Google Revamps AI Coding Strike Team to Close Gap With Anthropic

Google's AI coding strike team was already a sign of internal alarm when it formed in April 2026. Now it is being reorganized, and the timing is uncomfortable: two of the researchers heading to Anthropic worked on exactly the problems the team was assembled to fix.

The Information reported that Google is revamping the recently launched strike team focused on AI coding, as the company struggles to narrow the gap with Anthropic. The reorganization follows the planned departures of Jonas Adler, who worked on Google's AI coding effort, and Alexander Pritzel, who was involved in the process of training AI systems -- both described as key contributors to the Gemini model and both heading to Anthropic. Other recent losses include Nobel laureate John Jumper, also departing for Anthropic, and Gemini co-lead Noam Shazeer, heading to OpenAI.

The competitive backdrop makes those departures pointed. According to earlier reporting, Google's own CFO said Anthropic writes close to 100% of its code with AI assistance, while Google sits at approximately 50%. Sergey Brin, directly involved with the strike team alongside Google DeepMind Chief Technology Officer Koray Kavukcuoglu, put it bluntly in an internal memo: "To win the final sprint, we must urgently bridge the gap in agentic execution and turn our models into primary developers" of final code. Staff at DeepMind have reportedly raised concerns about the company lacking a clear developer product in AI coding, an area where Anthropic and OpenAI have gained substantial ground.

One structural complication is worth noting: reports indicate the strike team's work involves training on Google's own proprietary codebase, which would complicate any direct public release of those improvements. What the reporting does not specify is whether the reorganization expands explicitly into model training work or focuses primarily on the product and tooling layers. The departures are at least partly driven by Anthropic and OpenAI approaching IPOs, giving prospective joiners equity upside that large-company employment is hard-pressed to match. For teams choosing which AI coding tool to build around, the direction of talent flow is at least as informative as any benchmark leaderboard.