Gemini 3.5 Pro Delayed, Talent Continues to Leave: Google’s New AI Chief Takes Over a "Mess"

Google's AI business is undergoing a leadership change, but the hand dealt to the new chief is far from easy.

According to reports, Koray Kavukcuoglu will succeed DeepMind co-founder Demis Hassabis as Senior Vice President of Google DeepMind, reporting directly to Google CEO Sundar Pichai.

Currently, Google's flagship model, Gemini 3.5 Pro, has been delayed in its release. Issues such as shortcomings in programming capabilities in the competition for frontier models and the loss of high-end talent are all laid before this new leader.

For Kavukcuoglu, the new position means more than just a transfer of power; it is a "catch-up" lesson in AI competitiveness. With OpenAI and Anthropic continuously pushing the boundaries of frontier model capabilities, Google needs to prove as soon as possible that Gemini still has the ability to stand in the industry's first tier.

Falling Behind in Frontier Models, Programming Becomes the Biggest Shortcoming

The pressure Google faces in the AI frontier competition is transmitting from model release schedules to the market.

As previously reported by Bloomberg, Google postponed the release of its flagship model, Gemini 3.5 Pro, because the model still requires further improvement in key areas such as programming. This news had previously triggered market concerns and was one of the factors contributing to the decline in Alphabet's stock price in July.

Google has had its moments in the spotlight before. When Gemini 3 was released in November 2025, the market once believed it pushed frontier model capabilities to "new heights." However, after entering 2026, the competitive landscape changed again.

Malik Ahmed Khan, a senior equity analyst at Morningstar, stated that one important reason Google lags behind Anthropic and OpenAI is that the company invested more resources in commercialized areas such as search and multimodal directions, missing out on taking the lead in forming clear killer applications in programming scenarios.

In his view, Anthropic and OpenAI are already "more than a body length" ahead of Google in programming capabilities.

Google refuted the claim of "deviating from the main AGI line." The company stated that the path to AGI will not be limited to a single direction but will cover multiple fields such as agents, programming, robotics, and world models. To strengthen its programming capabilities, Google also formed an internal team called "Code Strike."

However, for the current model competition, how to translate these directions into faster and stronger products remains a question Google must answer.

Hassabis Hands Over the Baton, AI Strategy Shifts to "Execution"

Kavukcuoglu is not an external hire.

He joined DeepMind in 2012, prior to Google's acquisition of DeepMind in 2014. He subsequently served as DeepMind's Chief Technology Officer and Google's Chief AI Architect, making him one of the most senior technical managers in the company's AI business.

In fact, this transition did not happen suddenly. According to reports citing insiders, over the past year, Kavukcuoglu has gradually taken on more responsibilities from Hassabis, including model development and major Gemini product launches.

This formal succession is seen by the market as a signal that Google's AI business is shifting further from "research-driven" to "product execution."

Ray Wang, chief analyst at Constellation Research, believes that Kavukcuoglu's promotion means Google is placing greater importance on execution. In the future, we may see a faster product release cadence, more experimental explorations, and a more focused product roadmap.

In contrast, Hassabis previously emphasized an AGI path that transcended traditional large language models. Khan stated that during Hassabis's tenure, significant attention was paid to multimodal, world models, and health-related projects. With Kavukcuoglu taking over, "enhancing LLM capabilities will become a clear path forward."

This means the primary task facing the new leader may be quite clear: first, pull Gemini back into the competition for frontier models.

Talent Attrition, New Chief Must Also Fix the "People" Problem

Beyond model performance, another challenge Kavukcuoglu faces is talent.

Over the past period, Google's AI team has seen continuous departures of high-end talent, leading outsiders to worry about whether its programming and pre-training capabilities have been affected.

Nick Patience, head of AI at Futurum Group, offered straightforward advice: the first step is to deliver Gemini 3.5 Pro as soon as possible and demonstrate through a stable and predictable release cadence that this model is not a one-off product breakthrough. Meanwhile, Google also needs to rebuild its teams of programming and pre-training experts.

In other words, what Google needs to solve is not just "when a model will be released," but how to rebuild the talent and organizational capabilities required for continuous iteration of frontier models.

However, some believe that market concerns about talent attrition may be exaggerated.

Sebastian Mallaby, author of The Infinity Machine, stated in an interview that it is not objective to judge that Google's AI business is in overall crisis based on the departure of a few well-known figures. The Gemini team consists of thousands of people, and the departure of individual core figures should not be overinterpreted.