Moonshot AI, the company behind the popular Kimi assistant, operates like a classic tech startup fantasy — no job titles, extreme flat hierarchy, and an AI system that screens candidates for "abstract ability and paranoia." That's according to president Zhang Yutong, who shared internal details in a conversation at Peking University's Guanghua School of Management.
Organization-wise, Moonshot AI's roughly 300 employees have zero job titles. The hierarchy is brutally flat — any two people are at most one or two layers apart. Job boundaries are deliberately fuzzy: someone working on pre-training can switch to post-training, algorithm engineers can pivot to data, and marketers can help evaluate models. Zhang's explanation: people's underlying intelligence is transferable, and labels shouldn't define anyone.
On hiring, Moonshot AI already uses an AI interview system to screen candidates. The system ignores degrees and majors. It looks for two things: can you come up with original good ideas, and are you willing to iterate on an idea obsessively. The AI tracks how many times a candidate iterated, what different approaches they tried, and what time of day they were still at it — to filter for people with "abstract ability and paranoia."
Product strategy? Zhang made it clear: Kimi is strictly a productivity tool. It focuses on long, complex, high-economic-value tasks — no entertainment or lifestyle products. She also noted that when Moonshot open-sourced K2.6, it was already adapted for Huawei's Ascend chips, and she believes inference compute demand is exploding.
On open source versus safety, Zhang directly called out Anthropic: "If you think the model is risky, then having it in the hands of a few people might be even riskier." Moonshot AI chose open source because, in their view, more participants equals more safety, not less.