gives you a wide selection of models to work with (or for you) at a moment’s notice. And if you don’t want or need to look directly at your code, you can use the Agents view instead. It’s closer to a traditional chat interface and still lets you work across multiple projects in parallel. Review collected by and hosted on G2.com.

Cursor is an Electron app, and that comes with the downside of being very RAM-hungry when it runs for a long time; similar to Chrome and other Electron-based apps. Review collected by and hosted on G2.com.