Google DeepMind Mobilizes Strike Team to Narrow Coding Gap with Anthropic

Google has assembled a specialized strike team at DeepMind, led by researcher Sebastian Borgeaud and overseen by co-founder Sergey Brin and CTO Koray Kavukcuoglu, to rapidly close what executives acknowledge as a significant performance gap in agentic coding capabilities—particularly against Anthropic's Claude. The effort marks an unusually direct competitive move, with Brin's internal memo framing the initiative as essential to winning what he calls "the final sprint" in AI development. The underlying gap is stark: Anthropic reports that nearly 100% of its codebase is written with AI assistance, while Google's current rate stands at roughly 50%. DeepMind researchers internally rate Claude's code-writing abilities above Gemini's, a reality that prompted leadership to act with urgency. The strike team's strategy includes training Gemini directly on Google's proprietary internal codebase of over 2 billion lines of code—a data advantage the company hopes will translate into improved long-context reasoning for handling massive code repositories. Rather than positioning the effort as a direct product response to Claude Code (Anthropic's popular developer tool), Brin frames superior AI-assisted coding as an intermediate milestone toward a larger goal: achieving self-improving AI systems capable of training future iterations of themselves. Internally, Gemini engineers are already required to use Google's agentic tools for complex tasks, with usage tracked on a company leaderboard called Jetski, signaling how central this capability has become to the company's AI strategy. Developer and researcher sentiment reflects recognition that coding ability has become a key differentiator: many in the AI community view Anthropic's high usage rates as both a product advantage and a strategic asset for continuous model improvement. However, observers note that the reorganization's effectiveness remains uncertain; recent departures of senior staff members—including Jonas Adler from AI coding work and Alexander Pritzel from model training systems—have raised questions about whether structural changes alone can overcome the technical gap without retaining specialized talent. The competitive dynamic underscores a broader shift: major technology companies no longer compete only with other giants but now face intense pressure from well-focused AI-first startups, compelling them to dedicate elite teams to catching up on specific capabilities.