Princeton scientist says AI agents may turn knowledge work into supervision

In short: Some office jobs may shift from doing the work to supervising AI agents that do parts of it.

What's going on

Princeton computer scientist Arvind Narayanan said on The Ezra Klein Show that workplace AI might not just replace workers. It could also change what their days look like. Instead of writing, designing, researching, or analyzing, more people may spend time overseeing AI agents.

An AI agent is software that can take steps on its own to complete a task, like drafting a report and pulling information from documents. Narayanan’s point is that workers could end up acting more like supervisors. Think of it like moving from cooking a meal to watching a fast, messy helper in your kitchen and fixing mistakes before the food goes out.

Narayanan argued this shift could make work less satisfying. Managing people can be stressful, but it can also come with higher pay, more status, and the good feeling of mentoring others. Managing AI agents may not offer those upsides, especially if workers are still held responsible when the AI gets something wrong and they spend less time practicing the skills they trained for.

He also said the outcome depends on how companies deploy these tools. If AI handles routine chores and leaves workers in real control, it could help. If it turns skilled workers into constant monitors and error-catchers, it could feel like a downgrade.

What to watch

Narayanan was describing a plausible direction, not proving it is already happening everywhere. A key question is how employers measure performance and responsibility when AI is involved. Another is whether workplaces give people time to keep their own skills sharp, instead of treating them mainly as quality checkers.

Source: NYTimes