How Does It Feel to Program Your Own Demise?
Gustav Gillund
October 6, 2026
Image: studio thongden / Shutterstock
This article is translated from the original published in Klassekampen on 29 September 2026 and is republished with permission.
The private worries of a DeepSeek engineer strikes a chord with China’s young people.
Imagine you’re good at knitting sweaters. You’ve mastered all sorts of advanced patterns, you know how to pick the right colours, and you make sweaters that keep people warm and last a long time. People pay you good money for what you make and recognise your skill. You like to sit by the window with a pot of tea, knitting through the afternoon while you look out at the mountains, the water, the grazing animals and the chimney smoke.
One day someone invents a new machine. All it needs is yarn and a pattern, and within seconds it spits out a sweater as good as yours. You realise your competitors can now easily reach your level, so you have no choice but to use it too. Twenty years of skill means you would probably still have an edge over your competitors even when everyone uses the machine, but what you will definitely lose is the pleasure of the craft. The joy of sitting by the window, guiding the needles through the yarn and letting time pass slowly, has been crushed by the roar of the machine.
This scenario was first described by Liu Shengyu, an engineer in his early twenties at DeepSeek, one of China’s leading AI companies. In a personal WeChat blog, the young talent writes about what it feels like to help build a technology that, within six months to a year, will take away his joy.
“The rice bowl can be kept, but my old passion will probably have to be given up,” he writes.
The most common AI doomsday scenarios tend to involve a sudden event, such as a powerful autonomous weapons system turning its barrel the wrong way. Liu believes we are already living through a kind of doom, just a less spectacular one.
“I probably won’t become unemployed, but I’ll have to change my line of work. I can keep my rice bowl, but it may mean I never again get the chance to do the work I once loved,” he writes.
Liu has had a dream career. He studied in Peking University’s elite Turing Class, captained the university’s supercomputing team to an overall championship at the ASC Student Supercomputer Challenge, and co-authored a paper on large language model serving while still an undergraduate. He turned down PhD offers from universities abroad and joined DeepSeek in April 2025, where he designs and optimises kernels, the low-level code that makes AI models run efficiently on chips. It was originally a dream job.
“The moment I invent a new technique or see my kernel’s performance rise, my excitement is no less than that of a speedrunner breaking her own record. And when I see my kernel far outperform the vendor’s official one, I feel enormous pride,” he writes.
But the job has now started to leave a bitter aftertaste. When DeepSeek released V4.1 in September, Liu was initially proud: he had written its main attention kernel himself. But he knew that within six months to a year, AI would most likely write kernels as well as he does, or better.
The blog has sparked enormous debate among young people in China and become one of the month’s most-read posts on WeChat. There it has been described as the field notes of a foot soldier on the front line of a new great war.
Liu’s fear of losing his joy has struck a deep chord with a generation about to enter the world’s most demanding job market, shaped by intense competition from AI.
Liu challenges the official narrative in China of AI as overwhelmingly positive, and has also faced a lot of criticism. Many argue that Liu forgets what AI has done for China’s public health, and how it has narrowed the gap in access to knowledge between rich and poor.
Liu, however, stands his ground. He describes a reality in which human talent is buried, while we are reduced to alienated, knowledgeless operators of agents, robbed of important things that made us meaningful participants in society.
“When it comes to destroying itself, humanity has never hesitated,” he writes.
Nor does Liu’s critique stop there. Having alienated us, AI could lead to a dystopian class society in which a handful of tech companies control most resources, and very few people get access to the best AI technology. Only a small elite, he writes, would be able to use the most advanced AI and achieve something close to “mechanical ascension”, while the masses would be left with feeble AI. In the future he envisions, social mobility will no longer be achievable through work and talent, but will be determined by which AI agent you have access to. He calls this scenario “Cyberpunk 2077,” after a dystopian video game.
He contrasts it with another scenario he calls communism. Here, AI will greatly liberate the productive forces, and people’s living standards will rise markedly.
He leaves little doubt about which future he believes Anthropic is steering towards.
“What do you think will happen if Anthropic forever keeps hold of the world’s most advanced AI? Will the society of the future become communism or 2077? Go on, guess,” he writes.
That, he declares, is why he stays at DeepSeek: to fight the cyberpunk scenario, pulling the world as far as he can from “2077” and towards a future where frontier AI is an open, cheap and broadly accessible technology for everyone, a future that resembles his “communism”.
The conflict Liu sketches offers a unique glimpse into how the ideological dimensions of the US–China AI struggle are experienced in China’s most influential tech circles.
The blog was published the day after Anthropic CEO Dario Amodei’s essay “We Must Pace the Frontier,” which lays out a more liberal version of the AI rift. In Amodei’s world, liberal democracies must dominate AI to prevent the technology from falling into the hands of autocratic countries, so that AI is not used to spread autocracy and repression, and can instead be used to induce western style liberalization upon autocratic countries.
Whether it is pure ideology or different material conditions that have produced these divergent understandings of the AI struggle is debatable. For Liu, though, the struggle is real, and perhaps the most important of our time.
“I especially don’t want Anthropic to master the most advanced AI or AGI. To exaggerate a little, the seriousness would be no less than Hitler obtaining atomic bomb technology before the Allies,” he writes.
Related articles:
When the Guardrails Come Off (3-minute read)
When Algorithms Rewrite History: Governing the Digital Erosion of Democratic Memory (3-minute read)
