Sergey Brin commits DeepMind to a Claude catch-up

2026-04-21 · Source: The Rundown AI · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Software Development & Engineering · Depth: Intermediate, medium

Summary

Google co-founder Sergey Brin is reportedly leading a DeepMind "strike team" to enhance Gemini's coding capabilities, aiming to surpass Anthropic's Claude and achieve self-improving AI. Research engineer Sebastian Borgeaud heads this initiative under CTO Koray Kavukcuoglu. Internally, Claude's code-writing is rated higher than Gemini's, prompting Brin's focus. Gemini engineers are now required to use Google's internal agent tools for complex tasks, with their usage tracked on a leaderboard called Jetski. Concurrently, Moonshot AI open-sourced Kimi K2.6, an agentic coding model that performs comparably to or better than models like GPT-5.4, Opus 4.6, and Gemini 3.1 Pro on benchmarks like Humanity’s Last Exam and SWE-Bench Pro, at a lower cost. K2.6 supports long-horizon tasks, operating for over 12 hours with 4,000+ tool calls, and its agent swarms can spin up 300 parallel sub-agents.

Key takeaway

For AI Engineers focused on model development and integration, Google's push to enhance Gemini's coding capabilities highlights the critical role of code generation in advancing AI autonomy. You should prioritize integrating robust code-writing and agentic features into your models, as this capability is increasingly seen as foundational for achieving self-improving AI systems and automating complex tasks. Consider exploring open-source agentic models like Moonshot AI's Kimi K2.6 for cost-effective and powerful agentic workflows.

Key insights

Improving AI's coding ability is seen as the shortest path to achieving self-improving AI systems.

Principles

• Internal competition drives AI development.
• Agentic workflows enhance AI task execution.

Method

DeepMind is using a dedicated "strike team" and internal agent tools with usage tracking to improve Gemini's coding performance against competitors like Claude.

In practice

• Use Claude Design to generate website landing page mockups.
• Evaluate APIs using "time-to-useful-result" beyond raw latency.

Topics

• DeepMind AI Development
• Gemini Coding Capabilities
• Anthropic Claude
• Agentic AI Models
• Moonshot AI Kimi K2.6

Best for: AI Engineer, Machine Learning Engineer, NLP Engineer, Director of AI/ML, VP of Engineering/Data, Consultant

Related on AIssential

• Welcome to AIE CODE - Jed Borovik, Google DeepMind — The 2025 AI Engineering Code Summit focuses on advancing the critical field of AI coding across the industry.
• Kimi 2.6 Test | The Best Agentic Coding Open Model? | Coding, OCR, Image Understanding | 🔴 Live — Gemini K 2.6 is a large, multimodal open model excelling in agentic coding and visual design tasks.
• Applying an Agentic Coding Tool for Improving Published Algorithm Implementations — Agentic coding tools can significantly improve published algorithm implementations across diverse domains within a single workday.
• Unleash Your Development Superpowers: Refining the Core Coding Experience — Gemini Code Assist enhances developer productivity through AI-powered automation, collaboration, and customization features.
• Google makes an interesting choice with its new agent building tool for enterprises — Google's Gemini Enterprise Agent Platform enables scalable AI agent development for technical and business users.
• Demis Hassabis Thinks AI Job Cuts Are Dumb — Gemini 3.5 Flash excels at complex agentic coding, challenging views on AI-driven job displacement.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Rundown AI.