Is Sergey Brin Back in Founder Mode? Google Bets on AI Self‑Improvement

Google’s AI competition is moving toward a more aggressive direction: letting AI help create the next generation of AI. Reuters reports that co‑founder Sergey Brin, though no longer holding a formal title, is using his influence to push model training and to rally key AI staff around the Gemini effort.

One specific direction Brin is championing is Recursive Self‑Improvement (RSI). In May, Anthropic published “When AI builds itself” outlining its view of RSI progress, and in July OpenAI rehired former DeepMind co‑founder Wen‑Li to lead an RSI‑focused team. Recently, former Google chief scientist Jeff Dean left to start a venture also targeting RSI, aiming for AI to run experiments, evaluate results, and iteratively improve itself.

Meanwhile, Google’s own Gemini flagship has been delayed by about two months, and internal tests show the new model still lags behind rivals in programming and other critical tasks. This delay reflects broader structural issues that have repeatedly slowed Gemini’s iteration speed, such as insufficient compute, disagreements among project leads, and the lengthy decision‑making processes of a large company.

DeepMind’s leadership is also shifting. On August 5, Google announced that Demis Hassabis would become chairman, reducing his day‑to‑day duties, while Koray Kavukcuoglu, previously Google’s chief AI architect, took over DeepMind’s operations and continued overseeing Gemini. Reuters cites insiders saying Kavukcuoglu now receives Brin’s backing and has become a key decision‑maker for model development and resource allocation.

Because AI training relies on TPUs and Google Cloud also needs TPUs, internal friction over compute allocation has grown. Some Cloud executives hope the new DeepMind structure will ease these tensions, and CEO Sundar Pichai has pledged further AI infrastructure investment to alleviate TPU constraints.

The core dilemma Google faces is that model capability is one thing, but the ability to quickly marshal people, money, and chips toward the right direction is another. Past Gemini delays and internal criticism illustrate how these organizational bottlenecks can cause a “catch‑up‑be‑overtaken‑catch‑up” cycle.

DeepMind’s identity is becoming increasingly “Google‑ized.” After a company‑wide meeting on August 6, Hassabis and Kavukcuoglu said daily R&D would not change much, yet non‑technical teams are moving under Google’s reporting structure. Since its 2014 acquisition, DeepMind has balanced basic research with product integration; today the balance is tipping toward Gemini and commercial deployment, raising concerns among senior researchers about future research autonomy.

Brin’s push for RSI reflects a broader shift in Google’s AI strategy: not only to regain competitive rankings for Gemini but also to speed up the R&D process itself. Google has already experimented with adjacent approaches, such as AlphaEvolve, which uses large models and automated evaluation to search for better algorithms and to optimize engineering tasks. While AlphaEvolve is far from a fully autonomous self‑improving system, it demonstrates a possible model where AI first assists researchers and engineers, then gradually takes on experiment design, coding, optimization, and validation.

Other labs—including OpenAI, Anthropic, and several AI startups—are exploring similar AI‑assisted research pipelines. The industry conversation is moving from “Can AI write code?” to “To what extent can AI participate in building the next generation of AI?” However, most current systems still require humans to set goals, allocate compute, select experiment outcomes, and decide on deployment, so fully open, autonomous recursive self‑improvement remains unrealized.

Compared with Brin’s earlier “founder‑mode” comeback that focused on catching up to a single competitor, this round targets a deeper problem: whether AI research itself can accelerate in an environment where model capabilities improve rapidly.

What do you think of Google’s heavy bet on RSI and Brin’s renewed involvement? Share your thoughts in the comments.