We don't think it will lead to broad RSI or explosive progress
5 August 2026 (AI as Normal Technology). Princeton researchers Sayash Kapoor and Arvind Narayanan, writing on AI as Normal Technology in August 2026, after testing AI agents on open-ended research tasks and finding them unable to match human researchers.
AI agents can't yet do open-ended AI research
AI and science · AI and hype · 2026
Who else?
What it refers to: Faster progress from AI agents hill climbing on narrow, verifiable tasks such as improving efficiency, versus recursive self-improvement (RSI), the automation of AI research by AI agents.
Other voices
- “We are pretty far from the end state of RSI” Sundar Pichai, Builder,
- “We have built a machine that builds new models” Alexandr Wang, Builder,
- “Trying to fully automate AI R&D deserves a "what the fuck" reaction” Helen Toner,
- “Reliability lags capability” Towards a Science of AI Agent Reliability,
- “AI systems can one-shot a superficially-plausible philosophy paper” Seth Lazar,










