Most leaders have only ever managed people. Kate Kellogg, Professor of Work and Organization Studies at MIT Sloan, says that is changing. The next job is supervising AI agents, software that carries out tasks on its own, and nobody has learned it yet.
In a clip from the school’s Invent the Future event, she prefaces the claim by saying she does not want to get too futuristic, then makes it anyway: leaders “are going to become managers of agents, not of people.” So they need to understand how to do that. The problem is that they do not, and she has a name for it.
She calls it agent supervision ambiguity. With agentic AI, a manager is acting as a supervisor of agents, and people do not know how to do that yet. A career spent supervising people comes with no instructions for supervising agents, and the ambiguity is that nobody has written those instructions.
That leaves the question of what a leader can do, and she gives two answers. The first is to encourage people to have a growth mindset. She adds that she loves being at Sloan because everybody there has one.
The second is about volume. People need to be running a lot of agents, she says, and they need to think about how those agents work in parallel. Running many agents at once is a different way of working, and Kellogg describes it as a new skill set that people have to be trained to do.
Kellogg’s only hedge is that this sounds futuristic. On October 6, MIT Sloan listed a new course she leads, Leading Agentic AI Transformation in Organizations, built around five phases: architecting for agentic AI, co-creating agentic workflows, deploying agentic systems, monitoring them, and scaling them. The course description says the practices that took generative AI from experiment to scale are insufficient for agentic systems that plan, make decisions, use tools, and run multistep workflows on their own.
