On 16 October Anthropic published an internal employment report. By August 2026, interactive agent hours started by employees were six times employee hours. In March, the combined hours of autonomous and interactive agents passed human hours for the first time. These are hours, not a count of agents.

[1]
A stack of closed folders with charcoal covers and one muted-blue spine
A stack of closed folders, the covers painted charcoal and one spine muted blue, with pale paper edges., AI-generated illustration, not a news photograph

The figure note sets the rules. Only employee-directed interactive sessions count, and subagents are pooled with them. Agent time includes inference and active tool use. A wait longer than 30 minutes is excluded. Human hours are headcount times an assumed 40-hour week.

[1]

At five years old, Anthropic takes in 2 to 4 times as much revenue per employee as Meta and Google did at the crest of their own rapid growth. The report states revenue divided by employees. The body gives no dollar total and does not call the figure profit.

[1]

The share of junior employees in technical departments fell from 21% in early 2025 to 14%. The share of junior technical hires fell faster, from 27% to 9%. About a third of machine-learning managers and a fifth of software-engineering managers said they hired less because of AI. Technical departments fell from 61% to 56% of the company. Absolute technical hiring still grew. Sales, finance, and recruiting grew faster.

[1]

Code commits per engineer reached five times the pace of six months earlier. That is a company-wide description over time. Separately, the report follows 192 engineers as they adopt coding agents, against 74 who do not adopt during the study. After adoption, the code those engineers add rises sharply. For the same person, in the same month and directory, agent-written commits and human-written commits show no difference in how often lines are later removed or reverted. Adoption is not random. The report says the estimate is biased if someone adopts because a large project is about to start.

[1]

Even for non-technical employees, reading the output of their own Claude agents could take half a work week. Meta-work in interactive sessions, such as writing instructions, monitors, and checks on other agents, rose from 10% of the average employee’s sessions in fall 2025 to 22% in spring 2026. As of August, Anthropic employees expected AI to be able to do 75% of their work by 2030, against 28% for other Americans in the report. That is an expectation from that month, not a replacement that has already happened.

[1]