Sunday, September 6, 2026

News

New Google DeepMind Chief: Only Leading the AI Race Matters

ModelsPatryk Raba
New Google DeepMind Chief: Only Leading the AI Race Matters
Fot. Google DeepMind / Novoto Studio (Visualising AI), Pexels (Pexels License)

Koray Kavukcuoglu, who took over operational leadership of Google DeepMind from Demis Hassabis, says catching up with the top AI models is the company's only goal. He announces that Gemini 4 is the most ambitious training project in Google's history.

Contents
  1. Changing of the Guard at DeepMind
  2. Reason for the Reshuffle
  3. Gemini 4 as a Credibility Test
  4. What This Means for the Market

The new operational head of Google DeepMind, Koray Kavukcuoglu, cut short speculation about the company's priorities with a single sentence: only staying at the front of the race for the best artificial intelligence models matters. It is the first public statement from the new leadership since DeepMind's August restructuring, when Demis Hassabis moved into the role of Alphabet's chief scientist.

The comment came during a public discussion referenced by the-decoder.com, in response to questions about whether Google is genuinely putting frontier AI first or focusing instead on cloud and infrastructure. Kavukcuoglu left no doubt: "there is nothing more important to us than being at the front."

There is nothing else, apart from being at the front, that matters to us - Koray Kavukcuoglu, chief operating officer of Google DeepMind

Changing of the Guard at DeepMind

The August 2026 overhaul of DeepMind's leadership was one of the biggest personnel changes in the unit's history since the merger of Google Brain and DeepMind. Hassabis, the co-founder and longtime CEO, was formally promoted to Alphabet's chief scientist and chairman of DeepMind, but day-to-day operational decisions passed to Kavukcuoglu, previously the unit's chief technology officer and Alphabet's lead AI architect.

Kavukcuoglu took over oversight of Gemini model development, frontier AI research, the Gemini app, and the teams responsible for developer tools. At the same time, the decision-making center for Gemini development is moving to the San Francisco Bay Area, where Google is consolidating its research work.

Reason for the Reshuffle

The change did not happen in a vacuum. After the launch of Gemini 3.1 Pro in February 2026, Google's subsequent releases stopped setting a new standard, while OpenAI and Anthropic kept pushing the capabilities of their models further. Analysts were split on how to read the August reorganization: some saw it as a strategic pivot toward the more profitable cloud business, others read it as an admission that Google was falling behind in the race for the best models.

Kavukcuoglu himself fits into this second narrative, though he tries to frame it positively. He stresses that the company has the team, the computing resources, and the infrastructure needed to close the gap with the frontier, and treats the current lag as a transitional state rather than a structural problem.

Gemini 4 as a Credibility Test

The most concrete part of his statement concerns Gemini 4. Google had previously said it had begun the most ambitious pre-training run for a base model in its history, and Kavukcuoglu confirms that the project requires a significantly larger base model than previous generations in order to keep pace with the rest of the field. The priority is meant to be capabilities related to coding and autonomous agents, not general conversation or content generation.

Kavukcuoglu also describes the development of the Flash series as an evolution from classic language models toward coding agents, and calls software engineering the most important domain for future AI systems. That aligns with the direction competitors have taken: in 2026, both OpenAI and Anthropic invested heavily in tools for developers and code-writing agents, treating them as an arena for direct competition with Google.

What This Means for the Market

For companies using Google's models, the new leadership's statement effectively signals a faster release pace and a stronger push into developer tools, an area where Google has so far lagged behind Anthropic and OpenAI. Pichai himself had previously publicly acknowledged that Google was losing to the competition specifically in the AI coding category, which makes Kavukcuoglu's statements a direct response to that weakness.

For Polish companies and developers using Gemini through Google Cloud, this means the coming months will likely bring a series of model updates focused on coding and agentic work before the full version of Gemini 4 arrives. No official release date for Gemini 4 has been given.

Share: