Anthropic's research highlights how AI agents are dramatically altering the nature of work, with employees effectively logging six hours for every human hour worked. A significant challenge is the immense volume of information generated by these agents, leading to a "reading tax" that consumes much of a week's time. This shift is transforming workers into managers of AI systems rather than direct producers. To address this, Anthropic suggests practical steps to keep humans in control.Measuring time spent reviewing agent output is crucial, as a wealth of information can overwhelm attention. Agents should be designed to rank their requests by severity, similar to alarm management systems, so routine tasks don't demand immediate human intervention. The review process itself needs careful design, clearly showing changes and justifications to prevent automation bias.It's essential to document decisions in accessible formats, referred to as "context engineering," so agents can access and utilize this information effectively. The role of managing AI agents, or "meta-work," is growing, requiring skills in specifying tasks and overseeing agent interactions. Hiring for judgment and analytical thinking is becoming paramount, as AI handles more routine tasks.Finally, Anthropic's findings raise concerns about entry-level roles, as junior positions have declined significantly. Rebuilding these roles around agent review could provide necessary training and experience for future AI managers. This evolution mirrors past industrial agreements, suggesting a need for new frameworks to manage the human-AI workforce.