Hot Take about Cursor

Discussions
1

I’d rather pay a premium price for a truly premium product than watch it get watered down trying to serve everyone. Too many great tools lose their edge chasing mass adoption.

I’d rather pay more for excellence than settle for compromise. I respect products that double down on quality, even if it comes at a cost - because in the long run, that’s what delivers real value.

In my view, Cursor is one of the most powerful tools built for software engineers (and I mean software engineers, not pretenders). But with that power comes a decision point: do you chase wide adoption by lowering the barrier to entry, or do you double down on serving the users who truly. value deep functionality, stability, and innovation?

I would rather see Cursor continue refining its core experience, even if it means a higher price point, than stretch itself thin trying to cater to everyone. Building for the mass market often results in compromises. It’s the classic product trap: in trying to be everything to everyone, you risk becoming forgettable to everyone.

Premium pricing enables focused development, better support, and a more intentional roadmap. It funds not just infrastructure, but thoughtful design and long-term velocity.

We’ve seen this model work before:

Linear didn’t chase every user - they chased the best experience for focused teams. Their pricing reflects that.

Raycast unapologetically positions itself as a Mac-first, productivity-first tool. As a result, it has rabid fans and strong traction among professionals.

Superhuman went all-in on performance and polish over affordability, and it carved out a loyal base because of it.

Cursor is on that trajectory. It’s not just another AI tool, it’s a new kind of IDE. And if it wants to remain the best, it shouldn’t shy away from pricing that reflects that ambition.

Not every product needs to be cheap. Some should be worth it.
Cursor IS worth it.

10 Likes
2

Actually I disagree for the last sentence. I am on trial (which offer PRO functions for 2 weeks) as I do not marry a ■■■■■■■ the first date. Curzoe made me bilieve that there is a way for properly working Ais, despite its terrible interface from UX aspect. It did literally everything I asked. I wanted to make a YT video about how amazing it is.
Then day by day it became less and less usable, failing on even very simple tasks. I just made my 3rd test: Copilot made what I wanted in 20 minutes, Cursor runs the same and same circles. And lies. A lot. Tells ‘I checked the code’ (No, it did not). ‘I made the backup’ No, it did not. ‘This method could do that function you need’. No it could not. ‘I applied the rules’. No, it did not.
It is good in 1 thing: saying sorry for my wasted time. But it cannot give it back.

And as it does not show what model it uses , iyou do not even know why things happen. Not mentioning the ‘suspicious’ blocking from usage…

3

If you are now running on Auto, your requests are routed to different models. If you are still on your trial, you could disable the “Auto” toggle and try a specific model that you choose.

1 Like
4

I’m not discrediting anyone’s experience - I’ve seen similar concerns echoed elsewhere. But I can only speak from my own use. Cursor isn’t perfect, but it’s powerful. What I’ve realized over time is that I started relying on it too much. At first, it felt magical - and it was, to a degree. But the “magic” faded when I expected it to do everything for me. It’s not magic. It’s a tool.

My best results come when I approach it like a senior dev or architect: I plan out my session, create a clear ToDo list, and work through it iteratively. If I prompt it expecting perfect results on the first try, I get mixed outcomes. But when I guide it - milestone by milestone - it’s almost flawless.

Cursor shines when used by someone who knows what they’re doing. And honestly, trying to make it appeal to everyone might weaken what makes it special. It’s not meant to replace engineering skills - it amplifies them. As the saying goes, “It works even better if you’re already smart.”

To those frustrated: learn to code. Learn to design and architect systems. Then use Cursor to implement them piece by piece. That’s delayed gratification. Expecting it to build your dream app from a vague prompt isn’t just unrealistic - and worst of all, it devalues the craft.

This tool is phenomenal in the hands of professionals. Let’s not dilute it by dumbing it down. Just my take.

8 Likes
6

The logical decision from user aspect to see ehat task uses what as an ‘auto’ choice. Not using a specific model. I understand the business motive behind it, but I still do not know I have to use copilot for pretty simple tasks instead of using Cursor and using Cursor only as a ‘merge tool’ for code parts.

And of yourse , seeing the actual model wants delete my files for misinterpreting a command…that would help a lot…in what do NOT use.

7

God, I designed hell of an amount of things, but coders work under me, built the workflow/pipeline; what happens and when, where the bottlenecks are, how to save time and money. I am old-fashioned. I deliver working solutions, not half-baked ones. I offer quality in every aspect of my jobs over design trends, wrong approaches (like agile development usage for cases it is stupid to use). I grew up in an era people were forced to code well, put a system together on paper or on a whiteboard before writing a single line. You knew the start and you knew the end goal. No agile, Miro, Figma-like corporate jokes where dark/light themes matters more than an efficient solution for the customer.

1 Like
8

It sounds like you’re in mid-level management expecting Cursor to do the magic you saw people do for years. If what you’re saying is true, then Cursor should work great for you. Design your system, and implement each piece of those pipelines, exactly how those coders you saw would do. It’s here to write the syntax. You’re here to write the system. I don’t see the issue. You say “Hey Cursor implement a login page using my supabase auth”, and it writes the page. You work out the details in supabase, test the login… You’re expecting magic, and unfortunately I think the death of Cursor is catering to people like you. I’m sorry, that’s just my take. Don’t hate me for it AngryUser.

