- Liu Shengyu says AI could soon match his skills at writing and optimizing code
- The DeepSeek engineer says he is not predicting unemployment, but a shift from coder to “Agent pilot”
A DeepSeek engineer has sparked a global debate about AI and jobs after writing that the technology could soon outperform him at the very work he is helping to automate.
Liu Shengyu (刘胜与), a machine-learning systems engineer at DeepSeek, published an essay titled “I Have No Choice but to Bury My Talent in Yesterday” (《我不得不把才华埋葬在昨天》) on September 14.
It quickly topped the trending list of Zhihu, China’s answer to Quora, drew more than 100,000 WeChat views and was translated into multiple languages.
Related posts on X had reached 1.48 million views, according to Liu.
A 2021 graduate of Peking University’s Turing Class, an elite undergraduate honors program in computer science founded in 2017, Liu joined DeepSeek in April 2025.
He works on machine-learning systems and operator design and optimization, and has led development of the main Attention operator for DeepSeek V4.1.
‘Within six months to a year’
Liu said AI had advanced from helping him search documentation and find bugs to independently reading CUDA, PTX and SASS code, analyzing instruction stalls and optimizing operators.
He expects AI to match or surpass his ability to write operators within six months to a year.
That led him to describe what he called a “technology paradox”: the better he makes operators, the faster DeepSeek’s models improve — and the sooner AI could replace him.
But Liu said September 15 that the essay was not an expression of anxiety about unemployment. Rather, he was saying goodbye to the era of writing operators by hand.
He expects engineers to shift from writing code themselves to directing AI agents, becoming what he calls “Agent mecha pilots.”
A view from inside China’s AI race
The essay offers an unusual glimpse into how AI automation is being perceived by engineers at the cutting edge of China’s AI industry — not as an abstract threat to future workers, but as a technology already encroaching on highly specialized technical work.
Liu also criticized Anthropic CEO Dario Amodei’s recent call for “pacing the frontier” in AI, arguing that slowing down while competitors continue to accelerate would amount to “suicide.”
The contrast reflects a broader debate in the global AI race: whether to slow development to manage its risks, or keep pushing forward because falling behind may carry its own risks.
Header image credit: Unsplash
