Win McNamee/Getty Images
- Sergey Brin said Google's Gemini models are increasingly helping to build their successors.
- Recursive self-improvement, or RSI, is the next big race in AI.
- A Google safety report said a "majority" of code was being produced by AI agents.
Google may be approaching a pivotal moment in AI.
In a roundtable discussion about the creation of Google's new Gemini 4 model, Argon, company cofounder Sergey Brin said that AI development was accelerating internally and that models are now helping build their successors.
This concept, known as recursive self-improvement (RSI), means AI helps build better AI systems, which can set off a flywheel that speeds up the pace of building new models.
Brin pointed to the releases of Google's Flash models before Gemini 4 as evidence that this was happening, and said the company was "starting to get onto that RSI treadmill."
"And now our researchers and the model work hand in hand to make the next version of the model," he said during the roundtable, which Google shared on Wednesday.
Anthropic and OpenAI have also said that AI may be on the verge of RSI. A new report from the firm Gladstone AI, published this week, outlines the ways AI companies are moving toward RSI. Most researchers — who work at companies including Anthropic, OpenAI, and DeepMind — interviewed for the report believe RSI will start within the next 3 to 12 months. It could come sooner, they said, assuming there was no government intervention.
In a safety report for its new Argon model, Google said that some components of the AI research and development cycle were getting automated. "AI agents now operate as high-leverage assistants that author a majority of production code and bug fixes, though code changes remain subject to human review before submission," it noted.
RSI was at the heart of warnings from AI researchers last month that humans could lose control of the technology if sufficient guardrails weren't in place.
'The RSI flywheel has reached a new gear'
Brin has been increasingly involved in Google's Gemini efforts and has leveraged his power as cofounder to shift resources to projects he deems most effective, Business Insider previously reported. This has included pushing for more focus on RSI.
During the roundtable, Brin said he was excited by the cadence teams were seeing at Google, and that by the time people were watching the video about Argon's release, they would be a couple of models ahead internally.
Business Insider previously reported that Google employees had been testing a superior version of Argon, internally known as Carbon, which some staff had compared to being on par with Anthropic's Opus 5.5 model for coding tasks.
Others at Google appear to share Brin's perspective. In a post on X, Vihan Jain, an RSI leader for Gemini at Google DeepMind, remarked on how embedded the Argon model had become in its own evolution.
"When your post-training demand shifts from asking for raw chips to asking for Argon quota, you know the RSI flywheel has reached a new gear," he wrote.
Have something to share? Contact this reporter via email at hlangley@businessinsider.com or Signal at 628-228-1836. Use a personal email address and a non-work device; here's our guide to sharing information securely.