Sundar Pichai says Google is far from 'end state' of self-improving AI

Key facts

• Google CEO Sundar Pichai believes fully self-improving AI is still distant.
• AI agents are currently assisting Google researchers in model development.
• Google cofounder Sergey Brin sees AI assisting in building future models.
• A report suggests recursive self-improvement in AI could emerge within months.
• 75% of new code at Google is AI-generated, leading to more productive and better code.

Google CEO Sundar Pichai stated that the company is still a long way from achieving fully recursive self-improvement in AI. Speaking on the "Big Technology Podcast," Pichai explained that Google researchers are using AI agents to assist them in building new AI models. He anticipates this progress will continue into the next year.

Earlier this month, Google cofounder Sergey Brin commented that the company's Gemini 4 and Flash models demonstrate AI's role in its own development, suggesting Google is "starting to get onto that RSI treadmill." A report by Gladstone AI indicated that many researchers believe RSI could emerge within three to 12 months.

Pichai also noted that AI is deeply influencing code generation at Google, with 75% of new code now AI-generated. He added that AI tools are leading to more productive and higher-quality code, evidenced by a one-third decrease in code rollbacks.