Google reorganizes AI programming assault team to catch up with Anthropic According to Beating monitoring, Google is restructuring its AI programming assault team, which has only been established for a few months, to catch up with Anthropic. The restructuring will expand the scope of work of the assault team and adjust it from a short-term project team to a formal structure. During the restructuring period, Google experienced the loss of two core researchers. Last week, Noam Shazeer, one of the founders of Transformer, suddenly resigned to join OpenAI due to Google's merger of computing resources restricting Shazeer's access permissions. On Saturday, John Jumper, a Nobel laureate and Vice President of DeepMind, announced his joining Anthropic. Prior to his departure, Jumper had been transferred to the Programming Assault Team. The departure of core researchers led to a 5% drop in Alphabet's stock price on Monday. The core of restructuring is to adjust the model training process. Google has expanded its programming assault team into a mid training team, positioned between pre training and post training. The middle training team will take over the programming and model capability expansion that was previously handled by the post training team, while the post training team will focus on optimizing human-computer interaction. By segmenting stages and clarifying division of labor, Google attempts to accelerate the improvement of its model capabilities. Behind the restructuring is that programming has become the most valuable sector in AI services. Anthropic, with its advantage in programming tools, achieved an annualized revenue of $47 billion last month, more than three times the level in February. In contrast, Google had insufficient investment in programming in its early days, mistakenly believing that powerful foundational models would automatically master programming skills. Despite the personal involvement of co-founder Sergey Brin and DeepMind CTO Koray Kavukcuoglu in the programming assault team, the progress of Google's programming applications has not been smooth. In the early stages of the release of the Antigravity tool, the response was mediocre due to multiple vulnerabilities. The Gemini 3.5 Flash was also criticized by developers for its high pricing and catering to users, while the flagship model Gemini 3.5 Pro has not yet been released. [Original link]