Outcome
You stop typing one-line questions at AI and start briefing it like a capable new teammate — which is the single biggest jump in output quality you'll make in this whole course.
Concept
Most people's first AI habit looks like this: type a short question, get a short (often generic) answer, maybe try again. That's treating AI like a search engine. It works, but it leaves almost all of the value on the table.
The shift that changes everything is thinking of AI less like a search box and more like a capable new hire on their first week: smart, fast, and willing — but only as good as the brief you give them. A new hire who gets "write me a proposal" will guess wildly at what you want. A new hire who gets the goal, the context, and a way to check their work will come back with something genuinely useful.
We call this the delegation loop, and it has four parts:
- Describe the outcome. Not "help me with this email" — say what a good result actually looks like: "a reply that says no to this request but keeps the relationship warm."
- Give context. Anything a new hire would need and wouldn't know on their own: who you're talking to, what's already been said, any constraints.
- Set the guardrails. Tone, length, things to avoid, things to definitely include.
- Review the work. Read it like you'd review a colleague's draft — not as gospel, but as a strong starting point you can adjust.
This isn't a trick or a special phrase to memorize. It's just being as clear with AI as you'd be with a person you actually wanted to succeed.
A worked example: the meeting follow-up
Say you just finished a call and need to send a follow-up.
Bare prompt:
Write a follow-up email after my meeting.
Illustrative output: "Thank you for taking the time to meet today. It was great to discuss next steps. Please let me know if you have any questions." Polite, correct, and useless. It could have been sent after any meeting, to anyone, about anything.
Delegated prompt:
Outcome: a follow-up email that confirms what we agreed and makes the next step
obvious.
Context: I met with a client, [CLIENT NAME], about [PROJECT]. We agreed [DECISION
1] and [DECISION 2]. I owe them [MY ACTION] by [DATE]. They owe me [THEIR
ACTION].
Guardrails: friendly and direct, under 120 words, no filler like "I hope this
finds you well." End with one clear question.
Then list anything you assumed so I can check it.
Illustrative output: a short email that restates the two decisions, names who owes what by when, and ends with a single question about the date. Below it, a two-line note: "Assumed: the deadline applies to the full draft, not an outline."
That last line is the fourth step of the loop, built into the prompt. You asked the assistant to hand you its assumptions, so reviewing takes seconds.
Copy-paste delegation prompts
A general-purpose briefing template:
Outcome: [WHAT A GOOD RESULT LOOKS LIKE]
Context: [WHO IT'S FOR, WHAT'S ALREADY HAPPENED, ANYTHING THEY CAN'T GUESS]
Guardrails: [TONE, LENGTH, MUST-INCLUDE, MUST-AVOID]
Before you start, ask me up to three questions if anything is unclear.
When you finish, list your assumptions.
A prompt that turns a messy request into a brief for you:
I want help with this: [ROUGH REQUEST]. Interview me with the five questions you
would need answered to do this well, one at a time. Then write the result.
Build
Grab three prompts you've typed recently (or three you'd normally type) and rewrite each one as a real delegation:
Before: "Write a LinkedIn post about our product launch."
After: "Write a LinkedIn post announcing our product launch this Thursday. Audience is small business owners who don't consider themselves technical. Tone: confident but not salesy, one clear call to action to visit the site. Keep it under 150 words. Here's what we're launching: [details]."
Do this for your three prompts, run both the before and after versions, and compare. Most people are surprised by how much of the gap was just the briefing, not the tool.
Try it now (5 minutes)
Take one thing on your to-do list today that involves writing: an email, a message, a short description. Use the briefing template above and fill in all four lines honestly, even if one feels obvious.
Done when: you have a result you would genuinely send or use (after a light edit), and you can point to the one line of context that made the biggest difference.
Run
Keep a simple delegation checklist somewhere you'll actually see it — a sticky note, a pinned doc, whatever works. Four questions, every time:
- What does "good" look like here?
- What does the AI need to know that it can't guess?
- What are the guardrails (tone, length, must-avoid)?
- How will I check the result before I use it?
You don't need to write all four out every time — after a while it becomes automatic. But in the beginning, actually asking yourself these four questions before you hit enter is the difference between "meh" and "wow" results.
Common mistakes
- Delegating the thinking you haven't done. If you don't know what you want, say so and ask the assistant to interview you first.
- Briefing with a label instead of a goal. "Make it professional" is a label. "A client who is annoyed about a delay should feel heard and know the new date" is a goal.
- Skipping the review because the output looks polished. Polish is not accuracy (see Lesson 1).
- Over-stuffing. More context helps only when it's relevant. Cut the background a new hire wouldn't need.
Checkpoint
- I can name the four parts of the delegation loop from memory.
- I rewrote at least one real prompt as a delegation and compared the results.
- I've asked an assistant to list its assumptions or ask me questions first.
- I have a way to check results before I use them.
Next up: the specific techniques that make prompts even more reliable.