On October 16 Anthropic published a report about itself: how has work changed now that agents log more hours than people? The underlying data is unusually wide. Headcount, agent session telemetry, revenue and output metrics, employee surveys, calendar data, hiring decision records, plus 17 interviews.
The juniors go first
The share of junior roles in technical departments has fallen from 21% to 14% since early 2025. Among new hires the drop is sharper: 27% down to 9%. Postings for junior positions have been falling hard since December 2025.
Internally the subject surfaced in January 2026, first in the teams that lean on Claude most in their own work. By April, engineering and research had pulled back on hiring at the lower bands. A third of ML managers and a fifth of SWE managers said in a survey that they were hiring less because of AI. One quote in the report captures the moment: the uplift for performance engineers, in particular, changed overnight what they were looking for.
Meanwhile the technical share of the company as a whole is shrinking, from 61% to 56% over the past year. Sales, finance and recruiting are growing faster. Some of that is just a company scaling its business functions. Some of it is non-engineers now building their own software. One sales rep describes a custom dashboard with pipeline integrations that took three or four hours to put together.
Five times the commits, and no drop in quality
In March 2026 total agent work hours passed total human hours for the first time. By August the ratio was 6 to 1, counting only the agents individual employees start themselves.
Output tracks it. Commits per engineer are five times what they were six months ago. Lawyers process more contracts, accounts payable pushes through more invoices, sales closes more deals. Anthropic is growing fast anyway, and I wrote about the Q3 revenue two days ago—selling a strong model makes deals easier. So the report compares 192 engineers who adopted coding agents against 74 who did not during the period. After adoption, the amount of code they add climbs sharply.
And quality? The comparison holds the person, the month and the directory fixed, and contrasts code the employee wrote directly with code their agent wrote. On lines later deleted or commits later reverted, there is no difference.
Human attention is the new bottleneck
One engineer describes facing “an army of Claudes, with every one asking for my attention”. Even non-technical employees would need half their work week just to read everything their agents produce. The response is meta-work: building visibility tools and monitoring agent pipelines.
Hiring decisions show it too. Managers now weight autonomy and judgment above domain depth and implementation skill.
Anthropic is not a typical employer, which is exactly why this is worth reading
The company has built its workflows entirely around AI and sells the model it runs on. Generalising to the wider economy would be a mistake. As an early indicator it still holds up, and the uncomfortable detail sits in the first chart: the number of engineers is not falling, the entry point is. Anthropic responded with an early warning system for jobs at risk and a push toward internal redeployment. How other employers handle it is an open question.