Recently, a tool called Cursor has become very popular in the AI community. There’s even an 8-year-old girl using it to write programs. After seeing that video, I thought I’d give it a try myself. I fell in love with it after the first try—it’s really amazing. You don’t need to write code at all; just use natural language and the concept of building blocks to easily create your own applications. All the features are ones you use in your daily life, so you don’t have to open a bunch of tools.
More importantly, registering for Cursor now gives you a free 2-week Pro trial with no credit card required, including 500 Premium model uses. Even after the Pro trial ends, you can still use it for free, just with some limitations.
Cursor lets you build your own apps even if you can’t code.
Cursor can be thought of as a free tool that combines a code editor with major AI models. Currently, it supports the following models:
- GPT-4o（Premium model）
- Cursor-Small
- GPT-3.5（Premium model）
- Claude-3.5-sonnet（Premium model）
- GPT-4o-mini
The free tier now includes a 2-week Pro trial with no credit card required, meaning for 2 weeks after signing up, you’ll have access to Pro benefits. The biggest perk of Pro is 500 free uses of Premium models, so you can use GPT-4o or Claude-3.5-sonnet up to 500 times.
What does 500 times even mean? I’ve already made nearly 10 tools, and I still have over 100 tries left. In between, there were quite a lot of debugging and bug-fixing conversations, yet I still have this many remaining.
Although the interface is only in English, you can chat with the AI in Chinese, so it’s not too hard to get the hang of.
Moreover, there are many things Cursor can do; this article focuses on how to quickly build your own custom tools.
User interface
After installation, opening it will show a screen like this. On the left is the list of all files, and you can also create project folders. Hover the mouse over it, and you can choose to create a new project folder or a new file:
After setting up the project and files, press Control + L to bring up the AI chat window, located on the right side. By default, it will select your current document, meaning you can directly ask it anything about the content of this document:
Newly opened files are all in .txt plain text format, and you can later save them as other formats. Cursor supports all code-related formats, such as HTML, bat, Java, PHP, PY, and so on. It just depends on which file format the AI tells you at the time, and you save it accordingly.
Expand below to switch to other models. I almost always use Clause-3.5-sonnet, but you can also switch to GPT-4o—it’s up to your preference:
Alright, you just need to keep those key points in mind, and you can start telling the AI about the tool you want to build.
There are a few key points to keep in mind when making it; here they are for your reference:
- Please use a building-block approach: finish one feature first, then do another. It’s not recommended to ask the AI to build everything at once.
- After finishing each build, be sure to quickly try it out to see if there are any obvious bugs, such as features not functioning properly.
- Don’t worry about running into bugs — AI will help solve them.
- Every feature can be enhanced later, so you don’t need to get it perfect in one go—just build what you’ve thought of first.
- Every time AI responds, the answer is only the content that needs to be modified. You can simply click Apply in the top-right corner to apply it, without having to find and modify it yourself.
- Let your imagination run wild—many things you think you can’t do are actually possible.
- Some features can eventually be combined into one. For example, I first build a file conversion feature, then image cropping, and finally ask the AI to combine these two features. This is feasible.
How I created 5 commonly used tools in one hour
Because of my work, I often need to use image editing features, such as reducing the long edge of an image, adding a white border, adding a watermark, and converting to WebP. Previously, I used Lightroom and Photoshop.
Although I’m used to it now and the speed isn’t too slow, the operations are quite tedious, so I’ve been thinking about whether these editing actions can be completed directly with a single click.
Initially, I’ll start with the simplest one—requesting a converter for common image formats to WebP. Here are all the descriptions I gave:
製作一個能將常見圖片格式轉 Webp 的工具，支援拖曳圖片匯入或手動點擊檔案上傳二種方式，圖片匯入後就會立即開始轉檔 Webp，接著儲存並覆蓋原本的檔案。可批次處理，一次處理多個檔案
Within a few seconds, the AI gives me a large block of code. Once it finishes writing, you just need to click the Accept button at the top right of the code, and it will automatically be imported into the current file (remember to make sure that your current file is selected above the AI chat input field).
