
What happened
Anthropic's study of its own workforce found 80% of elapsed Claude session time was agents waiting on human input, and by August agent hours outnumbered human hours six to one.
Why it matters
Employees are producing more and moving faster, but human attention is now the constraint on agent output, so staff increasingly act as managers overseeing multiple AI agents.
WHO IT HITSAnthropic's own employees, including non-technical staff who now run personal agents for email and calendars, are affected first; other companies may see similar shifts as they adopt agents, though the report notes Anthropic's resources may make it atypical.
Summaries like this, in your inbox every morning.
Anthropic's report offers an early look at how an AI-heavy company actually operates, based on its own internal data rather than outside observation. The central finding is a reversal of the usual concern: rather than humans waiting on slow AI, the AI agents are fast enough that they spend most of their session time waiting for a person to respond. That has turned many employees into supervisors of multiple agents, a kind of work the report calls meta-work.
The report also documents how roles are shifting. A product design team that once wrote short website copy began using AI for that task this spring, then rotated into writing prompts and model evaluations for Claude. Anthropic says employees with at least a year at the company were 50% more likely to move across departments each month in 2026 than in 2025, and that it has tried to identify at-risk jobs and reassign people.
Hiring is changing alongside this. Managers now weigh autonomy and judgment more heavily than specific technical expertise, and the share of junior technical staff has fallen sharply. The report acknowledges that Anthropic's resources and its deep integration of AI into workflows may make it an outlier, so its experience may not predict what happens at other companies.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Tenstorrent, a 10-year-old Nvidia challenger focused on inference, raised $1.5 billion at a $6.2 billion valua…

PNY is offering NVIDIA RTX PRO 6000 Blackwell Workstation Edition and Max-Q Edition GPUs, configurable with up…

MiniMax released M3.1 Flash, which scores 42 points on the Artificial Analysis Intelligence Index, 13 points a…

The piece lays out IBM's five-type classification by mechanism — simple reflex, model-based reflex, goal-based…

AINOW published a comparison of 17 AI agents as of October 2026, covering pricing, free-trial availability, an…

The Nikkei column says the scramble for capital it flagged in June has become fiercer, with the 10-year US Tre…
