Cursor Reviews & Product Details

The best part about Cursor is that, when I’m vibe coding, I can use different AI models depending on what I need. For example, I can put the chat in Ask mode, which means it isn’t coding anything—it just answers my questions and gives explanations. The part I like most is when I switch to MCP and have Cursor AI take control in real time while it codes. That gives really good feedback to the AI, and I don’t need to copy and paste big chunks of data. It’s honestly the best. Review collected by and hosted on G2.com.

Overall, the only issue I’ve noticed is that it feels like the credit gets used up too quickly. Review collected by and hosted on G2.com.

Agent mode handles multi-file changes on a real codebase without losing the thread — it maintains my production Astro + Tailwind blog and its studio tooling. Codebase context is strong; it finds the relevant scripts on its own. It is also good at writing the boring safety code — idempotency guards, dry-run modes, audits — when asked properly. Review collected by and hosted on G2.com.

It will happily generate a lot of code fast, so review discipline is on you. It is occasionally confident about repo facts it has not checked, so claims need verifying against the actual files. Usage-based pricing can spike during heavy agent sessions. Review collected by and hosted on G2.com.

One of the best tools for agentic coding. It not only has its own AI models, like Grok/Composer, but also integrates with other models such as Opus/Sonnet and GPT. Performance is very strong, and I like having both a fast mode and a standard mode depending on what I’m doing. The ROI feels insane for what you get. Having Cursor model usage alongside other model usage also feels superb.

The UI is really cool, with both Agent mode and IDE mode, and it’s easy to switch between them. Support for the tool also seems widely available online, with plenty of tutorials and documentation. Review collected by and hosted on G2.com.

The UI looks cool, but at times it can feel a bit too complex. Also, aside from the Cursor models, integration with other models seems limited. For example, other open-source models like GLM 5.3 can be used via an API, but that still lacks native integration. Review collected by and hosted on G2.com.

I don't anymore. I used to like lots till they charged me without me choosing on demand services Review collected by and hosted on G2.com.

I had previously disabled Cursor's on-demand usage because I did not want additional pay-as-you-go charges.

After upgrading to the $60/month Pro+ plan, I discovered a $200 on-demand limit on my account and received an additional $77.88 invoice.

I asked Cursor Support when on-demand had been enabled, when the limit was created and what action authorised it. They told me their tools cannot provide that information, but they still refused to cancel or refund the charge.

They also told me on-demand cannot enable itself and must be explicitly activated.

I had disabled it, did not knowingly re-enable it, and they cannot show me when or how it was enabled.

I'm cancelling Pro+ and moving elsewhere. I would personally look at Claude, ChatGPT, GitHub Copilot, Windsurf or another alternative instead.

A company using pay-as-you-go billing should be able to show customers exactly when additional spending was authorised. Review collected by and hosted on G2.com.

A lot of times I know what I want to build, but I don't know exactly how to write the code for it. I don't always need to figure out the syntax or structure, I can explain the feature and let Cursor create a first version. Then I test it and tell it what needs changing... low the easy simple chat interface and all the integrations you could do in it.. Review collected by and hosted on G2.com.

Sometimes it changes some codes that was working fine... so I always have to careful before giving it a prompt.. Review collected by and hosted on G2.com.

The AI integration in Cursor feels genuinely woven into the coding workflow instead of bolted on as an afterthought. Inline code generation and chat-based editing pull context from the whole codebase, not just the open file, so suggestions actually match existing architecture and coding patterns rather than generic boilerplate. Tab-to-accept autocomplete is fast and often predicts multi-line edits correctly, saving a lot of repetitive typing during daily development.

The interface stays close to a familiar VS Code layout, so there's almost no learning curve coming from that ecosystem — getting started took minutes rather than a real onboarding process. Extensions and settings carry over smoothly, and the editor stays responsive even in larger projects with many open files. Integration with existing Git workflows and terminal usage feels seamless, and referencing specific files or symbols directly in a prompt makes debugging and refactoring noticeably quicker than switching to a separate AI tool.

On pricing, the value holds up well against the time it saves — faster iteration and fewer context switches easily justify the subscription cost for a small technical team. Support has been reliable when needed, with documentation that covers most common issues, so there's rarely a wait to keep moving. Overall it's become a core part of the day-to-day coding process. Review collected by and hosted on G2.com.

Pricing gets frustrating once usage scales — the fast request limits on the standard plan get consumed quickly on larger codebases, and hitting that ceiling mid-task means either slowing down to conserve requests or upgrading to a higher tier sooner than expected. More transparency around real-time usage consumption would help, since right now it's often only clear after the fact.

Performance can also dip on very large repositories — indexing takes noticeably longer, and the context window occasionally doesn't fully capture relevant files scattered across a big project, so suggestions miss dependencies living outside the immediate working directory. That means double-checking generated code more carefully on bigger builds than on smaller ones.

The AI still occasionally hallucinates function signatures or library APIs that don't actually exist, especially with less common packages or internal libraries it hasn't seen much of, so verification against actual documentation remains necessary rather than optional. Multi-file refactors sometimes need manual cleanup afterward since the model doesn't always catch every downstream reference that needs updating.

Onboarding new team members to the AI-specific features (custom rules, context management, model selection) takes more explanation than just picking up a standard editor, since getting real value out of it requires understanding how to prompt and scope context effectively. Minor learning curve, but it's there. Review collected by and hosted on G2.com.

What I like most about Cursor is that it feels like an IDE first and an AI tool second, which means I'm the guy in the chair. The AI is deeply integrated into the development workflow instead of feeling like a separate chatbot. Features like editing across multiple files, understanding the codebase context and making targeted changes save a lot time compared to constantly copying code back and forth.

I also appreciate that cursor doesn't try to replace the developer. It speeds up repetitive work, helps me navigate unfamiliar code, and lets me stay focused on solving the actual problem. It strikes a good balance between automation and keeping me in control, which makes it something i can use throughhout the day rather than occasionally Review collected by and hosted on G2.com.

Sometimes the AI makes broad changes when i want a small, targeted edit. It can also lose context in very large or complex codebases.

Occasionally, the suggested code looks correct but misses subtle project specific details. I would also like more transparency into why it chooses edits or approaches Review collected by and hosted on G2.com.

That the AI is embedded in the core of the editor and not attached with a side widget. The autocomplete with Tab is super fast and being able to launch refactorings that affect multiple files at once from Composer saves you a lot of time. Being a fork of VS Code, the learning curve is zero; you open your workspace, it indexes the project, and you're working from the first minute. Review collected by and hosted on G2.com.

RAM consumption gets spammed quite a bit when indexing large repositories, and sometimes the credit system falls short if you push Composer hard daily. It would also be cool if the context management allowed for a bit more manual fine-tuning of the files to ignore. Review collected by and hosted on G2.com.

I love that I can ask questions about my own code without having to explain the whole project every time. It helps me understand what’s going on, points me toward the relevant files, and suggests a few different ways to fix issues. The interface feels familiar, and it works smoothly with the terminal and Git integration. Review collected by and hosted on G2.com.

I've also seen it suggest changes that look reasonable but don't account for some edge case. Review collected by and hosted on G2.com.

It’s very smart and genuinely useful. The capabilities are strong, and it connects to many MCPs. It’s also relatively affordable, although a “Cursor Lite” option with a $10 subscription could be a great fit for people who still aren’t completely sold on it. Review collected by and hosted on G2.com.

Originally, it was simpler—an AI-powered IDE that worked amazingly well. I don’t fully love the other screen that’s focused only on the agents. Review collected by and hosted on G2.com.