3 Likes
9

I made this comment elsewhere – are they trying to be a fully autonomous agent, or a professional editing tool/partner to software engineers. Those are currently incompatible things.

Also true on quality versus cost. I agree with you 100%, let lovable own the $20 market. You want the $1000+ market. Tired of fighting the token efficiency nonsense in the prompts. Instead of getting the answer I get 55 file reads and inaccurate answers because it can’t piece things together. Max mode is max mode, change the underlying prompts accordingly.

2 Likes
10

So you’re saying only professionals should be able to use tools like Cursor and thus the pricing should be through the roof? Yeah, no. You’re forgetting a large group of people from less fortunate parts around the world who were able to create their dream application without the costs that usually comes with building one.

2 Likes
11

Auto mode successfully told everyone that cursor is not good

4 Likes
12

Auto mode is literally great, people just don’t know how to use it. And this is what I mean. People are relying on AI to do everything for them, literally EVERYTHING. Even the thinking aspect of this. Cursor is absolute magic in the hands of educated, seasoned engineers. The only issue I’ve had with it is the looping that happened last week. Other than that, auto mode is great. The cost of auto mode is practically nothing compared to if you’re using specific models explicitly. The reason people need to use models with advanced thinking is because they have no idea what they’re doing. Right now Cursor is magic for people who know how to use it. It’s proprietary software. ANYONE can fork VS-Code from GitHub if they really thought Cursor was another run of the mill program, but it’s not (they’ve already tried). The algorithm and methodology they use behind the scenes is absolute gold. It’s actually INSANE to me that people will get on here - spending their whole day on here - bashing it, belittling the devs, and trying to pass the product off as garbage. These people have NO IDEA what they’re doing, I guarantee you that much. They probably watched a YT influenzas video about it and thought they could code the next big thing without having to fire a single neuron…

2 Likes
13

That’s 100% what I’m saying. And the product would be better in the long run and that’s a tough pill for some people to swallow.

2 Likes
14

Yes bro. Cursor is quite strong right now, maybe a little more expensive is worth it, but remember: not every country is rich enough to use Cursor, to be able to compete globally, you need to have a good pricing strategy. And if Cursor only wants to compete in a certain area? Okay, go ahead and increase the price, it’s good right now.

1 Like
15

That’s because you don’t know how to “catch” the Claude 4 sonnet model in auto mode. I’ve used many powerful models, like Grok 4 thinking, but I still find Claude to be better because it maintains user rules better than almost all other AI models. Other AI models will often “forget” user rules when the conversation gets longer, but Claude remembers a rule repeatedly by memorizing it. Try Claude, it codes very well and also follows user rules very well

1 Like
16

So please explain how to use it, provide an example, talk is cheap.

If most people don’t know how to use a feature correctly and there’s no tutorial, how can you prove it’s a good feature?

17

You need to create a vision document for your project. Once you create that, then you need to create a milestones-to-mvp document that outlines what features and functionality you will need in order to have a minimally functioning product. Once you have that, then you iteratively generate micro-todo lists for each milestone. Whenever you’re working on the feature, use the micro-todo list as context. For each phase of the list you complete, update the todo list with what you’ve accomplished. By this time, it will be time for you to begin another conversation because you’ll be running out of context. Before you begin a new conversation, append your most recent work to the top of the todo list. Rinse and repeat until you reach your MVP. I don’t let the agent delete anything automatically. I always discuss package installations, fork-in-the-road type situations, etc. And honestly, if you’re not a professional, if you don’t have a degree or an education or any experience with this stuff, then you need to learn as much as possible while doing this so that you can control as much as possible. You don’t want to spend your time developing a product, only to get screwed over two months into launch when something breaks. You still need to know the product like the back of your hand. A one man company is totally possible with this tool, but you need to work with as much intent and focus as possible, otherwise you’re going to end up frustrated and be one of these people calling Cursor a bad product.

And if you really want some keen advice, you need to choose the most performant tech-stack you can. We’re really at a point where choosing inefficient stacks is a bad idea because now that tools like Cursor exist, implementing the code isn’t even the important part anymore. The important things are adhering to intelligent, tried-and-true engineering principles that will make your product performant and win the marathon.

3 Likes
18

Are you telling me to switch to Kiro IDE?

All your actions are just to make good use of traditional AI methods like GPT-4.1. If paying for Claude can solve the problem, why waste valuable time on this? Not all projects are built from scratch; many existing projects already have context that can be directly used.

2 Likes
19

We programmers actually have better option. Claude Code can mark milestones, can execute tasks with multiple agents, and it can even be replaced by the K2 API. The K2 after replacement is so cheap in price, it’s simply perfect. You can consider it as Cursor turning to serve managers because it can’t compete with Claude Code.This is not a question of price at all.

4 Likes
20

did the same, approach of Cursor team and hidding of problems without answears make it clear

1 Like
21

Asked in so many places a) what is the product idea of Cursor and b) why can’t someone implement planning features like CC has it. If I am to waste bunch of my tokens, I would rather spend it on planning (I am spending probably at least 50% of my tokens on planning in CC) vs being annoyed by LLM and whatever translation layer in between doing opposite of what I asked it to.