Happy new year, everyone!
I planned to make great progress during the Spring Festival, but it was so cold on this “detour” that I caught a cold. I’m still suffering from acute bronchitis today. I’m really speechless!!
However, I still mustered up the energy to optimize two of my projects using ByteDance’s Trae AI editor (https://www.trae.ai/?utm_source=juejin&utm_medium=juejin_trae&utm_campaign=techcall):
tinify — webp: It’s a terminal tool used for compressing images. By default, it converts images to the webp format, and you can also specify other image formats. The test result is amazing! A picture of more than 500 KB was compressed to only 45 KB!
md-note: This is a note-taking app I developed in 2023. Currently, I’m the only one using it. — It supports the What You See Is What You Get (WYSIWYG) feature for Markdown editing. — It enables one-click copying and pasting of articles to WeChat Official Accounts, Zhihu, and Juejin. — It allows you to insert images and videos. When inserting images, it automatically compresses and transcodes them. — It can be integrated with some open-source static site blogging systems, such as Hexo and Hugo, to achieve one-click blog publishing. — It integrates an AI feature to generate article descriptions.
The tinify — webp tool has been optimized and published on npm. If you’re interested, give it a try. As mentioned above, its image compression effect is truly amazing! My note — taking project has now met the standards for going live. However, I still want to containerize it, so the official introduction of it will be postponed by two days. So, in this sharing, I’ll mainly talk about two aspects: How good is Trae? How can DeepSeek be combined with AI — powered programming?
How good is Trae? Since the launch of Trae (https://www.trae.ai/), it has received both praise and criticism. Some people acclaim it as the shining star among domestic AI editors, claiming it outperforms Cursor and Windsurf. On the contrary, some people scold its design as extremely poor, complaining about bad user experiences here and there. These two viewpoints are the most extreme ones I’ve seen in related articles. I doubt whether those people have actually used Trae to work on real projects. I’m pretty sure most of them haven’t, because those who have truly used it know that Trae is good, but it hasn’t reached the level of Cursor and Windsurf yet. It does have its flaws, but it’s far from being as bad as they described. It seems that ever since the emergence of AI, exaggerated praise and slander have become all too common. I hope everyone can see clearly and not be misled by such information. During the holiday, I spent two days using Trae to refactor my previous note — taking project. The actual experience was remarkable: — I used the Builder to create a Vue3 project from scratch. — I redrew the UI pages and wrote common components and methods according to the previous project design. — I created a store to manage common variables among pages uniformly. — I re — encapsulated axios based on the APIs provided by the server and then made a unified plan for the APIs on the front — end. — It also handled the front — end and back — end joint debugging. — On the server — side, I optimized and rewrote the interaction with Github. — I added an AI — powered feature to generate article summaries. Combined with the popular DeepSeek during the Spring Festival, it only took two days to completely renovate my note — taking project, and the whole process cost me nothing! Trae definitely deserves to be my top “employee” during the Spring Festival. Besides Cursor, using Trae to access Claude — 3.5 — Sonnet for free is an incredibly pleasant experience! First of all, in terms of usability, it can understand what you want even when you input in Chinese. You don’t have to translate your ideas into English to communicate with it. Secondly, Trae supports specifying files, folders, or a specific section of code. In addition, you can upload pictures to make your requirements clearer to Trae, which is similar to dev0. Although there is no cursorrule — like feature, the coding process is still very smooth. Most importantly, as of now, Claude — 3.5 — Sonnet can be used for free. It’s like having a top — notch programming expert at your disposal. As long as your requirements are clear, it can implement them without charging you a single cent! Finally, I successfully refactored a project using Trae without encountering any glitches. Whenever there was a problem, Trae could promptly solve and fix it, which proves its capabilities as an AI editor. ### DeepSeek as a Solution Designer There isn’t an advanced way to combine DeepSeek and Trae as many people might expect. In fact, currently, Trae doesn’t even support DeepSeek. At first, I personally guided my “employee” Trae, but the efficiency was low. After DeepSeek became popular, I “recruited” it to assist with some extended tasks, such as: — Letting DeepSeek refine my ideas. — Having it clearly and systematically outline how to meet a certain requirement. — Finally, I sent the organized content to Trae for implementation. Since then, the efficiency has improved significantly. Handing over the AI — generated output to another AI for implementation shows a high level of synergy between them. We have to marvel at the capabilities of DeepSeek and Claude — 3.5 — Sonnet (the built — in model of Trae). They truly are among the top large — scale models in the AI field. DeepSeek can really analyze your ideas thoroughly and provide a reasonable solution. After I send this solution to Trae, all I need to do is verification. Throughout the process, I only act as the “boss” who makes decisions and conducts tests, while the rest of the work is done by DeepSeek and Trae. However, I’m still not entirely satisfied because the development process still relies heavily on human factors. The ideal state I envision is to have a workflow to complete the entire output process: — Provide an initial product design demo to DeepSeek. — Then DeepSeek enriches it and generates a usable Product Requirement Document (PRD). — Based on the PRD, it generates product design drafts. — Trae generates project code according to the product design drafts and PRD and conducts unit tests on the code. — Finally, humans test the project, find bugs, and hand them over to Trae for repair. — After confirmation, the project is launched and delivered.
If the development can follow this process, the efficiency of developing a product will be significantly improved. Of course, the assistance of professionals is still indispensable in this process. It would be quite difficult for complete beginners. That’s all about what I’ve been doing and my thoughts recently. If it’s helpful to you, please interact more!
T
AI
Top comments (0)