Collected from public review sites and shown with their original ratings, rebased to our 1–10 scale. They give extra context but do not count toward the Cursor community score.
santosh ramrakhianiTrustpilot
US · last month
4.0/10
2/5 on Trustpilot
Great development tool, but a very disappointing experience during a production incident
I have been using Cursor extensively for real CRM development and have completed a significant amount of development with it, including previously on the $20 and $60 plans. I upgraded to the Ultra plan because of the amount of development work we are doing.
Unfortunately, my recent experience during a production incident was extremely disappointing.
Our CRM suddenly became inaccessible to the entire team. During the investigation, Cursor's Agent took us deeply into an RLS/security diagnosis involving database policies, grants, permissions, repository analysis, repeated diagnostics and UAT. We relied on that direction because we were dealing with a live production outage.
A substantial amount of my included Ultra usage was consumed during this investigation, and our development team lost approximately one full working day dealing with the issue instead of continuing planned development.
The actual root cause ultimately turned out to be much simpler: our Supabase Nano instance was resource constrained/unhealthy. We upgraded from Nano to Micro, and the CRM immediately recovered. No RLS rollback was required.
What was particularly frustrating was that the security/RLS investigation had consumed a significant amount of Cursor usage and our team's time, while the actual production issue was infrastructure capacity.
I contacted Cursor Support and spent considerable time explaining the situation. To their credit, support reviewed the account and explained the usage figures in detail. However, despite acknowledging the production impact and the day of development time lost, Cursor ultimately refused to restore any of the consumed Ultra allowance or provide a $100 Ultra extension as a resolution.
I understand that the Agent genuinely performed the work and that Cursor incurred model-provider costs. My concern is that a production incident was taken down an unnecessarily expensive diagnostic path, resulting in both substantial usage consumption and real business/developer time being lost.
I was not asking for a refund of my subscription. I asked for a reasonable restoration of usage so I could continue using the Ultra plan I had already paid for.
The support experience was professional, but the final resolution was disappointing.
Cursor is a powerful development tool, and I have found real value in it. But when an AI-assisted production diagnosis goes significantly off track, I believe there should be a mechanism for reviewing exceptional cases where the customer suffers both substantial usage consumption and significant productivity loss.
I hope Cursor considers improving this aspect of its customer support and incident-handling process.
Asked for the staff a refund cause I run out of capacity in 2 days only.
They tell me : no cause you use capacity.
200 dollars for a month run, and I got an empty brain LLM.
AVOID buy on cursor, I send them a paper.
I think the reviews people are giving doesn't make sense. They seem to not know how to use it. A bad farmer quarrels will his tools. I and my team have been using Cursor for months now and its excellent. You need to know how to prompt, what rules to put etc to get the maximum out of it.
Multiple issues from 6 months of usage.
1. agent doesn't follow directions
2. rules are ignored
3. for that reason, a lot of time goes to fix the agents behavior
4. good features (such as docs) are constantly removed because *reasons*
5. get a MCP - observe your agent not using the tools EVER
6. doesn't follow output rules (such as caveman)
7. one time admitted to close to 20 violations to rules, skills, instructions, you name it, when tasked to enforce those rules
8. almost only way to actually enforce rules is creating hooks
9. however cursor isn't smart enough to actually craft working hook systems
10. mainly because everything it build is overly complicated
11. some version didn't even newline between functions
12. still doesn't comment code if not asked
13. doesn't use old, working code, if not asked to do so, and when asked, doesn't comply
14. constantly blocks itself from being able to do edits if editing hooks (such as blocks edit tools before enabling some other tool -> can't finish edits because blocked itself)
15. customer service doesn't answer messages
16. customer service flags messages pointing out problems
Those were the first few problems that came to my mind.
The support team doesn't care about customer problems.
Since the beginning of July, I've been having problems with Cursor regarding editing files in legacy projects with Latin1/Cp1252 encoding. The agent ignores the file and project settings and edits them in UTF-8 encoding, breaking the original encoding and eliminating accents related to the internationalization of languages like Portuguese and Spanish, for example.
On August 4th, I submitted a report on the company forum, responding to the most recent post on the subject and providing more specific details about the situation, and the subsequent responses were evasive.
