The AI-Era Manager’s Dilemma: When Users Become “Bad Emperors”

Summary

After spending half a month managing several AI agents, the author says the experience exposed his own poor management habits rather than simply revealing flaws in the tools. Vague requests such as asking for something to be “more advanced” led to repeated revisions because the desired result, acceptance criteria, scope and constraints had not been defined. The essay identifies seven recurring patterns: expecting an agent to infer unstated intentions, changing standards from turn to turn, withholding context and resources, adding requirements gradually, demanding results without measurable criteria, shifting responsibility for errors, and building elaborate management structures before doing the actual work. It argues that reliable agent workflows require a clear purpose, audience, deliverable, scope, prohibited actions, available materials and definition of success. Complex work should be decomposed into stages with checkpoints, while agents need the relevant files, context, tools, permissions and data. Status reporting and escalation rules should also specify what an agent may decide independently and when it must stop or ask for information. Because AI responds quickly and does not complain or resign, it removes many excuses that can be applied to human employees and sends repeated failures back to the person who designed the task. The author concludes that AI use is becoming a form of personal management: individuals may direct research, writing, coding and scheduling agents while setting priorities, checking results and accepting responsibility. The analogy has limits, since AI lacks employees’ emotions, rights and interests, but the symptoms of bad management can look strikingly similar.