When considering what AI models could make humanoid robots more autonomous, general-purpose machines, Wang has previously expressed skepticism of large world models as being too compute-intensive. But during Unitree’s recent IPO, he spoke of the company attempting to use large AI models as the foundation for a type of physical AI that could potentially achieve an “autonomous loop of perception, decision, execution, evaluation, learning, and evolution,” according to Caijing Magazine.
Extreme micromanagement and employee penalties
Wang, whose background is in hardware structural engineering, reportedly embodies the stereotype of the hard-charging tech founder by still posting in work chats at 2 or 3 am and spending most of his time in the office, including on weekends. Employees described Wang as a blunt communicator who is completely business-focused and will cut off employees after a few sentences.
But Wang’s micromanagement of the company and lack of a formal management structure mean that employees have to queue up to run decisions by him, according to Caijing Magazine. The founder’s emphasis on controlling costs even extends to personally approving any expense reimbursements exceeding 100 yuan (approximately $15), a Unitree employee said.
That leadership style may serve in a small startup, but Unitree has already grown to at least 480 employees this year, Rest of World reported.
Employees also claim that Unitree’s incentive system overwhelmingly takes the form of penalties rather than rewards. Unitree employees supposedly compare who gets less criticism, and Wang reportedly gave every senior executive a performance score of 1 from a range of 0 to 1.5.
One longtime employee told Caijing Magazine that the company has “seen the highest attrition of core staff in Unitree’s history” in 2025 and 2026.
Unitree did not provide any response when Caijing Magazine originally reached out for comment. The company later issued a broad statement to other Chinese media publications describing Caijing’s reporting as having substantial “misinformation” without providing specifics.