Sundar Pichai says Google is far from the ‘end state’ of self-improving AI
Researchers and AI collaborate
Pichai described recursive self-improvement as a process in which AI systems assist researchers in developing models, rather than a fully automated cycle. He said AI agents are already helping researchers work together on model building.
He expects this collaboration to continue advancing into next year. Pichai said researchers and models now work hand in hand on the next model version.
Different timelines for progress
A Gladstone AI report published earlier this month found that most researchers it interviewed from companies including Anthropic and DeepMind believed recursive self-improvement would begin within three to 12 months. Some thought it could arrive sooner without government intervention. Pichai’s assessment was more cautious: he said Google was still far from the end state of recursive self-improvement.
AI’s role in Google’s work
Pichai said AI now deeply influences every part of the code Google touches. He said Gemini 4 Argon would help people tackle harder problems, and reported a one-third decrease in code rollbacks—reversions caused by bugs or system issues. Pichai also told the Sources podcast this week that he uses AI to track Google’s model development and internal discussions.
Key points
- Pichai says Google remains far from fully self-improving AI.
- He expects AI and researchers to keep developing models together into next year.
- Gladstone AI reported that many researchers expect recursive self-improvement within three to 12 months.
- Pichai cited a one-third decrease in code rollbacks at Google.