Brin Steps In: Google Builds Elite Team to Challenge Anthropic in AI Coding

Google is getting serious this time.

The Information dropped a story yesterday: DeepMind has assembled an elite team specifically targeting Gemini coding capabilities. Led by Sebastian Borgeaud, who previously ran Gemini pre-training.

The interesting part? Sergey Brin himself is directly involved.

A company co-founder personally overseeing a specific technical direction is unusual for Google. The last time Brin was this hands-on was probably during the Waymo era.

Why Coding?

The answer is straightforward: Anthropic Claude has pulled ahead of everyone in coding.

Claude Code hit 80.8% on SWE-bench. Cursor 3 multi-agent collaboration is built on Claude capabilities. The entire AI coding toolchain is practically monopolized by one company.

Google has good models. Gemini 3.1 Pro performs well on many general tasks, but when it comes to complex engineering-level coding — building complete software projects from scratch, understanding large codebase context — it falls noticeably short.

My own experience confirms this. I tried using Gemini for a medium-scale refactoring task, and its understanding of cross-file dependencies was clearly a tier below Claude.

Can an Elite Team Fix This?

I am cautiously skeptical.

Coding ability cannot be brute-forced with headcount. It requires deep model understanding of code structure, long-context stability, and consistency across multi-step tasks. These improvements depend more on training data quality and methodological innovation than team size.

But Brin stepping in personally sends a signal: Google internally now treats AI coding as a strategic priority.

That is fundamentally different from their previous approach of doing a bit of everything without excelling at anything.

What This Means for Developers

Short term, not much changes. Gemini coding improvements will take time — probably second half of this year at the earliest.

But medium to long term, this is good for the entire AI coding ecosystem.

The current problem is Anthropic dominance. Claude Code works great, but it is not cheap, and your entire workflow gets locked into one company API. If Google can bring Gemini coding to the same level, developers gain real choice.

Competition always beats monopoly.

A Small Prediction

Google will likely announce a coding-optimized Gemini variant at Google I/O or some event later this year. Maybe called Gemini Code, maybe integrated into existing products.

The real test will not be benchmarks. It will be real development scenarios — large project refactoring, cross-file dependency analysis, long-horizon bug fixing — where it needs to genuinely compete with Claude.

High benchmark scores do not guarantee usability. Anyone who has shipped code knows this.