Google assembles elite team to catch up with Anthropic, even its own engineers are secretly using Claude Code

📄Full Article· Automatically extracted by trafilaturaGemini 翻譯1387 words

Google DeepMind has secretly assembled a "Strike Team," led by former Gemini pre-training lead Sebastian Borgeaud, with the goal of catching up to Anthropic's lead in the field of AI programming. (Previous coverage: Commentary: All vibe coding tools are selling the illusion of getting rich quick) (Background: Ethereum Foundation establishes "dAI team": To make Ethereum the preferred settlement and coordination layer for AI and the machine economy) Google CFO Anat Ashkenazi publicly revealed in early 2026 that approximately 50% of Google's code is currently written by AI agents. The numbers at Anthropic are from almost another universe: Claude Code lead Boris Cherny claims that "100% of his code" is already produced by Claude, and the entire company is "also almost 100%." This contrast has set off alarm bells within Google. According to a report by The Information, Google has quietly assembled an elite team of top researchers and engineers to respond directly to the pressure from Anthropic. The head of this "Strike Team" is Sebastian Borgeaud, whose previous role was the pre-training lead for the Gemini model. The logic behind this selection is clear: Borgeaud knows best where the capability ceiling of Gemini lies, and naturally, he knows best which links need to be broken through. Google co-founder Sergey Brin and DeepMind CTO Koray Kavukcuoglu are reportedly personally involved, which is uncommon in Google's organizational culture. Brin recently wrote in a memo to DeepMind employees: "To win the final sprint, we must urgently close the gap in agent execution and transform models into primary developers." In terms of technical direction, the focus of the Strike Team is not on short, quick, and simple auto-completion, but on "long-task programming." In plain terms, this means enabling models to not just write a single function, but to read a large codebase, understand the developer's intent, and deliver complete, usable software modules. Google has even established a leaderboard internally to track the usage rate of AI programming tools across teams, and Brin has mandated that employees use internal agent tools for multi-step tasks. Google plans to train a batch of specialized models for internal code under the Strike Team framework. Because these models contain trade secrets, they cannot be released publicly, but they can serve as a "waystation" for improving public versions. This approach is not new; Apple is following a similar path, also training AI tools primarily for internal development scenarios, which are also not released externally. The difference is that the source of Google's pressure is more explicit: Claude Code is not just an external competitor in the market, but a tool that Google's own engineers are secretly using. A report on Medium pointed out that Claude Code has become one of the most frequently used external AI programming tools by Google engineers. When a company's employees are using a competitor's product to fill the gaps in their own models, this signal is hard for senior management to ignore. For Google, this race has an asymmetrical starting point: Anthropic is an AI-native company that started by using AI to write code, while Google is a tech giant trying to reshape itself with AI. Scale brings inertia, but it also brings resources, which can be described as a double-edged sword. The existence of the Strike Team represents Google's choice to cut a fast track within the system, rather than waiting for the slow transformation of the entire organization.

Data Status✓ Full text extractedRead Original (動區 BlockTempo)

🔍Historical Similar Events· Keyword + Asset Matching6 items

2026-04-22

Google Stitch 開源設計系統：DESIGN.md 讓 Claude Code、Codex、Antigravity 生成高品質 UI

2026-04-23

OpenAI CEO Sam Altman slams Anthropic: Fear-mongering Claude Mythos is just to monopolize AI

2026-04-23

Introduction to Claude Code's new feature /ultrareview: Cloud-based multi-agent deep code review, now free for a limited time for Pro and Max users

Similarity 170%關鍵字 code/claude同分類 zh

2026-04-22

Bloomberg reveals Claude Mythos has been subject to unauthorized access! The hardest vulnerability for Anthropic to defend against will always be "people."

2026-04-19