AI Is Delegation

AI conversations become clearer when you stop thinking about prompts and start thinking about delegation.

In the ED, I match the level of delegation to the individual. I don’t hand a medical student the work I’d give a senior resident. I don’t ask a consultant to look up a single lab value. The mismatch would cost more than the task is worth.

AI works the same way.

You need to know what you want. You need to describe the goal, not just the steps. You need to match the ask to the capability of the model.

There are task-level asks: summarize this transcript. Extract the action items. Compare these two documents.

There are multi-step asks: review these meeting notes, then find the unresolved decisions, draft a follow-up agenda, and flag where you are uncertain.

And there are agentic asks: multiple capabilities working together, using your data, checking progress, returning work that still needs your judgment.

Many people are stuck at the first level, wondering why the results feel thin. The issue often isn’t the tool, it’s the level of delegation.

The disciplines that make you a good leader also make you good at using AI tools: know the outcome you want, match the task to the capability, build in checkpoints, stay in the loop on decisions that matter.

Good delegation includes stewardship. Match the investment to the work: sometimes that means building something new, and sometimes it means getting the right data, clearer access, or a familiar interface into tools people already use.

Trust matters too. If a model can’t handle task-level work reliably, don’t move it into multi-step or agentic work. First decide whether the issue is your framing, the model’s capability, or the fit between the two.

Use the same approach you would use when delegating to humans.

If you are getting unexpected results, it might be the model. It might also be your approach.