saved
How employment is changing at Anthropic
Hraness wrote this summary from a saved copy of the source. Quotations are taken word for word from the source.
gist
Nathan Wilmers and three Anthropic colleagues use internal data to show how AI agents are reshaping work and hiring at Anthropic. They report that junior technical hires fell from 27% to 9% of the total while agent hours grew to six times human hours by August 2026. Code commits per engineer reached 5x their pace of six months earlier, and output rose for lawyers, salespeople and other staff as agents spread. Human attention became the bottleneck, so employees moved into meta-work and hiring shifted toward autonomy and judgment.
ideas
- Junior technical hiring is shrinking. The authors report the junior share of technical headcount slipped from 21% to 14% since early 2025 and of technical hires from 27% to 9%, while administrative functions stayed small as the company tripled in size.
- Agent work now outweighs human work. Agent session hours passed human work hours in March 2026, and employee-directed agents ran six times their handlers' hours by August. Coding sessions by non-technical staff rose from 2 a month to 90.
- Output rose within the same person after adoption. Comparing 192 engineers who adopted coding agents with 74 who did not, the authors find code added rises sharply after adoption, with no difference in removal or revert rates for agent-written code. They caution that adoption is not random.
- Human attention is the bottleneck. Interactive sessions spend about 80% of their time waiting on people, reading agent output would take over half a non-technical employee's week, and meta-work grew from 10% to 22% of sessions. Median meeting time is up 20%.
- The authors propose sharing gains and redeploying staff. They point to broad stock ownership, an early warning system for jobs at risk and internal reallocation, while conceding none of this helps people who are never hired.
quotes
“This acceleration makes human attention the bottleneck for agent output.”
“The need to feed AI context requires observability of human work and decisions.”
“Employment protections for incumbents will not benefit those not hired to begin with.”