At the same weekly commit count, the 90th-percentile engineer burned 39 to 240 times the coding-agent tokens of the 10th. That is from 404 JetBrains engineers over five weeks. Within one person, doubling spend went with about 10% more commits, and work metrics explained 38% of the spread between people. The authors' advice: treat token spend as resource accounting, not productivity.https://arxiv.org/abs/2611.13396#AI #AIAgents #Programming #DevOps
Related
RE: https://flipboard.social/@newsguyusa/117605070767904233If the #INVIDIA guy can’t have affairs, you can’t have #AI ma...
RE: https://flipboard.social/@newsguyusa/117605070767904233If the #INVIDIA guy can’t have affairs, you can’t have #AI mail, meaning you can’t have mail at all. If the #INVIDIA guy ...
Google Launches Gemini 4 Hydrogen AI Model for Search and DevelopersGoogle has introduced Gemini 4 Hydrogen, a new AI mo...
Google Launches Gemini 4 Hydrogen AI Model for Search and DevelopersGoogle has introduced Gemini 4 Hydrogen, a new AI model for agentic workflows, now available across Google AI St...
A new version of nnetsauce, and a new Techtonique websitehttps://thierrymoudiki.github.io/blog/2020/09/11/misc/python/qu...
A new version of nnetsauce, and a new Techtonique websitehttps://thierrymoudiki.github.io/blog/2020/09/11/misc/python/quasirandomizednn/nnetsauce-new-website#Python #DataScience #M...