What working with AI can teach us about the requests we make of other people.
When AI gives you something you didn’t want, take a moment to look at what you actually asked for.
Was the outcome clear? Did you explain the context? Did you provide the information it needed, or assume it would somehow know?
I’m direct about this because I think it deserves our attention. If you repeatedly struggle to communicate what you need to AI, take a look in the mirror and consider how you’re communicating with people. Some of the same habits may be showing up there, too.
That doesn’t mean every poor response is your fault. AI can make mistakes, miss information, and give you a confident answer that is wrong. You still have to review the work. But before you rewrite the request for the fourth time, it’s worth examining the first one.
I use a simple exercise in workshops. I tell people, “I’m hungry. Make me a sandwich.”
They get to work designing a sandwich. Meanwhile, I have a very particular idea of what I want. I know the bread, the condiments, and how it should be put together. I even have preferences about how it should be cut.
What I have actually said is, “Make me a sandwich.”
You can probably imagine the difference between what I pictured and what someone else might make. They could put thought and care into their sandwich and still deliver something I don’t want.
Now consider the stories we might tell ourselves afterward. I might decide they should have asked. They might decide I should have explained. I could interpret their result as a lack of care. They could experience my disappointment as a lack of appreciation.
We’ve both added meaning to a situation in which important information was never shared.
This is familiar outside a workshop. We ask someone to prepare an update, organize a meeting, or take care of a task. We know the audience, the sensitivities, and what we hope will happen next. The person receiving the request may know very little of that.
If you’ve ever said, “That’s not what I meant,” you’ve probably had a version of the sandwich conversation.
I think about AI as a delegate or virtual employee. It needs a clear assignment and enough information to do useful work. I need to decide what it can handle and where I have to remain involved.
Before handing over a task, I find these questions useful:
Take a request to prepare a meeting agenda. “Make me an agenda” leaves a lot open to interpretation.
You could instead explain that the meeting is thirty minutes, the participants have already read the project update, and the purpose is to decide which two activities to prioritize. Give it the activities under consideration. Ask for a draft that leaves room for discussion and ends with decisions and owners. Tell it to flag missing information rather than fill in the gaps.
You’ve made several decisions before asking for help. That preparation takes effort. It also gives you something specific to review when the draft comes back.
The same preparation would help a colleague. They would understand why the meeting exists and what their agenda needs to support.
Clarity is a shared responsibility. The person making a request has a responsibility to explain it. The person receiving it has a responsibility to ask when something is unclear.
With AI, you can explicitly invite questions before it starts. You can ask it to identify assumptions and tell you what information is missing. You then need to read those questions and answer them.
With people, consider whether you make questions easy to ask. When someone checks their understanding, do you welcome it? Or do you sound impatient because you believe you’ve already explained?
Someone may move ahead with an incomplete picture because they’re trying to help quickly. They may also be reluctant to ask a question they think you’ll find obvious. Your response helps determine what they do the next time.
“Say what you need, and ask before you build” is a principle I return to. Both sides have a part to play.
Choose a task you’re about to hand over, either to AI or to a colleague. Spend a few minutes writing down what you have in mind before you make the request.
Explain the result you need and what it will be used for. Provide the relevant information. Be specific about what can be decided independently and what needs your review.
For example:
I need a draft update for our team meeting. Use the notes below to explain what has changed, what needs a decision, and what happens next. The audience already knows the project background. Keep it to one page. Flag anything you cannot establish from the notes. I will check the facts and decide what to share.
If you’re asking a person, finish by inviting them to tell you what they understand the assignment to be. With AI, ask it to confirm its understanding and surface questions before beginning.
Afterward, consider what needed correcting. Was information missing? Was a term too vague? Did you and your delegate understand “finished” differently?
Use what you learn in the next request. You don’t have to write an elaborate brief for every small task. Start by sharing the information that would otherwise remain in your head.
As you begin to delegate more effectively, consider what you want to make room for. Spend It on Purpose explores that next decision.
Before your next handoff, ask yourself: What am I expecting someone else to know that I haven’t actually said?
Start with the Paper Cut. Choose a small recurring task on which to practice a clearer request.
Spend It on Purpose. Connect better delegation to the way you organize and use your time.
Christina Levine is a Partner at Scimitar and the founder of Rehumanize Work™. With nearly twenty years of experience in consulting, strategic business management, and leadership development, she helps people reshape their days, teams, and organizations so they can do more of what they love. Her work brings Purpose, Presence, and Process into everyday practice through coaching, workshops, and practical changes to how work gets done.
If you or your team could use a partner in figuring out where to begin, reach out through scimitar.com. I’d welcome a conversation about what’s getting in your way and what we could work on together.