Strategic Report  ·  2026-11-20

Research acceleration: The jagged frontier

Strategic ReportMedium impactGlobal
OpenAI published internal measurements of how coding agents are accelerating its own AI R&D — a core indicator of recursive self-improvement (RSI). Agent success on tasks estimated to require two to four hours of human work rose from 23% to 46% between March and October 2026, and from 13% to 35% for four-to-eight-hour tasks; autonomous success was 79% for monitoring training runs, 35% for launching training runs, and 25% for writing inference-and-serving code — a jagged uplift with humans still central to the AI research loop. Median researcher daily agent usage reached $2,400 in October, up 62% from September, while the 90th percentile hit $23,000 (up 105%); the median researcher said they would trade 30% of team compute or 40% of team headcount to keep frontier coding agents for three months. Safety & Alignment saw the largest absolute increase in documented research effort across all of OpenAI's major research areas.
Frontier labs' own RSI indicators are becoming the key input to compute-governance and 'deceleration' policy debates; executives should work from the measured numbers rather than the competing claims.
Brief the board on the measured pace of AI-accelerated R&D and set an informed position for compute and RSI policy discussions.
OpenAI — Research acceleration: The jagged frontier
See this in the live feed Explore related AI security and governance findings — updated every morning.
Open the feed →