This problem has a history dating back to 2023, and apparently the company has no interest in addressing it, as any localization that doesn't use ANSI as the default character set seems to be unworthy of the product's attention.
A quick Google search revealed at least 9 similar open topics without a definitive solution, and in all of them the Cursor team evades the issue, saying there is no deadline to resolve the problem and that it is up to the user to work around the situation. The problem is that even following the suggestions, the problem persists, causing rework, loss of money in the form of token consumption, and the desire to switch to another product that better meets my needs.
Aug 12 they sent me "Some on-demand usage incurred on or after July 1, 2026 was not invoiced when the applicable billing cycle ended. This issue has since been resolved." When I asked Aug 15 for a refund - they said the most recent renewal was on July 29, which is already outside the 14-day refund window.
Technically it is true, mistakes occur. But why not consider the mentioned Cursor side mistake (backdate invoice correction) when calculating this days?
Cursor has been instrumental in helping developers I know, begin pair-coding with AI. Also I noticed that managing AI has improved every month, while the quality of the AI changes based on model. I appreciate their focus on creating their own models; in addition to, keeping up with new models in the industry. 5 stars for keeping things real.
It is very good for a prototype, but bad for big scale
It is very good for a prototype, but very bad for big scale product.
you always have to micro manage it, and it can improvise and implement and solution that is the opposite of what you already had implemented and was happy with. plus you have to test every time, since most fixes you have to try 3 or more times before it actually fixed it.
The benefit is that you can use it for quite cheap (if you are okay using slow mode most of the time)
Even though there are so many bad reviews about cursor on trustpilot, I have used it for just over a week and have had zero problems. The price is unbeatable (when using a paid subscription to claude lasts 5 minutes) and the speed of coding is quite good for £15.
I switched from Codex VSCode to Cursor because the usage limits make much more sense. With Codex, the agent usage gets used up way too quickly, sometimes a single prompt already consumes 10% of the quota, which is honestly unreasonable.
If Cursor can continue maintaining this pricing and usage model, or even reduce the usage limits further, I’ll probably keep using Cursor long-term, upgrade to a higher plan, and recommend it to my friends as well.
It's trash. Nothing works as intended, it's purely made for ripping people off. It makes mistakes like typos, so that people have to burn their usage % after these errors. At the end of the day, it restructured my complete project, ruined most of my bridging, and rolled out a completely different concept with tons of functional problems. I'd never use this again, even if they paid me. It wasted my time entirely.
What can i say Cursor has been useful, probably spent over $3,000 while using cursor over the year and honestly had no problems whatsoever as long as you prompt it well it'll design your task. The trial is a joke but over than that it's a good software.
Just today, the first time ever it blocked me as "all tokens used" without any info on when it will unfreeze. Previously it was promised to have unlimited usage - just slower. So they lied.
The tool is quite uncapable - You waste money, but job is not done. That sucks.
Its a great tool however the costs are absolutely insane. I just literally spent $4.58 on a single prompt! Can you image this day in and day out? It's not sustainable...
What initially seemed like a good deal turned into a nightmare.
I really like it for code exploration, and so on, but for coding it is completely useless - not that they did their job, but these AI tools make coding really annoying.
Also, the IDE got recently super slow, and as someone who uses vim regularly, this is absolutely unacceptable.
So while it is it definitely interesting and worth trying and has its target audence - I am not one of them.
i think it's great, it really is, but it is really expensive, i am currently on the Pro plan, and i'm paying 20 dollars a month, and i get limited FAST, only after a couple days, so i know the api usage to premium models isn't something cursor can control, but it is crazy expensive, if you want a chance to use the software for more than a couple days then you will have to go with the Ultra plan for 200 dollars a month, so if you are rich then i can absolutely recommend, but if you are not prepared to drop some serious cash on usage and prevent limitations then it's not for you.
I've had mixed experiences. On the one hand, the AI isn't very smart and often struggles to understand issues, requiring significant effort to fix bugs it creates. On the other hand, the pricing is quite reasonable. Compared to other AI IDEs, this one doesn’t use tokens and allows unlimited model usage, which is a big plus. However, it’s still not the best, as there are free alternatives that offer similar functionality.
