AI Without the Hype.
Chapter 9 of 21
Part Three · Make It Yours · Chapter 09

How to actually
delegate

By the end of this chapter you can hand a real task to Claude the way you would brief a capable colleague: plan first, reach a shared understanding, then let it run.

You have a constitution, a CLAUDE.md, and a skill. You have built the world. Now the move that turns all of it into real output: handing over real work and trusting it to run, without losing the judgment that makes it yours. This is the thesis of the whole course, made operational.

Most people delegate badly to the AI the same way they delegate badly to people. They throw a half-formed task over the wall and hope. Then they are surprised when it comes back wrong. The fix is not to do it yourself. The fix is to brief properly before you let go.

AVATAR OPENER · ~90s
Watch: briefing a task to ninety-five percent before letting it run
HeyGen avatar · generated, consistent presenter

Here is the trade you are making, and it is worth saying plainly. You keep the judgment. It brings the speed. You decide what good looks like, what matters, what is out of bounds. It does the fast, tireless execution. Hand over the judgment and you get bot-quality work. Keep the judgment and supply it clearly, and you get a partner that moves faster than you ever could alone.

Throw it over the wall
A half-formed task, no shared understanding. You hope it guessed your intent. It usually did not.
Result: confident, fast, and pointed at the wrong thing.
VS
Brief, then let go
Plan together, reach shared understanding, agree the bar, then it runs while you watch the edges.
Result: speed you could not match, aimed where you wanted.
You keep the judgment. It brings the speed. Give it the judgment by briefing it well, then hand over the speed. That is the whole trade, and it only works in that order.
THE THREE MOVES

Good delegation has a shape. It is the same whether you are handing off to a person or a model: agree the plan, close the gaps, then let it run. The mistake is skipping straight to the running.

Plan first
Ask for the plan before any work. Read it. Redirect cheaply, on paper.
Grill to ninety-five
Have it ask you questions until you both understand the task.
Then let it run
Once you share the picture, hand over the execution.

The middle move is the one beginners skip and the one that changes everything. Instead of you trying to write the perfect brief, you flip it: you ask the model to interview you. Tell it to ask whatever it needs until it understands the task as well as you do. This is sometimes called grilling, reaching shared understanding before any work happens. It surfaces the things you forgot to mention, while changing them is still free.

SEE IT

Here is the prompt that does it. Notice you are not briefing harder. You are asking to be questioned. The gaps in your own thinking come out in the answers.

The grill-me move
I want you to help me with [the task]. Before you do anything, interview me. Ask me one question at a time about what I actually want, who it is for, what done looks like, and what would make it wrong. Keep asking until you are about ninety-five percent sure you understand it as well as I do. Only when you are confident, show me the plan. Do not start the work until I say go.
Try it in Claude

What happens next is the useful part. The questions it asks are the brief you would not have known to write. Some you answer instantly. Some make you stop and realise you had not decided yet. That pause, before any work is done, is the cheapest possible place to find out. Once you both see the same picture, you say go, and now letting it run is safe, because the judgment is already in the plan.

NOW YOU TRY · APPLY
Delegate one real task properly

Take one real task you have been putting off because briefing it feels like as much work as doing it. Use the grill-me prompt. Answer its questions honestly, including the ones that make you pause. Read the plan it proposes. Redirect on the plan if it is off. Only then say go. Notice how much of the quality was decided before any work happened.

Right if at least one of its questions surfaced something you had not yet decided, and you settled it before any work began.
Show the worked solution
The drill works when a question stops you. Say the task is "plan a small launch for my new service." You expect to brief it in a line. Instead it asks: who exactly is this launch for, what does a successful first week look like in numbers you would actually be happy with, what are you not willing to do (no paid ads? no cold DMs?), and what is the one thing that would make the whole thing feel wrong. Two of those you answer in a second. One or two make you realise you had not decided. That is the value: you found the gap on paper, before a single asset was made, when fixing it costs nothing. Then you see a plan, you nudge it, you say go. Compare that to throwing "plan my launch" over the wall and getting a confident, generic plan aimed at the wrong audience that you then have to unpick. Same model, same minutes. The difference is entirely in the briefing, which is to say, the judgment, which is the part that stayed yours.
WATCH FOR
✗You skip straight to "do it". Ask for the plan first. Redirecting on paper is free. Redirecting finished work is not.
✗You try to write the perfect brief yourself. Flip it. Ask the model to interview you. It finds the gaps you would have missed.
✗You say go before you actually share the picture. If a question made you pause and you have not resolved it, you are not at go yet.
✗You hand over the judgment, not just the work. You decide what good looks like. It executes. Never outsource the call on quality.
WHAT YOU LEARNED
The takeaways
  • Delegating well is briefing well. The fix for bad results is not doing it yourself, it is reaching shared understanding first.
  • The three moves: plan first, grill to ninety-five percent shared understanding, then let it run.
  • The grill-me move flips the work: instead of writing the perfect brief, you ask the model to interview you until it gets it.
  • You keep the judgment, it brings the speed. That only works if you supply the judgment before you let go.
Your project · step nine

Take the next real piece of your thread project and delegate it with the grill-me move. Let it interview you, agree the plan, then let it run. You now have the full Part 3 stack working together: your constitution and CLAUDE.md supply the standing judgment, and delegation supplies the speed. From here we sharpen the skill that protects all of it: reading what comes back with a clear eye.

The amateur throws the task over the wall and hopes. The director briefs to ninety-five, then gets out of the way. The whole craft is in those two extra minutes.

← Chapter 8: What a Skill is Next → Chapter 10: Read output like an editor