Google Rearranges AI Programming SWAT Team to Catch Up With Anthropic

According to Dynamic AI Beating monitoring, Google is restructuring its AI SWAT team, established just a few months ago, to catch up with Anthropic. The reorganization will expand the SWAT team's scope of work and transition it from short-term projects to a formal structure.

During the restructuring, Google experienced the departure of two core researchers. Last week, one of the co-founders of Transformer, Noam Shazeer, abruptly left to join OpenAI, citing Google's merger of computing resources limiting Shazeer's access. On Saturday, Nobel laureate and DeepMind Vice President John Jumper announced his move to Anthropic. Prior to his departure, Jumper had been reassigned to the programming SWAT team. The departure of core researchers caused Alphabet's stock price to plummet by 5% on Monday.

The core of the reorganization is adjusting the model training process. Google is expanding the programming SWAT team into a mid-training team, positioned between pre-training and post-training. The mid-training team will take over the programming, among other model capabilities expansion tasks, previously handled by the post-training team, while the post-training team will focus on optimizing human-computer interaction. By segmenting the stages and clarifying roles, Google is attempting to accelerate model capabilities improvement.

Behind the reorganization is the fact that programming has become the most valuable segment in AI services. Leveraging its programming tools, Anthropic has achieved an annualized revenue of $47 billion last month, more than three times its February level. In contrast, Google's early investment in programming was inadequate, mistakenly believing that powerful base models would automatically acquire programming proficiency. Although co-founder Sergey Brin and DeepMind CTO Koray Kavukcuoglu personally participated in the programming SWAT team, Google's programming applications did not progress smoothly. Upon its initial release, the Antigravity tool received a lukewarm response due to numerous bugs, while the Gemini 3.5 Flash was criticized by developers for its high pricing and user-catered responses. The flagship model, Gemini 3.5 Pro, has yet to be launched.