The more I use AI, the more I notice something.
The center of what counts as "smart" is moving from knowing the answer to being able to decide what to ask.
It used to be that people who knew a lot had the edge. Good searchers, people with great memorization, people who could produce the right answer fast, all got rewarded.
But now, if you ask an AI like ChatGPT, you get general knowledge and neatly organized answers in seconds. Of course, AI answers need to be double-checked, but simply "coming up with some kind of answer" has become really easy.
So what's left for humans?
My take is that it's the ability to design questions.
Agenda design simply means deciding "what are we deciding today?"
You mostly hear the word "agenda" in the context of meetings.
What's on the table today? What are we going to discuss? What has to be settled by the end for the meeting to count as a success? Setting that is what an agenda does.
And it isn't just about meetings.
It's the same with work, with changing jobs, with questions you ask an AI, and with life's worries.
For example, if you just ask "Should I change jobs?", you'll get a thin answer.
But try asking it like this and things change.
- If I stay at my current company, what would have to improve for staying to be worth it?
- If I get transferred internally, what kind of track record should I go after there?
- If I leave, which comes first: salary, remote work, how drained I get, or the type of job?
- Should I move right now, or spend a set period building up results first?
Once you've broken it down this far, the AI stops being a fortune teller and becomes a tool that helps you make a decision.
In other words, using AI well isn't about knowing magic prompts.
It's about putting your situation, goal, constraints, and decision criteria into words so the AI can actually work with them.
Researchers and practitioners take "question design" seriously, too
This isn't just a gut feeling.
OpenAI's prompt engineering guide explains that a prompt is about writing "effective instructions" to get the result you want out of the model. In other words, what the AI gives back depends heavily on the instructions you put in.
Google's guide on framing a machine learning problem also stresses defining the ideal outcome, the model's goal, the output you need, and the success metrics first. That's about building AI, but it applies to ordinary work just as it is.
Research on learning and problem solving also has the idea of "metacognition." Put simply, it's the ability to look at your own thinking from one level up. MIT's teaching materials describe metacognition as using knowledge about the task, about how you learn, and about yourself to plan, check your progress, and evaluate the results.
So strong people aren't just thinking. They're watching things like:
"What exactly am I trying to solve right now?" "Is this approach actually getting me there?" "Wait, is the question itself off?"
This ability is going to matter a lot in the AI era.
People who can design questions use AI as a workshop machine, not a search box
People who only get shallow answers out of AI are often using it as a stand-in for Google.
"What do you recommend?" "What should I do?" "What's the right answer?"
Sure, you'll get a passable answer that way.
But the real value shows up when you load the question with context.
Say you want to build a reading habit. Instead of asking,
"How do I build a reading habit?"
try asking,
"I'm worn out from work on weekdays and I end up on my phone at night. I tend to quit things after three days. What kind of setup and rules would help me keep reading for just 10 minutes?"
The answer suddenly becomes something you can use.
That's less a difference in AI performance than a difference in the question.
You put your situation into words, state your constraints, set the goal, and even specify the output format you want.
By that point, you're already about halfway through solving the problem.
Why I can do this is probably because of the times I got burned in vague environments
In my case, how did I end up able to design questions?
Probably not just because of my personality.
I think the big factor is that I got hurt in vague environments.
The purpose is unclear. Nobody knows who makes the call. You can't see what "done" looks like. Afterward someone says, "That's not what I meant." Responsibility is blurry. You're judged on vibes and attitude.
In an environment like that, if you dive straight into the work, you burn out.
So I naturally started checking first.
- What's the purpose?
- Who decides?
- By when is it needed?
- What's the minimum I need to deliver?
- What are the priorities?
- What doesn't need to be done?
- What counts as finished?
This wasn't a polished business skill. It was a survival strategy.
If I didn't define the issues before starting, I was the one who got worn down. So agenda design got sharpened as a defensive move.
People who can't let a "something's off" feeling go are good at making questions
Another big factor is that I can't leave a sense of unease alone.
I can't stop at "doesn't this seem off?" I want to break it down further:
- What exactly is off?
- Which assumption is out of line?
- Whose responsibilities are getting mixed up?
- What do we need to decide to move forward?
- Is this an emotional problem or a structural one?
That habit is tiring.
But it pairs really well with AI.
Because the more context and conditions you hand over, the stronger AI gets.
Instead of just "What do you think about this?", you can say:
"Given this premise." "Within these constraints." "If the goal is this." "Separate the issues." "With the bare minimum." "For an interview." "In Markdown format." "As instructions for Codex."
People who can pass things along like that can use AI not as an answering machine but as a workshop for their thinking.
