Anthropic is pressing harder, and Google has formed a "task force" to improve its AI programming models.

```

Google is waging a catch-up battle in the field of AI programming. Facing Anthropic's sustained lead in code generation capabilities, Google has secretly assembled an elite team, focusing efforts on improving the competitiveness of its own AI programming models and linking this endeavor directly to its broader goal of "AI self-evolution."

According to a report by The Information on Monday, sources revealed that Google co-founder Sergey Brin and Google DeepMind CTO Koray Kavukcuoglu are both directly involved with the team, demonstrating the strategic priority of this action at the highest levels of Google's management. In a recent internal memo, Brin used urgent language, requiring Google DeepMind employees to "actively transform, close the gap in agent execution, and turn our models into the main developers of code."

This sense of urgency has a real basis. Google's CFO Anat Ashkenazi disclosed during an earnings call in February that AI programming tools are currently responsible for about 50% of Google's internal code writing; meanwhile, Boris Cherny, head of Claude Code at Anthropic, publicly stated in January that "nearly 100%" of Anthropic's code is written by AI. This gap is the direct trigger for Google's emergency response.

Elite Team Established, Aiming for the Ultimate Goal of AI "Self-Evolution"

Google has drawn top talent from its researchers and engineers to form a "task force" focused on improving AI programming models. Three sources confirmed that the team is led by Google DeepMind research engineer Sebastian Borgeaud, who previously led pretraining work at Google DeepMind.

The core mission of the team is to improve the model's performance on long-term programming tasks, such as writing a complete software system from scratch. These tasks are highly complex and require models to read large numbers of files and accurately understand user intent.

Brin regards enhancing AI programming skills as a key step toward achieving "AI takeoff"—a stage where AI can self-improve and independently advance research. He said that once advanced programming agents are combined with AIs capable of solving mathematical problems and running experiments autonomously, it will be possible to automate the work of AI researchers and engineers on a large scale in the future. OpenAI has already internally deployed similar tools to accelerate the productivity of their AI researchers, including automatically generating code for model experiments.

Strategic Focus Shifts, Internal Code Training Becomes Key

This initiative is accompanied by a notable strategic shift: Google is moving its programming model focus away from external customers and toward serving internal company needs.

The technical logic behind this shift is that Google's private codebase differs significantly from the external code typically used to train general-purpose programming agents, so it is necessary to specifically train models using internal code to improve actual performance. However, models trained on internal code cannot be released publicly, but in theory, they can support the release of higher-quality public models in the future.

A Google spokesperson stated that internal programming tools have seen "widespread adoption," greatly accelerating the development of our models and AI tools—which is our true focus.

Internal Rollout Accelerates, Employee Usage Rate Included in Assessment

Google is also ramping up efforts to promote internal adoption. According to two sources, Google has followed Meta's example by establishing internal leaderboards to track employee usage of the internal programming tool Jetski. In addition, certain teams outside Google DeepMind have implemented mandatory AI training for engineers.

In the memo, Brin explicitly required every Gemini engineer to be mandatorily required to use internal agent tools for complex multi-step tasks.

This internal mobilization reflects the competitive landscape across the frontier AI field. Programming capabilities have become the key battleground for major AI labs this year—OpenAI last month announced the shutdown of its Sora video generator to reallocate resources toward programming and enterprise business; while Anthropic, with its accumulated advantage in code generation, continues to put pressure on Google and OpenAI. Google's formation of this "task force" is the latest sign of this intensifying contest.

Risk Warning and DisclaimerThe market has risks; investment requires caution. This article does not constitute personal investment advice, nor does it take into account individual users’ specific investment goals, financial situation, or needs. Users should consider whether any opinions, views, or conclusions in this article suit their particular circumstances. Investment based on this is at your own risk. ```