Thesis: Anthropic’s account suggests that agent adoption can shift the constraint from producing work toward allocating human attention to oversight.
Mechanism: As agents take on more execution, employees spend more time monitoring and reviewing their work, shifting organizational attention toward oversight. The company says agents worked more hours than employees by March 2026 and six times as many hours by August. It also reports more code per person, more sales closed per sales representative, and more contracts processed; it cautions that code volume is an imperfect productivity measure. Anthropic says employees increasingly spend time building ways to monitor agent work, while its hiring has shifted toward a smaller share of junior positions. The reported sequence is consistent with a workflow in which more delegated execution increases the need to inspect, route, and own outputs.
Why now: The report puts a date and scale on Anthropic’s internal agent adoption and pairs it with a survey showing different expectations among staff and the US public. Its companion survey reveals a sharp gap: 91% versus 39% expect a Nobel-level AI discovery by 2030, and more than half of staff versus 5% of US adults expect AI to perform at least 80% of their current work.
Implications: Teams adopting agents may need to plan for review capacity and role design, not only faster production. Anthropic’s account may be an early signal for AI-native organizations, but it does not establish that other employers will see the same results. Evidence boundary: These are company-reported workplace observations and survey responses; they do not prove economy-wide productivity effects or causation. Watch: Whether Anthropic or independent follow-up research measures the quality, cost, and review burden of completed agent-assisted work across other organizations.