Google co-founder Sergey Brin has taken a hands-on role, assembling and leading an AI “tiger team” aimed at rapidly improving its artificial intelligence models’ programming capabilities to catch up with the perceived lead of rival Anthropic. According to a report by The Information, this initiative is considered one of the highest-priority projects within Google, underscoring that autonomous AI programming is becoming a new core competitive arena for tech giants.
Informed sources revealed that Google has assembled a tiger team of researchers and engineers, led by DeepMind research engineer Sebastian Borgeaud, reporting directly to Brin and DeepMind Chief Technology Officer Koray Kavukcuoglu. The team’s core mission is to optimize Google’s AI programming models, tackle complex tasks, with the ultimate goal of achieving an “AI leap”—creating AI systems capable of self-improvement.
The direct catalyst for this strategic shift is the formidable strength Anthropic has demonstrated in AI programming. In January, Boris Cherny, head of Anthropic’s Claude Code project, claimed that nearly 100% of the company’s code was written by AI. In contrast, Google Chief Financial Officer Anat Ashkenazi revealed during an earnings call in February that approximately 50% of Google’s code is generated by programming agents. Multiple sources indicated that researchers within Google DeepMind believe Anthropic’s programming tools have surpassed Google’s own Gemini models in capability.
Brin made the urgency clear in a recent internal memo, demanding the team accelerate its efforts. He wrote, “To win the final sprint, we must quickly close the gap in agent execution capabilities and turn our models into the core developers of code.” He specifically emphasized the importance of developing “Agentic AI”—systems capable of autonomously handling multi-step, complex tasks—and mandated that all engineers working on the Gemini project must use internal agent tools.
To achieve this catch-up, Google is taking a series of aggressive measures. First, the company has shifted its model development strategy from primarily serving external customers to prioritizing the creation of code-generation models for internal use. This means new models will be trained on Google’s vast, private code repository, which differs significantly from the public code repositories commonly used to train general-purpose programming models. While models trained on internal code cannot be released directly, this approach could theoretically help Google build superior foundational models that could later be adapted for the public market.
Second, Google is aggressively promoting the use of its programming tools internally. Similar to Meta, Google maintains internal leaderboards tracking the usage of its programming tool, “Jetski.” The company is also forcing engineers in certain teams to attend training courses on AI tool usage to boost adoption and efficiency.
CompanyAI Code Generation RatioKey InitiativesAnthropicClose to 100%Claims the vast majority of internal engineering needs are met by AI programming tools.Google~50%Formed a tiger team, pivoted to training on internal code, mandated employee tool use.OpenAIRatio not publicly disclosedHas shut down projects like Sora to reallocate resources to programming and enterprise business.
Note: Data based on company statements and reporting by The Information.
A Google spokesperson confirmed the company’s focus on AI programming tools in a statement: “These tools have dramatically accelerated our model and AI tool development work, and we are focusing intensely on this area.”
Industry observers note that Google’s move is not an isolated event. Last month, OpenAI announced it was shutting down its video generation model Sora to re-focus resources on programming and enterprise business. Meanwhile, Anthropic continues to double down on AI programming. These trends collectively indicate that AI programming capability—which enhances a company’s own R&D efficiency—is gradually replacing mere text or image generation as a key metric for measuring the core competitiveness of AI labs.
The significance of this race around AI programming extends far beyond improving software development efficiency. Analysis suggests that if AI can efficiently write and optimize code, especially code for AI research itself, it could significantly accelerate the iteration speed of artificial intelligence technology, potentially having a profound impact on the path and landscape toward achieving Artificial General Intelligence (AGI). The “AI leap” goal emphasized by Brin is based on this very logic.
For investors, Google’s efforts to catch up on foundational AI capabilities are crucial. Despite competitive pressures, Wall Street analysts remain broadly optimistic about Google parent Alphabet (GOOGL). According to data from TipRanks, over the past three months, 26 analysts have issued a “Buy” rating, with 5 issuing a “Hold,” forming a “Strong Buy” consensus. The average price target is $386.31, implying roughly 13.7% upside potential.
As the world’s leading AI labs channel core resources into the field of autonomous AI programming, a deep-seated competition that could reshape software development and even AI R&D paradigms has begun. Whether Google’s founder-led “tiger team” assault can enable it to come from behind will be a key bellwether for observing the evolution of the industry landscape.