6 minBusiness
Anthropic Study Finds Human Attention Is the Bottleneck in an AI-Run Workplace
Anthropic's internal study of its own workforce found that AI agents now outnumber human working hours six to one, but employees increasingly spend their days managing bots rather than doing the work themselves, making human attention the limiting factor on output.
Anthropic has published a study of its own workforce that offers an early look at what an AI-dominated company may look like in practice, and the central finding is that human attention, not computing power, has become the constraint on how much work gets done. According to the report, 80% of the elapsed time in Claude sessions was spent with the AI agents waiting on a human to respond, a dynamic the company described as making human attention the bottleneck for agent output.
Employees are no longer simply doing the work themselves. Even staff members who do not supervise anyone are increasingly acting as managers of multiple AI agents that complete tasks but periodically require direct human input. One employee described the experience as managing «an army of Claudes, with every one asking for my attention.»
The bottleneck is the flip side of a sharp rise in output. Anthropic found that AI agents collectively surpassed human working hours for the first time in March, and by August agent hours outnumbered human hours six to one. Software engineers' code commits increased fivefold over the past six months. Non-technical employees now have personal agents monitoring their email and calendars, compiling to-do lists from messages, and briefing them on the purpose of upcoming meetings.
Those changes have removed some manual labor, but they have also reshaped the workday. Managing agents is becoming a job in itself. The share of employees' interactive Claude sessions devoted to this kind of «meta-work» rose from 10% to 22%, according to the report. For non-technical employees, simply reading everything their agents produce would consume roughly half the work week. The report also noted that private work conducted in unlogged meetings and direct messages is «invisible, confusing and discouraged in an AI-heavy workplace.»
Meetings are taking up more time as well. The median employee now spends 20% more time in meetings than a year ago. Part of that reflects the company's growth, with Anthropic now employing more than 3,500 people, and part reflects an emphasis on hiring in non-technical roles. But meeting time is up 16% even among technical individual contributors, a sign that humans are devoting more of their day to coordination and decisions while agents handle tasks such as repetitive coding on their own.
The pace continues to ratchet up. As soon as a new model becomes available, Anthropic employees start tackling harder problems, the report found. One employee described an «eternal pressure at Anthropic to move as fast as possible and use Claude for everything.»
In some cases, employees have used AI to streamline their own responsibilities away. A product design team that previously wrote short pieces of copy for the company's website, including greetings displayed on Claude.ai, began using AI to write the copy this spring. The team then rotated into agent prompting and began writing model evaluations and prompts for Claude. The company said employees who have worked there for at least a year were 50% more likely to move across departments each month in 2026 than in 2025, though it said it has tried to identify at-risk jobs and reassign employees to other positions.
Hiring is changing too. Managers increasingly emphasize applicants' autonomy and judgment over specific technical expertise, and fewer entry-level people are being hired. The share of junior-level employees in technical departments fell to 14% from 21% in early 2025, while the share of junior technical hires dropped from 27% to 9%. Open positions for those junior roles have also plummeted since December, a trend likely to add to anxiety about AI job displacement, which is running especially high among young people. A survey from recruiter platform iCIMS published earlier this year found that 51% of Gen Z respondents considered AI the greatest threat to their job security.
Anthropic employees expect AI to take on more of their daily responsibilities over time. As of August, they said they expect AI to be able to complete three-quarters of their work by 2030, compared with 28% predicted by other U.S. workers in a separate Anthropic study. It is unclear whether the changes at Anthropic reflect how the average workplace will evolve. The company has massive resources and is nearing a potential $2 trillion IPO that could break records, and it has embedded AI in its workflows more deeply than most employers have. Still, the report describes a transformation in which the strategic portion of the job, once a small slice of the workday, becomes most of it. As one employee put it, «that 20% that was strategic becomes 90% of the job.»
2
