Former ByteDance Engineer: Chinese AI Practitioners Widely Rely on US Agent Tools, Domestic Code Agent "Completely Unusable"

According to Dongcha Beating monitoring, Zhang Chi mentioned in an interview that he uses the Agent mode of Claude Code and GitHub Copilot in his daily routine. His students use Claude Code for training monitoring: automatically check the GPU status, debug and restart automatically upon errors, perform inspections every 30 minutes, and then "we can go to sleep".

When asked if all Chinese AI practitioners are using Claude Code, he first said "not all," indicating that he prefers to use Copilot himself. However, he later confirmed that Chinese AI practitioners generally use American Agent tools. He bluntly stated that the coding Agent for Chinese models is "completely impractical," based on his personal user experience rather than systematic evaluation. ByteDance's overseas team can directly use Cursor, leading him to speculate that ByteDance's future large models may be partially assisted in construction by American Agent tools. Zhang Chi also mentioned a personal dilemma: Claude Code is so easy to use that he is considering "whether he still needs to train Ph.D. students," but he is worried that without training the next generation of researchers, there will be no one to do research in the future.