ByteDance's AI Coding Ambitions: Can Trae Challenge Cursor?
I was chatting with a friend who works on AI at ByteDance yesterday, and he asked me an interesting question: ‘Do you think Trae will make it?’
I said: ‘Depends on how you define making it.’
If ‘making it’ means becoming the second-best AI coding tool in China, I’d say the odds are pretty good. But if ‘making it’ means surpassing Cursor—that’s a completely different story.
So what is Trae? Simply put, it’s ByteDance’s AI programming assistant, built around the concept of an ‘AI-native IDE.’ Not a plugin, not a Copilot-style code completion tool in VS Code, but a complete development environment designed from the ground up for AI.
Sounds cool, right? But here’s the thing—Cursor already did this.
Cursor’s core competitive advantage isn’t ‘AI can write code.’ It’s ‘AI can understand your entire codebase.’ You open a project, Cursor indexes all files in seconds, and when you ask ‘what are the dependencies of this function,’ it gives you an accurate answer.
This ‘global context understanding’ capability is why Cursor is valued at $20 billion.
Where is Trae right now? Honestly, I didn’t get beta access (my ByteDance friend said there’s a waitlist), but from leaked videos online, the basic features are there—code completion, natural language code generation, conversational programming. But Cursor, GitHub Copilot, and even Alibaba’s Tongyi Lingma already have these.
The real question is: What does Trae have that others don’t?
ByteDance has two advantages:
First, data. ByteDance has one of the world’s largest code repositories—Douyin, TikTok, Feishu, Volcano Engine, all massive engineering projects. If Trae can internalize this data to train better code models, that’s genuinely a moat.
Second, user scenarios. ByteDance has enormous developer user bases—Volcano Engine customers, Douyin mini-program developers, Feishu app developers. If all these users are funneled to Trae, user base won’t be a problem.
But both advantages come with risks.
The data risk: internal code and open-source code are completely different beasts. A model trained on Douyin’s code might crush e-commerce systems but fail completely on financial systems. Code model generalization has always been tricky.
The user scenario risk: developer tools aren’t something you can solve with ‘user traffic.’ VS Code succeeded because of its powerful plugin ecosystem. Cursor succeeded because it nailed the product experience from day one. Trying to push an IDE just because you have lots of users—history is littered with failures (remember Alibaba’s Dawn?).
Plus, AI Coding is already a red ocean.
Domestically: Alibaba has Tongyi Lingma, Tencent has CodeBuddy, Baidu has Wenxin Kuaima, Huawei has CodeArts. Internationally: GitHub Copilot dominates, Cursor is growing like crazy, even OpenAI is rumored to be building their own IDE.
ByteDance is entering relatively late in the game.
So why do I still say Trae ‘has a shot’?
Because ByteDance has something others don’t: patience.
Look at ByteDance’s product history—from Douyin to Feishu to Volcano Engine, none were ‘instant hits.’ Douyin was mocked as ‘a Kuaishou clone’ in its early days. Feishu was criticized as ‘too complex, nobody uses it.’ But ByteDance is willing to keep investing, iterating for three years, five years, until the product matures.
AI Coding isn’t a quick business. It requires:
- Massive model training investment
- Deep IDE feature polishing
- Long-term developer community building
All three need time and money. And time and money are exactly what ByteDance has in abundance.
So I told my friend: ‘Whether Trae makes it isn’t about year-one metrics—it’s about whether ByteDance is still willing to invest in year three.’
The biggest risk for big tech making AI tools is ‘kill the project if it doesn’t show results in six months.’ As long as ByteDance avoids that trap, Trae can at least become a ‘useful product.’
As for challenging Cursor—that’s not a product problem, it’s an ecosystem problem. And ecosystems take much longer to build.