Delegating to AI agents is the new managing up

You ask an AI agent to make your deck better. It comes back with brighter colors and shorter bullets, when what you wanted was a different story for a skeptical CFO. You never mentioned the CFO.

Most disappointing AI output starts with a vague handoff.

Why delegating to AI is hard for most people

For a while the advice was to learn prompting and find the right magic words. Then models got better at reading sloppy instructions, and how well you hand off the work started to matter more than how you phrase it.

Managers learn delegation, usually by doing it badly for a couple of years. Individual contributors spend their careers on the other side, taking a task, filling in the missing context from experience and delivering.

An AI agent leaves everyone managing capacity they didn't have before, and people make the same mistakes first-time managers make.

The vague handoff. "Make this better." Better how, and for whom? The agent guesses, like a new hire would, and you blame it for guessing wrong.

The context you kept to yourself. You know the client hates jargon, the deadline moved and the last version died in legal review. You didn't mention any of it, so the output ignores all of it.

No definition of done. You'll know it when you see it. Twelve rounds later you still haven't seen it, because you never wrote down what you wanted.

Skipping the review. The output looks polished, so it goes out unread. Polish makes unreviewed work risky, because the errors look finished too.

The good news is that these are delegation mistakes, and delegation can be learned.

What to include when you delegate to an AI agent

A good handoff has the same parts whether it goes to a junior teammate or an AI agent. Start with the objective stated as an outcome, like "get the client to agree to a call without offering a discount," rather than "draft an email." Add the context that lives only in your head, the boundaries on what must not happen, a definition of done you could check the result against, and a checkpoint where the work comes back to you.

That's close to the briefing you'd give a capable new hire who has read everything and knows nothing about your situation. People who already delegate well, like a seasoned chief of staff or executive assistant, tend to get good results from an agent quickly.

Say you need a follow-up to a client who went quiet.

The usual way: You ask for "a follow-up to the client." You get a polite, generic nudge that offers a discount you'd never approve, and you end up rewriting it yourself.

With a brief: You say the goal is a 20-minute call, the client went quiet after legal flagged the contract terms, discounts are off the table, and you want a short draft to review before anything is sent. The first draft needs one edit.

Why delegating to AI is like managing up

Managing up means getting what you need from someone whose time and attention you don't control. It builds the same habits a good brief needs: packing context into a few lines, being explicit about the ask and guessing where you'll be misunderstood.

Both also make you understand your own work well enough to explain it. Someone who can't brief an agent usually can't brief a person either, and the agent shows it faster.

How to get better at delegating to AI agents

A better prompt gets you one better answer. A better delegation habit improves every handoff after it, and the pieces add up: context docs you reuse, definitions of done that become team standards.

Write the brief before every handoff that matters, covering the objective, context, boundaries, done and checkpoint. It takes two minutes.

When the output disappoints, reread what you asked for before blaming the agent. The missing piece is often still in your head.

Keep a one-line log per task with what you handed off, how it went and what you'd brief differently. Your blind spots show up within a few weeks.

Briefs get better when the whole team can see them. In Hops, the Collaborative AI™ workspace, you brief the AI agent in a shared thread and it works in your team's docs and tables, so teammates can see the request, add the context you forgot and pick up the result. What one person teaches Hops, like how your clients want follow-ups written, the whole team can use. Anything that goes outside the company, like that client email, waits for a person to approve it.

Delegating used to depend on your spot on the org chart. Now it depends on how well you write a brief.