Welcome back to SavvyMonk, your one-stop for AI and tech news that actually matters.

Today's story is about a leaked memo, a new strike team inside DeepMind, and one awkward fact Google doesn't want to talk about. Claude writes better code than Gemini.

Let's get into it.

Want an A+ when it comes to AI? Subscribe to The Current written by Kim Komando and get a free AI Prompt Hack Pack filled with powerful prompts you can enter into ChatGPT to save you hours (and money) every week.

Get The Free AI Guide Here

Sergey Brin stepped back from day-to-day work at Google years ago, then quietly returned in 2023 to focus on AI as the race heated up. For a while he was more of a presence than a leader, drifting between projects and filing the occasional code request. That changed over the last few weeks.

According to reporting by The Information, Brin is now personally leading a new DeepMind group built for one purpose, which is to close the coding gap between Gemini and Anthropic's Claude.

The group is being called a strike team internally. It is led by Sebastian Borgeaud, a research engineer who previously ran pretraining for Gemini, and it reports up through DeepMind CTO Koray Kavukcuoglu and Brin himself. The trigger was not external pressure from investors or the press. It was internal. DeepMind researchers have been saying out loud that Claude writes better code than Gemini does.

The Memo

In a recent memo to DeepMind staff, Brin laid out the stakes in language that sounds more like a wartime directive than a product roadmap. He wrote that to win the final sprint, the team must urgently bridge the gap in agentic execution and turn the models into primary developers.

He framed coding as the stepping stone to something much bigger, which is AI that can train and improve the next generation of AI without humans in the loop. That is the self-improving loop every frontier lab is racing toward, and Brin told his team that coding is the capability that unlocks it.

The Gap Inside Google

The numbers help explain why Brin is so worked up. Google reportedly generates about half of its code internally with AI. Anthropic and OpenAI are operating closer to 70 to 100 percent, with some senior engineers at Anthropic saying they no longer write code by hand at all.

Claude Code, Anthropic's coding agent, has become the default tool inside its own walls and has rocketed to the top of external developer surveys since launching in May 2025.

Google has its own internal coding tool called Jetski, which is its version of the public Antigravity IDE, and the company now tracks how often engineers use it through an internal leaderboard that ranks teams by adoption. Some Gemini engineers are now required to use these internal agents on complex tasks.

Why This Matters

This is not a product launch story. Brin is not promising a new Gemini model or a flashy consumer feature. He is trying to change how Google builds Google. The strike team's real job is to train coding models on Google's internal codebase, which is one of the largest and most proprietary in the world.

Those models cannot be released to the public because they are trained on private code, but they can accelerate everything Google builds, including the next public version of Gemini.

The deeper point is what Anthropic and OpenAI have already figured out. If your AI can write your AI, you compound faster than any competitor who is still writing things by hand. That is the loop Brin is chasing, and coding is the first rung on the ladder.

The Bottom Line

Brin's involvement tells you something important about where the AI race actually sits right now. Google has more data, more compute, and more researchers than almost anyone, and yet the co-founder felt the need to personally lead a team to catch up on one specific capability.

Coding is no longer just a feature but the leverage point for everything else, and the companies that win it first will build faster than the ones still trying to get there.

Category: Codebase Review

"Act as a senior software engineer reviewing my codebase for automation opportunities. I will paste in a directory structure and key files. Identify the three tasks in this project that are most repetitive, most rule-based, and most time-consuming for a human to do manually. For each one, explain what an AI coding agent could realistically automate today, what would still need human review, and the estimated time saved per week. Ask me for the codebase now."

When a co-founder who hasn't run a team in years comes back to personally lead one, it tells you how the people closest to the action actually feel about the race. Brin could have just sent a memo, but instead he is in the room, and that should tell you something about the pace AI is moving at right now and how much of it is happening below the surface of what gets announced.

Hit reply, I read every response.

See you in the next one.

— Vivek

P.S. If you know a developer, indie hacker, or anyone building with AI tools, forward this issue along. They can subscribe at https://savvymonk.beehiiv.com/