AI agents need managers, not just users—companies lag in training

The article presents a critical shift in how companies must approach AI in the workplace. For the past two years, the focus has been on getting employees comfortable with AI through basic training in prompting and usage—a foundational step that has succeeded in reducing friction. But the newest generation of AI systems no longer fits that model. Autonomous agents that pursue goals, coordinate tasks, and trigger actions autonomously require a fundamentally different relationship: employees must become managers of AI rather than merely users of it.

The research cited in the article reveals the scale of the gap. Seventy-six percent of executives already view agentic AI as a coworker, yet most organizations have not yet taught employees how to set boundaries, calibrate trust, or maintain human accountability for agent-driven work. This creates a window of risk. The danger is not that agents will fail obviously—it is that they will be competent and confident enough that workers relax their oversight at precisely the moment judgment is most critical. The Boston University research showing 18% fewer errors when agents are labeled "AI employee" rather than "chatbot" suggests that framing shapes behavior, and that the language of teamwork carries real responsibilities that most companies have not yet operationalized.

The article's framework of five mental models—tool, intern, service provider, teammate, and expert—offers a practical way to think about this. Each model carries different expectations about autonomy, trust, and supervision. What makes the framework powerful is its flexibility: the same agent may serve different roles depending on the task and context. BNY's approach of treating digital employees with the same governance rigor as enterprise systems illustrates how this discipline can be embedded from the start. But the article makes clear that building this capability is not a training problem alone—it is an organizational one. Companies must make explicit what each model permits, when it applies, and how oversight changes. Without that clarity, the shift to agentic AI will create new sources of error and risk rather than the productivity gains it promises.