Overview
- This week a Gladstone AI report and comments from Google leaders updated the picture by describing meaningful automation in the research loop and saying industry teams are actively building tools that let models help design successors.
- Google has reported that AI agents now author a majority of production code and bug fixes for some projects, though those changes remain subject to human review before release.
- Gladstone’s interviews with company insiders found many researchers expect recursive self‑improvement could begin within three to twelve months if there is no government intervention.
- Leading researchers stress that full, human‑independent recursive self‑improvement has not been achieved and that current systems mainly automate routine, iterative tasks known as “day science,” not higher‑order problem selection or conceptual breakthroughs called “night science.”
- A wave of startups and DeepMind/OpenAI spinoffs are chasing automated research, and experts warn that concentrated compute access plus rapid automation could speed capability gains and complicate regulatory oversight and safety checks.