Newsletter Subscribe
Enter your email address below and subscribe to our newsletter
[forminator_form id="25163"]

openai.com.openai.com.openai.com.On Nov. 18, OpenAI said its coding agents now complete nearly half of mid-length research tasks with no human help. It added that the gains are uneven and that people "remain a core part of the AI research loop." In its report, Research acceleration: The jagged frontier, the company estimated that the agents' zero-intervention success rate on tasks needing two to four hours of human work rose from 23% in March to 46% in October. On tasks estimated at four to eight hours, the rate rose from 13% to 35%.openai.com
OpenAI also renewed its call for AI labs to be required to track and publicly report indicators of recursive self-improvement, meaning AI speeding up the development of future AI. It said its own reporting "should not be read as an intention to pursue full RSI".openai.com
Results varied widely by type of work. From Aug. 1 to Oct. 18, agents succeeded without intervention 79% of the time when monitoring training runs, 35% when launching them and 25% when writing code for inference and serving stacks. The company cautioned that researchers may have changed which tasks they hand to agents within each duration band, so rising success rates do not reflect model improvements alone. The task durations describe estimated human effort, not how long the agents ran.thedefiant.io+1
Agent use grew quickly. Valued at API list prices, the median researcher used about $2,400 of agent work per day in October, up 62% from September, and the 90th percentile used about $23,000, up 105%. Agent workdays per researcher, with each eight hours of agent use counted as one workday, rose elevenfold from 0.5 to 5.7. In a survey of 101 research staff, the median respondent said they would give up 30% of team compute, or 40% of team headcount, to keep frontier coding agents for three months.openai.com
Interviewed researchers said agents still showed poor research taste and produced "AI slop" pull requests that needed close oversight. One described the workflow this way: "Go make one. I'll review it and make edits. Then go do the rest".openai.com
Documented safety and alignment workweeks rose 77% from January to September, the largest absolute increase of any research area. OpenAI linked the acceleration to the aftermath of a Hugging Face incident in July. OpenAI researcher David Holtz wrote on X that researchers are "devoting more and more time to safety and alignment."x.com+1
The disclosure push builds on OpenAI's June governance blueprint. That plan proposed federal rules requiring transparency reports on RSI tracking and asked the U.S. Center for AI Standards and Innovation to develop methods for measuring it. The report itself notes that more code and more experiments "do not necessarily imply faster research progress".openai.com+1