The Biggest Innovation of Cursor Is Not AI but IDE

Recently, a new product in the AI programming field, Cursor AI, has sparked heated discussions. I searched online for evaluations of Cursor AI using Mita AI, and overall, it is an advanced AI programming assistant tool that can enhance developers’ efficiency through intelligent code generation and editing. In terms of performance, it excels, achieving industry-leading levels in startup speed, response time, and stability. Below is a mind map provided by Mita AI for reference.

However, today I want to discuss that after extensive use and experience, I believe the biggest innovation of Cursor is not AI, but the IDE.

| Previous AIGC Programming Products Before Cursor

Since the launch of ChatGPT, many AIGC-assisted programming products have emerged, such as GitHub’s Copilot, which was released early on. Following that, domestic AI giants have launched countless AI programming products, such as SenseTime’s Code Little Panda, Kunlun Wanwei’s SkyCode, ByteDance’s MarsCode, Baidu’s Wenxin Quick Code (Baidu Comate), Alibaba’s Tongyi Lingma, Tsinghua Zhipu’s CodeGeeX, iFlyTek’s iFlyCode, etc. Additionally, there is a useful AI programming product from a non-major company called Fitten Code by Beijing Feishi Technology.

Since VS Code is the most widely used IDE globally, everyone has chosen to release their products in the form of VS Code plugins. This means that you can directly search and install them in the plugin section of VS Code.

Then, it provides real-time programming assistance in the form of a sidebar, along with real-time code prediction prompts that follow the programmer.

I have used all the AI programming products mentioned above, and I feel that providing AI programming assistance through the VS Code plugin method can greatly assist our programming process and significantly increase our programming speed. Since we are already very familiar with the operation of VS Code, we can essentially use AI programming with no learning cost.

In practice, the accuracy of code prediction is generally not too bad; what affects the user experience more is the speed of code prediction and answering questions. Foreign AI programming products are definitely slower, while domestic ones are much faster. Among these domestic products, Fitten Code from Feishi Technology is the fastest.

However, Cursor AI takes a different approach by releasing its own programming IDE instead of as a VS Code plugin. This is quite strange. To experience the super capabilities of Cursor AI, I have to download and install their IDE, which raises the threshold, doesn’t it?

Or is it that what Cursor AI aims to achieve is difficult to accomplish through a VS Code plugin? With these doubts, I deeply explored the effects of Cursor IDE + AI and finally had an epiphany!

| Recommendations for Modifying Original Code

When Cursor recommends how to modify the code, it cleverly processes the unchanged parts in this way. There are two benefits: first, it makes the response clearer and more concise; second, it cleverly speeds up the response time. Other AI programming tools output the original code again in the response. This is indeed an innovation of Cursor!

| Suggestions for New Code in Multiple Files

If you need to add code in multiple files to achieve your goal, AI tools in the form of VS Code plugins can only tell you in their responses where and which files to modify. You still have to rely on yourself to find these places and manually insert the changes.

Cursor, however, is quite impressive; it directly lists the specific modifications for each file. As you can see, the left side shows the names of the files to be modified, and below are the new codes to be added. The key point is that the upper right corner provides an Apply function, which allows direct modification of the target file. The IDE will automatically open this file to apply the change. This is indeed much more convenient and efficient.

| Discussion and Optimization of Code Snippets in Answers

Another commendable aspect is that Cursor provides the ability to Ask again in the upper right corner of the recommended code, allowing for further discussion on this code to achieve optimization.

Previously, all AI tools could not discuss and confirm a specific piece of code in the response; they could only helplessly apply the entire code from the current response to the code file and then select that piece of code for new discussions.

Cursor fully supports this situation, which is actually encountered quite often. You just need to click Ask to discuss and optimize confirmation for this piece of code in the response, significantly speeding up the integration of AI recommendations and our programming experience.

| Change Tracking

Another highlight is that Cursor IDE has Git enabled by default to track code changes. If changes are made based on the code recommended by Cursor, there will be automatic markings, even if the current project has not been initialized with git.

At this point, I suddenly understand why Cursor AI wants to create its own IDE. These more detailed functions require deep customization of the IDE, which cannot be achieved with a VS Code plugin.

| Deleting Code

This might be Cursor’s trump card! It outshines VS Code plugin AI tools!

All VS Code plugin AI tools can only tell you where to delete certain code when recommending deletions. You need to open these files yourself, find the corresponding locations, and delete the code segments indicated in the answers. This means that in this regard, AI can only guide you, and you have to do it step by step.

Cursor has made a revolutionary innovation in this scenario. Why? Because Cursor directly lists the answers according to the files to be modified, and this UI is very clear about which files should have deletions. The key point is that each file provides an Apply option in the upper right corner to directly handle these deletion actions.

After clicking Apply, it first highlights the differences between the old and new code, providing Reject and Accept options to make the final decision.

Have you noticed that this operation is very similar to merging code? This type of merge interaction is very familiar to all programmers, and this new code is provided by AI, which perfectly solves the interaction problem of how AI recommended code can quickly combine with programmers’ experience and judgment.

I have to say that this is indeed the biggest innovation of Cursor IDE!

To give you a deeper understanding of this innovative advantage, let’s look at how other VS Code plugin AI programming tools handle code modifications. They can only provide the modified results. Can you quickly see what changes the AI made to the Home() code in the interface below? Isn’t it very tedious?

The upper right corner of the AI suggestion result provides an Insert button. You have to select the content on the right that is to be replaced before you can click Insert; otherwise, it will directly insert it as new content instead of replacing the source code.

After clicking Insert, the AI recommended code directly replaces the original code, but can you see exactly what changes were made?

This is a common problem faced by all AI tools based on VS Code plugins. Therefore, when we use these AI programming tools to fix or optimize code, most of the time we still need to carefully check the AI recommended code to find the key points and manually copy them to the correct place in the source code.

In contrast, Cursor IDE effectively solves this problem, allowing us to quickly advance code modifications through a merge approach. This is indeed a very valuable and significant innovation point.

The following example shows how Cursor IDE, based on a Next.js scaffold project, creates a gradient background stage and a 3D gradient sphere through dialogue with the AI. Here I only post one image, but it is actually a continuously changing gradient background with a rotating 3D sphere with light and shadow. As we all know, gradients, animations, and light effects require a lot of CSS code to achieve. By conversing with Cursor, allowing AI to handle these tedious tasks, it was indeed a pleasant process.

I am an AI productivity mentor~ Teacher Liao. A master’s graduate from the University of Electronic Science and Technology, with twenty years of software development and management experience, and a three-time internet entrepreneur, proficient in AIGC large model principles and algorithms, familiar with various AI applications and skills. Currently focusing on AI’s new quality productivity, spreading knowledge and skills to make AI functional, smooth, and utilized to the extreme, empowering millions of workers to turn AI into workplace and career productivity!

If you want to learn AI efficiently or discuss AI-related questions, feel free to add my personal WeChat on the left side of the image below, and note “AIGC”.