File: 4 October 2026 · Source date: September 17, 2026 · Anthropic report
What the new report says
Anthropic offers three measurement areas to monitor the AI development process: automation in research and development, supervision of agents, and distribution of computational resources. In the company's August 2026 measurement, Claude leads 26% of jobs with human supervision; it does not operate completely autonomously in any subtask group measured. These rates depend on the company's methodology and job descriptions; It does not show the situation of the whole sector.
How to read numbers?
The report describes the lack of common methodology and the risk of error in using models to evaluate their own models. Independent verification is therefore important. Our editorial assessment: automation rate alone is not a measure of accuracy, security or awareness. Whether a system can do more work and whether its work is more reliable are separate questions.
Okurun kontrol listesi
When evaluating a product announcement, write out the task, success criteria, and error cost separately. Gains from a coding role should not automatically be carried over into healthcare consulting. Who is the source, who did the testing, under what conditions was the comparison made? These questions provide a more permanent evaluation method than model names.
keep thinking
What decision would you leave to a research assistant; What decision would you require human approval on? Share your reasoning in the reply field below.
Resources and further reading
Anthropic — measuring development speed ↗Share your answer
You must log in to reply.
No response to this article yet.
Established science, active research, and philosophical speculation are clearly separated.