Spend more than 10 billion yuan! Google was exposed to poach its AI programming team, which was established nearly a year ago

Compiled by Zhidongzhi | Editor by Eggplant | Cheng Qian Zhidongzhi reported on August 6 that Business Insider reported on August 5, citing four people familiar with the matter, that Google is currently negotiating a potential deal worth more than US$1.5 billion (approximately RMB 10.2 billion) with Mech anize, an American AI programming evaluation startup. According to people familiar with the matter, according to the current transaction design, Mechanize will continue to retain the ownership of the company's main body and technology, while Google will obtain the "non-exclusive right to use related technologies" and absorb some Mechanize employees to participate in model evaluation and development. Some people familiar with the matter said that the deal is still under negotiation and the transaction structure and amount may be adjusted. Google and Mechanize have currently declined to respond to relevant messages. Mechanize was founded in 2025 and mainly develops training environments and evaluation tools for AI programming agents. The company raised US$9.1 million (approximately RMB 61.8 million) in April 2026, and was valued at US$500 million (approximately RMB 3.4 billion) at the time. Google and Mechanize both declined to comment. 1. Mechanize designs an "examination room" for AI programming agents, and Google may introduce a team to strengthen the model. Mechanize does not directly develop AI programming assistants for ordinary users, but builds a software engineering training environment and evaluation system for cutting-edge AI models. Mechanize builds a virtual software engineering simulation sandbox for AI programming agents. AI programming agents are placed into this digital environment to complete real development work such as developing new functions, deploying applications, and debugging unfamiliar code libraries; the system will automatically judge the effectiveness of task completion and use the results for model training and capability evaluation. This type of tool is equivalent to designing an “examination room” for AI programming agents that is closer to real work. Unlike traditional benchmarks that only check whether a piece of code can run, Mechanize hopes to test whether the model can handle complex projects for a long time, understand existing code, and complete the development and debugging process independently. Mec