Anthropic Pushes Google to the Brink: Brin Returns to Lead 'Strike Team' in Hunt for Claude

Reported by New Intelligence Element

Editors: Hao Kun, Tao Zi

Google has formed a specialized strike team dedicated to hunting down Anthropic!

According to a report by The Information, Google DeepMind has quietly assembled a "strike team" composed of researchers and engineers.

Co-founder Sergey Brin has made another limited-time return to the front lines.

There is only one goal: to reclaim the lead in programming capabilities that has been lost to Anthropic.

In a memo to all DeepMind staff, Brin issued a grave warning:

The last time such measures were taken was in 2023, when Google sounded "Code Red" and brought back Larry Page and Sergey Brin to take charge.

After nearly three years of deep cultivation, Google Gemini finally surged past OpenAI to reclaim the throne.

Currently, Claude is sharp and dominant in the field of programming; even OpenAI Codex struggles to shake its position, let alone Google.

After all, Anthropic has achieved 100% internal coding via Claude, while Google's AI coding adoption remains at 50%.

This personal supervision by Brin marks his most high-profile intervention in the AGI race to date.

Strike Team Fully Equipped, Brin Watching Closely

The strike team is led by Sebastian Borgeaud, a Google DeepMind research engineer who previously oversaw pre-training for the Gemini models.

However, the leader is not the focus. The focus is on who is watching.

Both Brin and DeepMind Chief Technology Officer Koray Kavukcuoglu are personally involved in the strike team's work.

It is extremely rare for a Google co-founder to descend to the level of a frontline project team; this configuration is unprecedented within Google.

The strike team is focused on long-term coding tasks, specifically building complete software from scratch. The requirement is for the model to read multiple files comprehensively and precisely understand user intent—currently the toughest bone for AI to chew.

Google Admits Defeat: Coding Capabilities Lag Behind Claude

This is not an evaluation by external media. Researchers within DeepMind themselves have admitted that Anthropic's programming tools have surpassed Gemini.

How large is the gap? A single set of numbers suffices.

Boris Cherny, head of Claude Code, publicly stated this January that "almost 100%" of Anthropic's code is written by AI.

In contrast, Google CFO Anat Ashkenazi revealed during the February earnings call that Google's figure is approximately 50%.

100% versus 50%.

Claude Code's commercial performance is even more explosive: officially opened to the public only in May 2025, its annualized revenue surpassed $2.5 billion by February 2026—going from zero to $2.5 billion in just nine months.

Enterprise subscriptions have quadrupled since the beginning of 2026. One in five enterprise users on the Ramp platform is now paying for Anthropic, a ratio that was one in twenty-five just a year ago.

When looking at Anthropic's overall portfolio, the numbers are even madder.

Annualized revenue hit $1 billion at the end of 2024, surged to $4 billion by mid-2025, reached $9 billion by year-end, and was officially announced at $14 billion in February 2026. That is a 14-fold increase in 14 months.

This February, the company completed a $30 billion Series G financing round, achieving a valuation of $380 billion, making it the third-largest unicorn globally.

Claude Code has already captured a 54% share of the enterprise programming market. OpenAI holds 21%. Google was not even listed separately.

On GitHub, 4% of public commits are already generated by Claude Code, with projections exceeding 20% by year-end.

More critically, there is the habit effect. A Google Developer Expert (GDE) stated openly in a letter that being forced to use Claude Code for 8 hours a day at work has caused these habits to "permeate personal projects."

Even though he prefers Gemini's context window, he cannot stop the power of muscle memory.

Internal Codebase: Google's Ultimate Trump Card

The strike team's strategy involves a key pivot: no longer primarily building coding models for external customers, but prioritizing the training of models capable of writing Google's internal code.

Google's internal private codebase is worlds apart from public code. To allow AI to take the position of "primary developer" within Google, the model must ingest Google's own code.

While such models cannot be publicly released, they can produce superior derivative models that can be面向 the public.

A Google spokesperson stated that the "adoption rate" of internal programming tools is "extremely astonishing," and that large-scale use of these tools has "installed a powerful engine for model and AI tool development."

On the implementation level, all three major initiatives are being maximized.

Google's internal programming tool, Jetski, has established a leaderboard to track usage across teams, similar to Meta's approach.

Another asynchronous AI Agent, "Agent Smith," can continue executing tasks even when employees are offline; it became so popular that Google一度 had to restrict access.

In his memo, Brin explicitly required that every Gemini engineer must use internal agents when handling complex, multi-step tasks. "Mandatory," not suggested. Some teams have already incorporated the use of AI tools into performance reviews.

Sundar Pichai is also increasing pressure. Reports indicate that some non-technical employees have been told that AI use is "no longer encouraged, but expected."

This recalls an incident he mentioned last year at the All-in Summit. Upon returning to Google, he discovered that Gemini was actually on the ban list for internal programming tools. His exact words were, "I can't explain what kind of ridiculous reason this is; the whole thing left me dumbfounded." It eventually required Pichai's personal intervention to resolve.

Google's own AI was banned by Google itself from writing code. The existence of such internal friction is sufficient to explain Google's lag in the coding track.

All Three Giants Go All-In

Google's anxiety is not an isolated case.

OpenAI has just cut its Sora video generation project, shutting down its website and app on April 26, reallocating all computing power and engineering resources to coding and enterprise products.

Foreign media put it bluntly: Anthropic's Claude Code is "stealing OpenAI's lunch."

Sora was burning millions of dollars daily, with users dropping from a peak of one million to under 500,000, while Claude Code achieved $2.5 billion in annualized revenue in the same period over nine months. Disney's $1 billion collaboration deal for Sora was notified less than an hour before the shutdown news was announced.

In the same week, three executives—OpenAI's CPO, the head of Sora, and the Corporate CTO—resigned on the same day. OpenAI internally refers to these non-core projects as "side quests" and is clearing them out one by one.

All three major AI giants have unanimously pushed coding capabilities to the highest priority.

In his memo, Brin positioned coding capability as the inevitable path to "AI takeoff." So-called AI takeoff refers to enabling AI to evolve itself. A sufficiently strong code agent, combined with AI capable of solving math problems and running experiments, could theoretically automate the work of AI researchers and engineers on a large scale. OpenAI is already using similar tools internally to improve researcher efficiency, including automatically generating code for model training experiments.

Whoever builds the strongest coding AI first may be the first to create an AI capable of improving itself. This is what Brin is truly racing for.

Brin's Two Letters

This is not the first time he has panicked over this issue.

In February 2025, he sent a memo to the Gemini team, stating, "Competition has accelerated dramatically; the final race to AGI has begun," and suggested engineers work 60 hours a week.

The tone back then was mobilization. This time, it has become an order: "Rapidly close the gap" and "Mandatory use."

Between mobilization and orders lies 14 months. In those 14 months, Anthropic's annualized revenue grew from $1 billion to $14 billion, Claude Code went from nothing to $2.5 billion, and captured over half the market share in enterprise coding.

When ChatGPT emerged in late 2022, Google sounded a "Code Red," with Page and Brin reviewing code overnight.

Three and a half years later, a new Code Red has arrived. Only this time, the opponent is not OpenAI, but Anthropic—the company founded by former OpenAI VP of Research Dario Amodei along with a group of core researchers who left.

Those who left Google built OpenAI; those who left OpenAI built Anthropic. Now, Anthropic has forced Google to form a strike team in response.

The food chain of the AI industry has come full circle.