Google is currently utilizing its advanced AI models to assist in the development of their successors. Co-founder Sergey Brin confirmed that the company has adopted a recursive self-improvement (RSI) approach, where current models work alongside researchers to build future versions.
The company has already implemented automation in certain research and development tasks. AI agents act as high-performance assistants, generating significant portions of software code and fixes, which are subsequently reviewed by human developers before final approval.
Industry peers, including OpenAI and Anthropic, are also nearing the implementation of recursive self-improvement. According to developers surveyed by Gladstone AI, this practice is expected to become standard within the next 3 to 12 months, provided that government regulations do not impose delays.
Sergey Brin is personally overseeing the Gemini project and prioritizing the RSI strategy. Reports indicate that Google is already testing a new model, Carbon, which is reportedly capable of matching the coding performance of the Anthropic Opus 5.5 model.
