This article is written based on publicly available information and is intended solely for informational exchange, not as any investment advice.

On April 20, according to The Information, Google has assembled a 'strike team' to improve its AI Coding capabilities; Sergey Brin informed DeepMind employees in an internal memo that they must 'make a decisive shift' to catch up with Agent technology.

This news is brief, just over a hundred words, but its signal significance far exceeds the literal meaning.

When a tech giant with a market capitalization exceeding $4 trillion uses militaristic terms like 'strike team' to describe internal technological efforts, and when its co-founder demands that engineers 'mandatorily use internal agent tools for complex multi-step tasks,' it indicates that certain developments are transcending the framework of conventional competition.

01 Signal: From 'Fifty Percent' to 'Nearly All'

Google CFO Anat Ashkenazi stated during a February 2026 earnings call that approximately half of Google's code is generated by AI Coding tools. This figure would be astonishing in any other context, implying that a company with tens of thousands of engineers has delegated half of its code production to AI.

However, without comparison, there is no harm. The reference point has shifted too rapidly.

Boris Cherny, head of Claude Code at Anthropic, publicly stated in January that 'nearly all' of the company's code is generated by AI. Google DeepMind researchers subsequently concluded that Anthropic's Coding tools have surpassed Google's Gemini model in programming capabilities.

The gap between 'half' and 'nearly all' represents a chasm at the engineering system level. It suggests that Anthropic has established an internal process closed loop (closed loop) centered on AI as the core of development. Meanwhile, Google, despite massive investments in AI infrastructure, lags by an order of magnitude in this critical metric.

A noteworthy timeline: On April 16, 2026, Anthropic released Claude Opus 4.7, surpassing GPT-5.4 and Google's Gemini 3.1 Pro in benchmark tests for AI Coding, scalable tool invocation, and Agent computer usage.

Just four days later, news of Google's strike team surfaced.

This proximity in timing is no coincidence. Google's move was undoubtedly triggered by the recent series of new model releases from Anthropic.

02 Strategic Shift: From 'External Focus' to 'Internal First'

Google's formation of the strike team involves a noteworthy strategic choice: prioritizing internal code over external clients.

According to three informed sources, the team focuses on optimizing model performance in long-term code tasks, such as writing entirely new software. Models will be trained on Google's internal codebase rather than public code, meaning they cannot be directly released externally but can serve as 'internal expertise' to enhance public model optimization.

The rationale behind this strategy is straightforward: as coding AI evolves from 'completing a few lines of code' to 'understanding entire engineering systems,' models no longer require general programming corpora but rather contextual understanding deeply coupled with actual development environments. The training value of Google's proprietary codebase lies in enabling models to comprehend Google-scale engineering structures—code review processes, dependency management, testing frameworks, deployment pipelines—rather than merely syntax generation.

A longer-term goal has also been disclosed: the project ultimately aims to achieve autonomous AI iteration, where AI can optimize itself. This is a direction Brin personally emphasizes; he has told employees that enhancing Google's AI Coding capabilities is a 'critical step' toward this ultimate objective.

The leap from 'having AI write code' to 'having AI write AI (self-evolution)' represents a qualitative transformation. If successful, it would mean exponential improvements in R&D efficiency, with models not only generating product code but also participating in their own iteration and refinement.

03 Brin's Memo: No Longer Gentle

Since his return in 2023, Sergey Brin has been deeply involved in Google's AI development. Demis Hassabis, head of DeepMind, confirms that Brin 'personally programs' and writes code almost daily. Brin drove the merger of Google's two major AI divisions, reshaped an engineer-centric decision-making culture, and accelerated the Gemini project significantly from the second half of 2025. Following the release of Gemini 3 Pro, Alphabet's market capitalization reclaimed the global third spot for the first time in seven years, surpassing Microsoft.

However, the tone of his recent memo suggests dissatisfaction with the current competitive landscape.

In a recent internal memo, Brin demanded that DeepMind employees 'make a decisive shift to catch up with Agent technology.' He wrote: 'To win the final sprint, we must swiftly close the gap in Agent execution capabilities, making our models the core developers of code.'

The positioning of 'core developers' is intriguing. It implies upgrading AI's role from a supporting tool to an active participant, from 'helping you write' to 'doing it for you.' Brin also mandated that all engineers involved in the Gemini project use internal Agent tools for complex multi-step tasks—not as a suggestion but as a requirement.

Recalling 2017, Google published the Transformer paper but failed to commercialize it due to 'excessive caution,' ultimately paving the way for OpenAI. Brin acknowledged this in a public dialogue at Stanford. Now, he is personally leading the charge, shifting DeepMind from a research-oriented to a product- and competition-oriented approach, essentially correcting the decision-making inertia of the past.

04 Conclusion: Coding Has Become the Core Battleground

From OpenAI and Alibaba's QianWen earlier on, to Grok later, and now Google, their profound reflections and strategic reshaping along the Coding technology route all prove that coding capabilities have become the core battleground for major AI model companies in 2026.

Behind this lies the fact that since the second half of 2025, Anthropic has leveraged Claude's long-term deep cultivation (deep cultivation) in coding to establish a formidable competitive barrier that rivals fear most.

The reason AI Coding has become the main battleground is simple yet compelling: it is the scenario closest to productivity. While chat tools boast large user bases, their stickiness and willingness to pay are limited. In contrast, improvements in software development efficiency directly translate into commercial value. Whoever first makes Agent products based on Coding capabilities indispensable to users will secure a high-frequency, high-value, and sustainably paying user base.

With Google forming a strike team, focusing on internal code, mandating the use of agent tools, Brin personally overseeing operations, and beginning to define internal R&D rhythms with military terminology, we should recognize that the intensity of the AI large model Coding battle will surpass everyone's expectations.