Cursor saves me 90% of my time

The thing is, half a year ago, I already knew about the emergence of AI, and I occasionally used products such as Kimi and Ali Qianwen. They were all mediocre, at most adding icing on the cake to my work, and even many AIs were a bit stupid.

Until I used Cursor, my view on AI changed completely. After I used it in depth and developed one application after another, what shocked me most was its large engineering capabilities.

For example, if I want to develop an application now, I tell him my ideas and needs. He will not only analyze the description, but also conduct a comprehensive analysis based on the content of existing projects.

This is not enough to show its power. Because what I just described is more like the common use of AI for drawing pictures, writing texts, and taking notes. I just need to tell it what I need? What do I need it to do for me? Or help me find and organize something. Just tell it roughly, and it can do something.

However, Curosr focuses on programming. At first, we thought that the emergence of this thing would replace many programmers. For example, a product manager only needs to describe his needs, and Cursor can help him develop the product. But in actual use, it is obviously unrealistic.

How to use step-by-step verification

The problem with AI programming is that in the early stages, you can easily and quickly describe your ideas to AI, and it will quickly build a variety of functions based on your ideas. We can ignore its degree of restoration for now. But when the construction is completed, there will be a big problem, such as some complex bugs appearing during the software operation. At this time, if you ask AI to search and fix the problem, it will most likely modify the existing code or function, causing more strange problems.

I once encountered a very strange bug. At that time, I used the cursor to help me fix it. I kept optimizing its prompt words, that is, the description of the dialogue. The description of this dialogue is not only a description of some ideas or requirements, but also a lot of engineering code and logic description. In short, it is very complicated, but it still cannot fix the bug.

What I did at that time was to calm down and read the code it wrote carefully to sort out the logic, which actually helped it find the problem. When I found the problem, I described it to it, and it realized that it was a problem. Before that, it just kept adding or deleting code while fixing the problem, and didn't see the problem.

Although it has strong engineering capabilities, it still has many loopholes in the details. In summary, in the early stage of use, you should verify its code step by step, and it is best to constantly understand the logic of its code writing. If you leave it alone, it will eventually get out of control.

Significant efficiency improvement

This efficiency improvement is reflected in the use of it to do some repetitive, obscure, and troublesome things. For example, quickly set up a network request, quickly build a UI model, these are very simple. According to actual tests, the efficiency improvement is at least 50%. The team members are very powerful. The best ones can use Cursor 90% of the time and write high-quality code.

Learn to describe and ask questions

During the period of using Cursor, our team has communicated many times. The key to using Cursor well or not is how to communicate, or more complicatedly, how to describe your problem. This is actually very good. We can understand the requirements or problems first, and then form a description that the machine can understand better. This step completes the in-depth understanding of the requirements.