Luo Yonghao sells AI coding platform TRAE on a Double 11 livestream for 9.9 yuan

On November 11, an AI coding tool turned up in a shopping livestream. TRAE, billed as a full-chain AI development platform, was listed in Luo Yonghao's Double 11 broadcast at 9.9 yuan for the first month, with the host projecting his screen and running demos in front of tens of thousands of viewers.

His framing was built for reach: he said he cannot write a single line of code, that building software used to require assembling designers, product managers and engineers with painful coordination costs, and that this tool would not flinch even after a hundred revisions. The claim had weight because everything that followed happened live, with no room for editing.

The first demo was a short-drama market report. Asked to analyze the Chinese short-drama market from January through October, TRAE produced a plain-text draft in under four minutes. Luo then played the demanding boss, ordering it into Feishu with pie charts where pie charts fit and bar charts where they did not. His laptop froze mid-task; after rebooting he simply told the tool the machine had crashed and to continue, and it resumed and delivered a formatted document with charts minutes later.

The second demo was long-document rewriting. He fed it a document of more than 50,000 words and asked for two deck versions: a speaker edition with only key points, diagrams and cases on each slide and the detailed reasoning pushed into speaker notes, and a self-reading edition that stands on its own. He also demanded it run quietly in the background without seizing the interface, reformat to a 21:9 widescreen ratio and switch to a blue theme. Both files came back as specified.

The third demo was the most interesting. Because his mouse is nicknamed the "steamed bun," he wanted to put its idle middle scroll wheel to work: hold the button in any application, get a floating window, dictate text, then drag to a punctuation mark or the send button to finish input. He handed over a barely legible hand-drawn wireframe, and TRAE not only reconstructed the interaction but also broke out button functions, failure fallbacks and cross-application coverage. The resulting local app works in browsers, chat windows and Feishu, and the project can be exported for further development.

What makes that possible is engineering packaging rather than model magic: multi-file awareness, automatic environment configuration and local code orchestration underneath. The chores that scare off non-programmers, setting up environments, installing dependencies, scaffolding projects and fixing configuration errors, are handled silently by the system, leaving the user to state intent.

The user numbers explain why a vendor would put this on a livestream shelf. According to QbitAI, TRAE had more than 10 million registered users worldwide and 1.6 million monthly actives as of June, with a substantial share opening it nearly five days a week. That is no longer a tool that only serves professional developers.

The broader context sharpens the signal. Gartner's 2026 generative AI maturity curve shows the market past the concept-inflation stage and into results-driven adoption, and GitHub's State of Octoverse 2026 report says more than 40% of pull requests on the platform are now opened by AI agents. In another livestream the same day, the product on sale was token credit priced by volume; Luo was selling a finished tool. The contrast sits exactly on the line between selling conversation and selling outcomes. The open questions are whether 9.9-yuan newcomers are still paying in month two, and whether a livestream channel can genuinely convert non-programmers into subscribers.