DeepSeek Engineer Says AI Will Surpass His Coding Skills Within a Year

Shengyu Liu, a machine learning systems engineer at DeepSeek, said in a recent article on his personal WeChat account that AI-written operators will likely match or surpass his coding ability within six to twelve months.

Liu, a member of Peking University's 2021 Turing Class, developed the main attention operator for DeepSeek's newly released V4.1, a core component that directly determines inference efficiency. He acknowledged that AI will soon be able to produce operators at least as good as his own.

He described a "technological paradox": the better operators he writes, the faster DeepSeek's models train and infer, the quicker the models evolve, and the sooner he could be replaced. He said AI can think 300 tokens per second, write a command in half a second, and finish code in 20 seconds.

His comments reflect broader adoption of AI coding tools. According to Sonar's 2026 Developer Survey, 72 percent of developers who use AI coding tools use them daily, and 42 percent of submitted code is AI-generated or heavily AI-assisted, up from 6 percent in 2023.

At a press conference held by China's Ministry of Industry and Information Technology, China Software Industry Association Secretary-General Lv Weifeng said pure coding skills will become less important, while requirements analysis, architecture design, system integration, and quality assurance will become more critical.