Have You Heard of Cursor?

AI is now an everyday topic in startups. Yet, when I ask this question, reactions vary significantly. Most non-developers haven't heard of it, which makes sense since Cursor is an AI Code Editor.

Cursor Homepage Main Screen

Cursor is a tool that assists developers with coding tasks, similar to Microsoft's Copilot. Looking at the service demo on the homepage, it's clearly a service for developers. They received $8 million in investment from OpenAI. While generative AI has shown exceptional performance in coding, Cursor stands out as one of the most 'hot' services these days.

I started using Cursor because I wanted to try 'vibe coding.' I'm self-taught and can do a bit of development, but I'm far from being confident enough to make it my profession. It's just a field I enjoy exploring on my own, even without being asked.

This experience has given me basic development skills. I understand what coding is and what it 'feels like' to create a service using a programming language. But the idea that even a non-professional like me could create a service through 'vibe coding' was exciting!

I'll write about this in another post, but I created this website through vibe coding all by myself.

Being close to the AI industry, I'm more aware of AI's current reality. I initially thought there might be some hype. I installed Cursor just to check out a new tool, which I enjoy doing. And then, in the office, I let out an audible gasp. "Is this thing a ghost?"

'Vibe Coding' - I Thought It Was Just a Saying, But It Might Actually Work?

When I need to analyze statistics with over 100,000 lines of data, I write Python code to solve it. Google Sheets becomes unbearably slow with 100,000 lines. I opened Cursor but didn't actually send any prompts to the AI. I was writing Python code as usual. With Cursor's code suggestion option turned on, as I write code (even just one sentence), Cursor suggests the next line of code at each moment.

The key here is that Cursor predicts what I'm going to do and makes suggestions. For example, if I group data by User ID, Cursor immediately suggests grouping by Workspace ID after I write that code. And at that moment, I was actually thinking, "Now let's do it by Workspace ID..."

An image created by GPT since I couldn't find an appropriate meme. The actual feeling was much more spine-tingling.

Of course, statistical calculations have a fairly limited set of possible operations. The fact that the operation I was thinking of was selected next might be somewhat coincidental. Still, this was the first time I experienced an AI that accurately predicted my thoughts.

So Why is Cursor a Tool for Non-Developers?

There are many more stories about how well Cursor performs, but let's get back to our main topic. I've explained how Cursor is excellent as a Code Editor. But tools are what you make of them. I use Cursor for broader purposes. It's like a file explorer that can converse with generative AI. Let me explain exactly what's good about using Cursor.

Advantage 1. You Can Use GPT, Claude, and Gemini with Just One Cursor Payment.

In these times when AI models and tools are overflowing, the biggest barrier to entry is surprisingly 'cost.' Since they're usually subscription-based, using just 2-3 $10 tools can easily exceed 50,000 won per month (especially with the current crazy exchange rate), and thinking about it annually, that's 600,000 won.

• GPT Plus: $20
• Claude Pro: $17
• Gemini Advanced: ₩29,000

Setting AI Models in Cursor

• (1) In Cursor's AI chat window, there's an Auto button. Clicking this lets you choose which generative AI model to use.
• (2) If you don't want to specify a model, you can turn on Auto-select. Then Cursor will automatically choose the optimal model to get prompt results.

Cursor Pro costs $20 per month. There's a limitation that high-performance models can only be called 500 times per month, but this is plenty unless you're using AI at an expert level. At the May 2nd exchange rate (1,397 won), using GPT, Claude, and Gemini separately would cost 80,689 won, but using Cursor costs 27,940 won, saving you 52,749 won.

Advantage 2. Using Cursor is Like Having AI Directly Attached to Your Computer's File Explorer.

For example, let's say you asked the GPT desktop app to polish a blog post. GPT can't execute commands like "Save the improved result to ./home/my_folder/blog/YYYY-MM-DD.md on my computer." Simply put, the GPT desktop app doesn't have access to your local files.

On the other hand, Cursor opens your local folder as an Editor, with a generative AI chat window open on the side. It's much smoother for both retrieving and saving files from your local storage.

You Can Save Frequently Used Prompts as .txt Files and Add Them to Your Current Prompt.

The @ button supports not only files but also folder paths and internet links.

• (1) Clicking the @ button lets you specify files to pass to the AI during chat.
• (2) If you have fixed prompt phrases, you can save them as .txt files and specify them using the @ button.

@ = Reference Material for AI = Fixed Prompts

• (1) As shown in the image example, you can write guidelines (SEO-appropriate description tag writing methods) in .txt and use them as fixed prompts.
• (2) Then add your current requirements (what post to write tags for) in the chat window.

Most generative AI services provide this 'provide materials' feature. For example, Claude allows you to create a Project and add files that serve as Knowledge. The difference with Cursor is that if these material files are already organized on your computer, you can use Cursor as if AI were directly attached to your file explorer.

Advantage 3. You Can Easily Experiment with Prompts by Changing AI Models with Just One Option Button.

Combining advantages 1 and 2 means Cursor is a convenient tool for prompt experimentation. Since different models produce different results for the same prompt, there are times when you want to compare results across models. Or you might experiment by slightly modifying prompts to see if you get better results.

For smooth prompt experimentation, you need to (1) be able to easily change models and (2) have convenient version management for prompts and their results.

You could open the GPT desktop app, Claude desktop app, and Gemini in a browser window and move between them to input prompts, but this process is surprisingly cumbersome. Plus, to compare prompt results, you'd need to scroll back and forth or copy and paste them somewhere to save them.

Cursor Saves Well Even Without Being Asked

On the other hand, with Cursor, you can change models with one button click, and you can always type "Save the result to ./results/filename.md" in the chat window. Or you can write it in the fixed prompt.txt file specified with the @ button. Sometimes, as shown in the image above, Cursor even saves things well without being asked. Since Cursor handles saving results automatically, you can really feel the speed increase when you're seriously experimenting with prompts.

Conclusion

Recently, I saw an article advising against immediately using Cursor for vibe coding beginners. There were definitely relatable points. Since I have some self-taught coding experience, I was already familiar with programming editor UIs, which helped me adapt easily. For someone with no development experience, this usage concept might be quite unfamiliar.

However, I think people who are serious about utilizing AI usually have good tool utilization skills or IT familiarity, so they don't need to think of it as too difficult. Personally, as a tool enthusiast (?), discovering unintended uses like this is the most fun moment. Thanks to Cursor, I was able to experience this kind of fun after a long time.

These days, my LinkedIn feed is full of AI and vibe coding. I recently saw a post about effectively using Cursor's rule feature. At first glance, I thought, "This is probably too much for me." Cursor allows you to set AI guidelines that apply to your entire project. For example, after installing Cursor, `Always answer in Korean.` is automatically set as a rule. If @ serves as knowledge, this serves as guidelines.

While it might be difficult to use right away, after seeing this post, I became even more convinced that Cursor is a very convenient prompt editor. When you think about having AI with settable rules and knowledge attached to your file explorer, it's like using a computer with generative AI built-in.

I hope this post can serve as a useful tip for non-developers like me who want to utilize generative AI more thoroughly.