A DeepSeek engineer says AI will probably match or surpass the kernels he writes by hand

The substitution Liu describes happens inside the job. He wrote that in about a year, AI went from a tool that helped him search documentation and spot bugs to one capable of reading GPU code and optimizing the software operators ^[4]. He said he is "well aware" that the operators written by AI "will most likely be as good as mine, or even surpass mine" ^[5]. What he expects to lose is the hands-on part, becoming someone who mainly steers AI agents rather than writing operators himself ^[8].

His stated reason for continuing applies to his employer as much as to him. Liu wrote that even if he lay flat, or deliberately threw a wrench in the works to slow down model training, rival companies would keep advancing their models and he would still end up being replaced all the same ^[6]. He gave a second, simpler reason: "writing kernels feels like playing a video game to me. I get an enormous thrill from it," he wrote ^[7].

One post from one engineer is not evidence of a retention problem. Liu said he does not think he will be out of work ^[8], and he declined to comment beyond confirming to Business Insider that he wrote the post ^[3]. The post documents a cost a pay rise cannot fix. He compared the change to a talented knitter who gives up making sweaters by hand and uses a faster machine because everyone else in the trade has one ^[9], and wrote: "But the pleasure of listening to the rain by the window, threading a needle, and slowly passing the time was ultimately crushed by the roar of the machines" ^[10].

The two people named in the account went in different directions. Jacob Coxon, a former researcher at Anthropic and OpenAI, announced last week in a post on X that he had quit Anthropic over AI safety concerns ^[15]. Liu stayed, and described staying as a choice: he wrote that he would rather not be overthrown by the revolution, but that if revolution is inevitable he would rather be the one to revolutionize himself ^[16]. Coxon's stated reason was safety. Liu did not resign ^[17].

The post also makes a control argument that has nothing to do with his employment. Liu warned against frontier AI being held by a handful of companies, and wrote that this could widen inequality by giving those with access to the most capable models an even greater advantage ^[12]. "Therefore, I still believe that cutting-edge intelligence should be available to everyone in an open and affordable way," he wrote ^[13]. He then named rivals: "I don't trust Anthropic or OpenAI to do this, and I especially don't want Anthropic to master the most advanced artificial intelligence or AGI; to exaggerate, the seriousness of that would be no less than Hitler acquiring atomic bomb technology before the Allies," he wrote ^[14].

The only interval in the post is that year, from documentation search to operator optimization ^[4]. Liu does not date the replacement he anticipates. That leaves engineering managers a narrower question this quarter than the one the essay poses: whether hand-written kernel work stays on the roadmap for anyone, given that the people who can check an agent's operator today are the people who wrote operators themselves ^[8].