Anthropic study tracks a shrinking share of junior technical hires
Anthropic published two reports on October 16 examining how AI is changing work inside its own company and how its employees expect the wider economy to change. The research announcement describes a smaller share of junior hiring and growing pressure on employees to supervise agent output.
Junior hiring changes within a growing company
The employment study reports that junior employees’ share of technical departments fell from 21% to 14% since early 2025. Juniors’ share of technical hires dropped from 27% to 9%. Technical hiring nevertheless continued growing in absolute terms, so the changing shares should not be read as evidence of layoffs.
By August, employee directed interactive agents accumulated six times estimated human work hours. The measure counts inference and active tool use, excludes waits longer than 30 minutes and assumes employees work 40 hours weekly. It measures runtime rather than agent headcount or a sixfold productivity increase.
The study combines internal data with 17 employee interviews. It finds higher output following coding agent adoption, but adoption was not randomized. Employees starting larger projects might also be more likely to adopt tools. Company growth complicates the interpretation, and more code does not necessarily establish better code or greater business value.
The study estimates that reading all agent output would take nontechnical employees more than half a workweek. That calculation uses assumed reading speeds and should not be mistaken for measured time spent reviewing output.
Different expectations for the same decade
The expectations report compares August surveys of 10,980 US adults and 1,334 Anthropic employees. Staff expected AI to be capable of handling 75% of their current work hours by 2030 on average. Employed US respondents put that share at 28%.
These figures describe respondents’ beliefs about future capability. They do not measure future job losses or establish a company forecast. Within both groups, people who expected AI to handle more of their own work also anticipated higher unemployment.
The methods matter. Morning Consult’s US results were weighted, but its online panel was voluntary. Anthropic’s internal poll was unweighted. Staff used a slider to estimate their future work exposure, while US respondents selected ranges that researchers converted to midpoints. Half of the US sample received questions about 2027 rather than 2030, while all staff were asked about 2030.
At the same time, 95% of surveyed Anthropic staff expected company headcount to grow over the next year. Those personal expectations are not hiring plans. The responses show that expected task automation can coexist with expectations of business expansion.
Human attention becomes a practical constraint
Anthropic’s overview says employees increasingly build tools to monitor the work their agents produce. It presents the company as a possible early indicator of changes in work organization, while recognizing that a frontier AI lab differs from traditional businesses. The findings describe this workplace, leaving broader effects to be tested elsewhere.
Illustrative archival workplace photograph published in May 2024. The people shown are unrelated to the study. Photo by Compagnons on Unsplash, used under the Unsplash License.
