OpenAI's AI Coding Agents Now Finish Nearly Half of Complex Research Jobs
Original headline: Research acceleration: The jagged frontier — openai.com
More in AI and machine learning news
OpenAI released a new report detailing how its AI coding agents are being used internally to accelerate AI research and development. The key finding is that these agents have become significantly more capable, now completing 46% of research tasks estimated to take humans two to four hours, up from 23% in March. However, success rates vary wildly depending on the task, ranging from 79% for monitoring training runs to just 25% for writing inference code, illustrating what OpenAI calls the 'jagged frontier' of automation.
The impact is a dramatic increase in both agent usage and researcher productivity, with median daily API spending per researcher hitting $2,400 and heavy users pulling further ahead. Notably, OpenAI reports the largest absolute increase in documented research effort is now directed toward Safety and Alignment work, which grew by 77% this year. The takeaway is that while AI is rapidly automating parts of its own creation, skilled human researchers remain essential for steering the process, especially as the stakes get higher.
Read the original at openai.com
Satirical commentary on real tech news, summarised with AI. Read the original source for the facts. How ZeroDaily works