As shown in the figure below, and since this code is HTML, I saved this document as .html, and after saving, I opened the file with a browser:
But unfortunately, the first attempt failed. I dragged in a PNG to convert to WebP, and it showed conversion failed:
At this point I went back to the AI and said, “After I tried it, it showed conversion failed, JPG’.” It then immediately checked and gave me the code that needed to be modified. Similarly, after waiting for the AI to input everything, just click Apply at the top right. It will automatically detect the document content, find the parts to modify, delete the erroneous ones, and add new ones. During the generation process, “Generating” will appear at the top right of the AI. Remember to wait until everything finishes running; only then is the entire check complete. The green sections are newly added, and the red sections are deletions. After generation finishes, click Apply at the top right again to fully apply it. You can also accept modifications section by section:
After applying, the red and green blocks will disappear, at which point you can save the file and test again:
The second test worked properly, so I started requesting the second feature. Below is changing the image’s longer side to 750px:
加入第二項功能，可在右上方切換。第二項功能是「修改圖片長邊」，會自動把把圖片的長邊改成 750px，另一邊根據原來的長寬比來修改，修改好後自動儲存，一樣支援拖曳和手動上傳
And after testing, I found that if the long edge is below 750px, it gets scaled up to 750px, so I added another condition for it:
修改圖片長邊加入一個條件，如果長邊小於 750px，那就不做任何動作，直接儲存
Currently both functions are working normally, but I don’t really like the design of the function menu, so I requested to change it to buttons:
切換功能的選單，幫我改成按鈕，切換之後該功能主頁的標題也會跟著變
Although it was indeed changed to a button, this single-button toggle isn’t what I want, so I’ll give it new conditions again:
按鈕幫我各別獨立，按下之後，該按鈕會亮起
This is the key point I mentioned earlier—you don’t have to get it right all at once; it’s fine to make gradual changes step by step.
Indeed, it became two buttons as I requested, with the currently active one lit up. However, I didn’t really like the position at the top right, so I later changed that part to the center:
Next, I also need another tool that modifies the long side to 1000px, so I asked to directly copy the original 750px one, change the long side to 1000px, and then change the button name:
修改圖片長邊名字改成「修改圖片長邊（750px）」，再幫我製作一個新工具，內容是「修改圖片長邊（1000px）」，自動把圖片長邊改成 1000px，低於 1000px 的圖片不需要修改任何內容，直接儲存
This feature is very easy, no problems at all, so I’m requesting other new features.
At first, I wasn’t sure if HTML could add white borders to all four sides of an image, so I asked about it. The AI said no problem and gave me the code:
你可以做到一個工具是，幫任何圖片四個邊加上 1px 的白邊嗎？如果可以，請添加
After I tested and it worked normally, I continued to request new tools,
As for the next watermark, I’m not sure whether it can be done, or whether the position can be specified — but the AI’s answer was affirmative:
那你可以為圖片加上浮水印嗎？位於每張圖片右下角，距離邊際 20px 的位置，並且我可以指定 Logo，Logo 也會透明化 50%
Even this feature that calculates whether the Logo should be enlarged based on the current image ratio, and remembers the last specified Logo image, is no problem:
Logo 自動幫我根據圖片比例，放大成圖片的 1/20，並且會記住我上次指定的 Logo 圖片
Of course, you will inevitably run into bugs, features suddenly disappearing, or the layout breaking. When that happens, don’t panic—just ask the AI to check, or even ask it to give you the fixed code or the full code from the previous step.
Once all the above tools were ready, I started trying to combine them all to create an exclusive “Computer King Ada” feature that automates every action, making it more convenient for me to use later. The AI again said no problem and provided me with the code:
製作一個新功能，名稱是「電腦王阿達專用」，把修改圖片長邊（1000px）、添加 1px 白邊、添加浮水印功能全部結合，這可以做到嗎
Afterwards, adding the conversion to WebP format for 電腦王阿達專用 also succeeded:
And these are the 5 commonly used tools I made within an hour, including a tool that automates all the actions:
After that, I didn’t stop either. Whenever I thought of a feature I needed, I kept adding it. For example, I added a side-by-side screenshot feature that lets you arrange multiple screenshots together with proper spacing, and also includes editing tools like rectangles, arrows, and mosaic blur. For all of this, I had no understanding of the code at all—it was all done with AI:
Afterwards, I also used this tool to create “PDF to Word format”, “track article keyword rankings in Google Search”, “automatically convert a URL into Jina AI’s Reader API format, and open the webpage to save as PDF”, and so on.
Ask whatever comes to mind, and you can make almost anything—just with different combinations. Some things require Python to implement. If you don’t know how to use Python, that’s fine—just ask AI, and it will guide you through every step, including which application to open and what commands to enter, all very clearly.
Additionally, I also noticed that it’s not recommended for each tool’s code to exceed 700 lines; if it does, the AI will almost always make mistakes when applying and modifying the current file’s code.
The biggest difference from AI chatbots like ChatGPT is that Cursor can read the content of the current file at any time and automatically generate modifications.
Finally, here’s also a video of an 8-year-old girl using Cursor to create her own app:
Source: KOCPC Chinese