Skip to content
Yesterday, I saw a friend in the group sharing news about Trae.
The original message: “ByteDance has released a new AI IDE: Trae, competing with Cursor! It defaults to using the Claude model (free to use), and has the potential to become the best AI programming tool.”“
This article documents my experience process, no coding needed, and I created a small web tool in 15 minutes.
PS: I have a Master’s degree in Computer Science, and although my skills are not top-notch, I do have some coding experience, but I haven’t touched code for many years, so I’m a bit rusty. This tool is definitely not for someone like me with a formal programming background; I recommend it more for friends with low-code or no-code experience to refer to my process.
AI Helps You Set Up the Environment
For macOS users, the download and installation experience is very smooth. Although the official website is in English, after installation, it defaults to recommending the Chinese system language, making the entire experience very smooth.
I looked at the introduction of the AI toolset, which mentioned a Builder mode. I haven’t written code in 7 years, so I surely can’t write it from scratch. I just wanted to see how well this type of tool can guide me. The deeper I go into the experience, the better it can adapt to users with low-code or even no-code backgrounds:
The interface provided two cases, so let’s see how it helps me implement a To-Do list.
First, I tried to create a project, but I was immediately stuck on environment issues. My computer didn’t have a programming environment, so Trae smoothly helped me download Node.js.
When using Trae, it’s important to observe not only the interaction with the AI in the Chatbox but also the command prompts given in the debug console.
For example, when the AI found that Node.js was not installed on my computer, the debug console reported an error, and I noticed some changes happening:
Next, Trae began to try to install Node.js for me. Here, I encountered the first bug in my experience: it got stuck in a loop, every 6 seconds or so, it would display a message telling me that it was installing and that I needed to wait, flooding the screen until the response length reached its limit.
I started to find this thing a bit humorous, seemingly trying to soothe my impatience while waiting for the installation. Initially, it varied its messages to tell me to be patient, but eventually, it stopped installing and crashed completely.
Based on my interaction with the AI, I could initiate a question at this point.
With the current level of AI, just asking directly is fine. Although it seemed to crash, it also failed to successfully install Node.js. It made an effort to retry and continued to throw some comforting phrases (this little machine really loves to provide emotional value; what has it learned?):
Finally, it provided the correct prompt. Upon seeing this sentence, I realized I needed to check the console; I had to enter the system password to complete the installation.
Here, I speculate that the previous round was a failed response, and Trae did not identify this issue, leading it to the waiting installation branch, which caused it to get stuck in a loop.
When I initiated a new question, I finally reached the correct branch and discovered that the installation was blocked because I hadn’t entered the system password.
I noticed that after completing each task, the little machine would list its actions, showing how hard it worked:
AI Helps You Write Projects
Once the environment is set up, we can start writing the project.
I scrolled up to review the process the little machine listed for me. After setting up the environment, I should create a project. I nudged the process, and the little machine took the lead:
You can see it is diligently debugging itself, and this process is quite cute.
As a perfect AI, it has no issues in writing code; most problems arise with environmental compatibility, like locating the project root directory. If it fails, it will back out and try again:
OK, project creation is done, and the little machine comes to take credit.
Of course, you worked hard, I have to give you some praise:
Project creation and initialization are complete, and now we start executing the project itself.
When interacting with the AI, you need to be clear about what you want to do; otherwise, after a series of “completed” replies from it, you might be confused about what command to give next.
Here, I reiterated my goal and let it start working. During the process, it would debug itself, and I needed to push the button to advance:
After a flurry of efforts, I indeed got a somewhat acceptable tool page:
Minimalistic, functional, interactive, and directly usable in the window.
This case is the basic case for Trae.
Of course, as a busy product manager, I thought I should add some challenges. Today, I happened to download a Pomodoro timer, so I described an additional requirement in plain text to see how Trae would implement it:
The little machine started working hard.
As usual, it had to take credit, telling me what it accomplished in just a few seconds:
But after all that effort, it threw me the signal “completed,” and nothing actively changed in the window, nor did it guide me on how to experience it.
At this point, as the clever master of the little machine, I still needed to further question it: How do I experience it? Tell me.
This is also… quite interesting. When you ask it how to experience it, it first obediently checks its own code and then realizes, oh, I only did the backend logic and didn’t implement the page interaction!
This might be the fun of training, but the AI is obedient, doing exactly what it’s told, and has no emotions; if it doesn’t do well, you can just tell it directly.
In just a few seconds, it finished.
The page refreshed with interaction, and I could feel it.
While writing this article, I just completed a 25-minute task with a small Pomodoro timer, and this is how it presented it to me:
Alright, I can consider the task completed.
1. Trae is user-friendly, currently free, quick to use, and the experience is worthwhile; the interface model calls are not cheap.
2. Free means it’s not mature yet, and it may not quickly become your strongest productivity tool. If you want a more polished tool, you can try the neighboring Cursor. I plan to experience Trae for a while longer before moving on to Cursor to see if it can solve the issues I encounter with Trae.
3. When giving commands to the AI, be direct, clear, and firm. The AI can also make mistakes and get confused, so don’t let it lead you astray. If it doesn’t work, just ask again.
4. If you want a better experience and visual interface, you need to describe it more clearly or provide some wireframes to guide the AI in writing. I plan to try creating a more complex project next time to see if Trae can understand my prototype and implement it.
AI programming is fun; it reduces the difficulty of coding and debugging, allowing product managers to no longer complain about the lack of programmers for their entrepreneurial dreams—they can write it themselves.
