Anthropic's heavy use of AI agents is changing not just how its employees work, but the kinds of employees it hires. In an internal employment report published Oct. 16, the company said the share of junior-level employees in technical departments fell from 21% to 14% since early 2025. Junior workers' share of technical hires dropped more sharply, from 27% to 9%.
The report does not describe an absolute retreat from technical hiring. Anthropic said technical headcount continued to grow as the company expanded, but sales, finance and recruiting grew faster. Technical departments' share of total employment fell from 61% to 56% over the previous year.
Anthropic based the report on internal employment, AI-use, output, survey, communication, calendar and hiring records, supplemented by 17 employee interviews. That gives the company an unusually detailed view of its own workplace. It also makes the report a company account that relies on Anthropic's own definitions and measurements.
Agent hours outpace employee hours
The company said employee-directed interactive AI-agent hours reached six times employee work hours by August 2026. That is a comparison of hours, not a count of agents and people. Anthropic's calculation pools subagents, counts inference and active tool use, excludes waits longer than 30 minutes, and estimates human work using headcount and a 40-hour week.
Output measures rose alongside agent use. Code commits per engineer reached about five times their level six months earlier, according to the report, while lawyers processed more contracts, accounts payable handled more invoices and sales representatives closed more deals.
Anthropic tried to separate AI use from its rapid business growth by comparing 192 engineers who adopted coding agents with 74 who did not during the study period. Code added by adopters rose sharply after adoption. The company also found no difference in selected removal and revert measures between code written directly by employees and code written by their agents.
Those results come with important limits. Anthropic acknowledged that adoption was not random, and engineers who adopted fully autonomous agents were already on a steep upward output trend. The report also cautioned that more code does not necessarily mean better code or greater business impact, while its quality checks captured only a limited set of possible problems.
Human attention becomes the bottleneck
More agent capacity did not remove the need for people. Anthropic said interactive Claude sessions spent about 80% of their wall-clock time idle while waiting for human input. Using a token-to-word conversion and an assumed reading speed of 250 words per minute, the company estimated that reading all agent output would consume more than half the workweek even for nontechnical employees.
