DeepSeek engineer explains why he's building an AI that will replace him – so the world doesn't become like Cyberpunk 2077

An engineer at Chinese AI lab DeepSeek published a lengthy post acknowledging the inevitability of his own replacement by artificial intelligence.

The developer, known online as interestingLSY/intlsy, works on the company's kernel systems and says he keeps developing the technology precisely because of competitive pressure. His reasoning goes that stepping away from the work wouldn't change anything – it would just hand the advantage to other players.

Shengyu Liu joined the debate over AI development speed, a conversation that heated up after Anthropic CEO Dario Amodei's statement about the need to slow things down. Jensen Huang from NVIDIA pushed back on Amodei position, while Elon Musk and Sam Altman from OpenAI backed the idea of caution.

Mark Zuckerberg also put out a lengthy post explaining that his company had internally slowed model development due to safety requirements, just without making a big announcement about it.

Liu opens his argument by describing how quickly AI role in his daily work has changed. Not long ago it was a simple assistant for checking code and reading documentation, but now the system can navigate complex software platforms like NVIDIA CUDA and analyze them through operational tools.

The engineer then lays out specific speed estimates. By his calculations, within six months to a year, AI will start producing output comparable in quality to his own.

The model "thinks" at roughly 300 tokens per second, types out a line of commands in a second, and writes a piece of code in 20 seconds – far outpacing human capability.

Liu admits that from a purely personal standpoint, he has little reason to speed this process along. Beyond the thrill of pushing the technology forward and pride in building products that exceed expectations, he points to the main argument – even if he slows down, competitors will push their models to the point where his work becomes unnecessary anyway.

The engineer then moves to a broader argument, describing two possible futures for society. One scenario mirrors the world of Cyberpunk 2077, the other follows a communist model of resource distribution.

In the cyberpunk scenario, the most powerful AI systems end up controlled by a narrow group of players, with top-tier technology access limited to a handful of people while everyone else gets stripped-down, lower-performance versions. Liu directly asks what the world would look like if the strongest AI ended up in Anthropic hands.

Since he doesn't believe OpenAI or Anthropic will open up cutting-edge intelligence to everyone at an affordable price and without restrictions, working at DeepSeek looks to him like the only way to make the strongest technologies open.

The engineer got particularly harsh toward Anthropic, comparing the lab potential lead in building artificial general intelligence (AGI) to handing a weapon of mass destruction to Hitler before the Allies could get their hands on the technology.

Liu also predicts a shift in professional roles across the industry. Value will move away from those who write code by hand toward those who can use AI to rapidly build high-performance software.