Stop asking. Start delegating.
The difference isn't which AI you use. It's whether you ask for a small answer or hand over a well-defined piece of work.
Asking or delegating
Asking
You supply every detail and work out the next steps yourself.
Delegating
AI prepares the work. You approve what happens next.
Why this works
It can't read your mind, but it can read your material
Most bad AI answers aren't the model being dumb. They're the model guessing at everything you didn't say. Every fact you hand over is one less guess.
A small sample is cheap, a wrong batch is expensive
If the first draft is off, you've lost two minutes. If a hundred are off, you've lost the afternoon and maybe trust in the tool. Checking early is the cheapest quality control there is.
Drafts keep you in charge without doing the work
Reviewing a good draft takes a fraction of the time it takes to write it. You still decide what goes out, so a mistake costs you an edit, not a client.
A worked example, start to finish
A small business owner got a long email from a client asking three questions and complaining about a missed deadline. Here's how the conversation goes when you delegate instead of ask.
Write a reply to this client email.
Dear client, thank you for reaching out. We apologize for any inconvenience and value your business. We will look into this and get back to you shortly.
That's useless, and it's my fault. Here's what you need. The client is a 10-year customer. The deadline slipped because their files arrived late, but we never told them that. Answer all three of their questions using the notes I'm pasting below. Own the communication miss, not the delay. Keep it under 150 words, in my voice: direct and friendly. Don't send it. Show me the draft.
Hi Dana, you're right that we went quiet, and that's on me. Here's where things stand. 1) The new date is Thursday. 2) Yes, the price stays the same. 3) We'll send updates every Monday until it's done. The delay came from files arriving on the 14th, and I should have told you then instead of now. I'll call you tomorrow to walk through it. Sam
Good. Change Monday to Tuesday, that's when we meet. And add that the files are now complete so there's no further risk.
This is an illustrative example. Names and details are made up.
Seven habits
1. Talk, don't write specifications
Explain what you want in your own words. Turning it into a plan is the AI's job.
2. Give it the whole problem
The people, the history, the constraints, and what good looks like.
3. It drafts, you decide
Anything that leaves your hands stays under your control.
4. Turn repeat mistakes into rules
Correct the same thing twice, then make it standing.
5. Push back
Ask what it checked and what a skeptic would say.
6. Give it real material
A recording or a thread beats a description. Share only what you're allowed to.
7. Check the first sample
Answer its questions and look at ten before it does a hundred.
Common mistakes
Asking for the output without the inputs
Fix: Paste the thread, the notes or the file. Then say what you want.
Accepting the first answer
Fix: Treat it as a first draft. Say what's wrong and why. The second answer is usually much better.
Letting it act before you've seen a sample
Fix: Ask for one or ten before it does a hundred, and before anything is sent.
Never writing down what you keep correcting
Fix: When you fix the same thing twice, save it as a standing instruction in your AI's settings or project.
Check yourself
Pick the better request. You'll see why right away.
1. You need a follow-up after a sales call.
2. You want a list of 200 local businesses to contact.
Exercise
Rewrite this request so the AI can actually do it well: "Make me a social media post about our new service." Write yours first, then compare.
Show a strong answer
Notice what it adds: what the service is, who it's for, why they care, where it's going, how long, the tone, the one fact that must be in it, and a first step that's small enough to check.
What Jason actually did
I have AI watching two inboxes with different jobs. One gets the alerts and the noise: it labels, files and only brings me decisions. The other is client-facing: it drafts replies to customers and saves them as drafts. Nothing sends until I read it and click Send. Most of the drafts are right the first time.
Practice: turn a request into an assignment
Pick a kind of task, describe the situation, and get an assignment to paste into any AI. The coach checks it as you go. Nothing you type leaves this page.
Try it today
Private to you. Saved only in this browser, on this device. Nothing here is sent anywhere, and nobody else can see it. Clearing your browser data erases it.
Your saved prompts
Nothing saved yet. Use "Save this prompt" on any example above.
Go deeper
The sources worth your time if you want the reasoning behind this lesson.
- Be clear and direct (Anthropic): a short guide on giving AI the context a new colleague would need.
- Prompt engineering guide (OpenAI): the same ideas from the ChatGPT side, with examples.
- Prompting strategies (Google Gemini): Google's version, useful if Gemini is your tool.