To narrow the gap with Anthropic's Claude model in the field of AI programming, Google DeepMind has recently reorganized its resources and established an elite team called Gemini, led by Sebastian Borgeaud. The core mission of this team is clear: to tackle complex and long-term programming challenges, and comprehensively enhance the practical capabilities of AI in engineering development.

As a co-founder of Google, Sergey Brin has shown great interest in this project. He not only personally oversees it but also participated in the planning process together with DeepMind's Chief Technology Officer. Brin emphasized internally that Google must close the gap with Anthropic in terms of AI agent execution capability. He believes that strong programming ability is not only a benchmark for measuring AI intelligence, but also a crucial cornerstone for achieving AI "self-improvement".

Currently, Google is tracking the usage rate of agents through a tool called "Jetski" and using this data to quantitatively rank team performance. To accelerate the AI transformation of the development process, Google has also organized a large number of engineers to participate in specialized training, requiring them to use internal agent tools to improve daily development efficiency.

In fact, Google's layout in the AI field goes beyond programming. Recent developments show that its technical reach is expanding to all corners: from developing native applications for Apple Macs, to deeply integrating Gemini into document and spreadsheet office suites, and providing non-classified AI assistants for the U.S. Department of Defense. In addition, Google is also increasing its investment in physical AI, improving the quality of AI Overviews answers, and plans to introduce large models into the smart TV industry.

This series of intensive actions sends a clear signal: Google is accelerating from the technical foundation to the application terminal, trying to regain control in the efficiency competition of AI agents.