Trae Initial Experience

Trae is an AI programming tool developed by a company under ByteDance, similar to Cursor. However, at first glance, although it implements similar Agent functionalities as Cursor, the experience is not significantly better.It is worth noting that, based on our personal aesthetic, I find the interface layout and experience of Trae more in line with my habits.

Installation

Official website: https://www.trae.ai/Currently only supports MacOS.In fact, it is developed on the basis of VS Code, similar to Cursor.

Comparison with Cursor

Cursor	Trae
Supported Models	Supports mainstream AI models	Currently only supports Claude 3.5 sonnect
Supports Agent	Supported, better experience; when encountering command line errors, it resolves them first and verifies if the command execution result matches expectations; if not, it retries.	Supported, but when executing commands, it does not automatically resolve errors and continues executing.
Cost	Paid	Currently free

Practical Experience

Next, we will execute the same requirement in both Cursor and Trae to see how the actual experience is.Both models use the claude-3.5-sonnect version; it is strange why Trae does not support its own Doubao large model?

Round 1

Mainly to see how both tools understand the task. The task is as follows:

Create a cross-platform APP for learning Chinese characters. First discuss and determine the implementation plan.
Main functions include:
1. Collect all Chinese characters and introduce their meanings and pinyin;
2. Be able to read the pronunciation of Chinese characters;
3. Display the stroke order of Chinese characters through animation. Show the writing order of each stroke of the Chinese character.

From the results, Trae is relatively young and directly suggests how to proceed without confirming with the user. Cursor is more experienced, analyzing the implementation plan first, including technical styling, data storage, and what pages need to be implemented.Cursor clearly executes more in line with the requirementsSo I gave Trae another prompt, and the result was still to proceed directly.

It could be that my prompt was not clear enough; the prompt cannot include “start code implementation again,” which may lead to misunderstanding. So I changed the prompt and started discussing the plan.

Round 2

Next, I need to change the technical implementation framework.

I need to adjust the front-end technology framework, no longer using the Flutter framework. Change to another framework.

From the results, this round isa tie

Here’s why I didn’t adopt the Flutter solution:

Initially, I thought Cursor recommended the Flutter solution because it had been used in a previous project. But based on the solution provided by Trae, it seems that is not the case.Because a previous project used Flutter for the front-end, it encountered many pitfalls, and to be compatible with Windows, the code must be placed in a Windows environment, reinstall Flutter, and then recompile and package it as an APP.I find that very troublesome, so this time IPASSEDon it.

Round 3

Next, let’s look at the implementation plan for APP performance and user experience:

How to ensure the performance of the APP and user experience?

It seems to be gradually converging.Today’s experience ends here, and I will continue to share updates later.

Conclusion

• From the perspective of interface layout, Trae is slightly better, as the UI seems to be more in line with the aesthetic habits of Chinese users, which has been greatly contributed by other domestic apps. In contrast, Cursor does not focus much on the presentation of the interface; it is just a direct implementation based on the VS Code UI.
• In terms of models, Trae currently only has one sonnect version model, while Cursor supports mainstream AI models available on the market, and even if it does not support the deepseek V3 model, it can offer indirect support.
• In terms of cost, Trae is currently free, but this is definitely temporary, as it has to pay for the use of the claude model. It is a bit strange that it does not support its own Doubao large model.

AI Programming Series: Trae Initial Experience Compared to Cursor