Brin says Google is nearing a self-improvement milestone

Google cofounder Sergey Brin says the company is moving toward recursive self-improvement, or RSI, as its AI models increasingly help build their successors. Speaking during a roundtable about Gemini 4, known as Argon, Brin said researchers and models now work together on the next version. He pointed to earlier Flash releases as evidence that development was accelerating.

Google says parts of its AI research and development cycle are becoming automated. AI agents write a majority of production code and bug fixes, although changes remain subject to human review before submission. Brin has pushed for greater focus on RSI and said internal teams would be a couple of models ahead by the time viewers watched the Argon release discussion.

Employees have also been testing Carbon, a more capable internal version of Argon. Some staff compared its coding performance with Anthropic’s Opus 5.5. Google DeepMind RSI leader Vihan Jain said demand for Argon access in post-training work showed how deeply the model had become embedded in development.

Anthropic and OpenAI have also suggested RSI may be approaching. Most researchers interviewed by Gladstone AI expected it to begin within the next 3 to 12 months, potentially sooner without government intervention. AI researchers have warned that insufficient guardrails could leave humans unable to control the technology.