The side effect of being able to make questions: "too many questions"
That said, being able to make questions has a side effect.
People who can't make questions freeze because they don't know what to think about.
People who make too many questions freeze because the questions pile up.
For example, while thinking about a work problem, it can balloon all at once into:
- Should I stay at my current job?
- What should I gain from an internal transfer?
- If I leave, what kind of job should I aim for?
- Is my current exhaustion temporary or structural?
- Do I put salary first or freedom first?
- How does this connect to building assets for the future?
- What do I even want to protect in my life?
That is a real ability.
But when you're tired and try to process all of it at once, your brain shuts down.
So what people who generate lots of questions need isn't the power to ask even more.
It's the power to drop "the questions you don't need to ask right now."
What you need next is to sort questions into four types
If you generate lots of questions, it's better not to treat them all as the same weight.
For me, splitting them into these four makes things much easier to sort out.
1. Questions to decide now
Questions tied to what you do today.
Examples: What do I turn in by tomorrow? Do I apply this week? Do I sleep or work tonight?
Handle these.
2. Questions to think about later
Important, but you don't need a conclusion right now.
Examples: Your career a few months out, your views on marriage, the big picture of building assets.
Park these in a note.
3. Questions that aren't worth thinking about
Someone's true feelings, a fork in the past, the perfect answer, and so on.
You might learn a little from thinking about them, but they rarely lead to action now.
Don't spend long on these.
4. Questions better handled by sleeping
The heavy questions that show up at night.
Questions that are amplified by fatigue, hunger, loneliness, anger, or anxiety often shrink a lot by the next morning.
Don't analyze these. Send them to sleep.
The strength that counts in the AI era is "framing the problem," not "producing the answer"
If humans try to compete with AI head-on in the AI era, it's rough.
On knowledge, speed, and coverage, it's hard to beat AI.
But if you're on the side that uses AI, the picture changes.
What you need isn't to give orders in polished language.
It's to sort out your situation, set the goal, state the constraints, build decision criteria, and specify the output format.
In other words, the ability to frame the problem well.
Not "the person who produces answers" but "the person who can decide what to ask."
Not "the person with the model answer" but "the person who can put into words what to decide right here, right now, to move forward."
This ability becomes more valuable the smarter AI gets.
If the question is poor, the answer will be poor.
If the question is specific, AI becomes a very strong partner.
So what we should train from here isn't just prompt tricks.
It's the ability to look at your own reality, form a question, and, when needed, throw the question away.
People who can build an agenda are pretty strong, even in the AI era.
But if you're someone who builds too many, ask yourself this at the end:
"So which question do I actually need to decide right now?"
If you can come back to that one question, your thinking will keep moving forward.
Internal link ideas
- What wears you out at work is less the "workload" and more "unclear purpose" and "changes sprung on you later"
- When you can't decide whether to change jobs, first define the conditions under which you'd stay
- People who don't get results from AI are stuck on sorting out their premises, not on prompts
- When rumination won't stop, separate the questions to think through from the ones to sleep on
- What you need at a life fork around age 30 is building decision criteria, not hunting for the right answer
Related article ideas
- "For people tired of too many questions: just sorting your thoughts into four types makes things much easier"
- "Work skills for the AI era: five things to decide before you ask ChatGPT"
- "A requirements memo so vague instructions don't wear you down"
- "What is metacognition? The ability to see your own thinking from one level up"
- "Designing the agenda when you're torn between changing jobs, transferring, or staying put"
References and source notes
- OpenAI API Docs, “Prompt engineering”
https://developers.openai.com/api/docs/guides/prompt-engineering - OpenAI Cookbook, “GPT-5 prompting guide”
https://developers.openai.com/cookbook/examples/gpt-5/gpt-5_prompting_guide - Google for Developers, “Framing an ML problem”
https://developers.google.com/machine-learning/problem-framing/ml-framing - MIT Teaching + Learning Lab, “Metacognition”
https://tll.mit.edu/teaching-resources/how-people-learn/metacognition/ - Harvard Business Review, “AI Prompt Engineering Isn’t the Future”
https://hbr.org/2023/06/ai-prompt-engineering-isnt-the-future - García et al. (2014), “The effects of question-generation training on metacognitive knowledge, self regulation and learning approaches in Science”
https://pubmed.ncbi.nlm.nih.gov/25069559/
Notes before publishing
- No workplace-specific names, people's names, or concrete internal circumstances are included, so it's easy to publish.
- That said, the part about "getting burned in vague environments" is fairly personal, so if you publish while job hunting, you could generalize it further.
- SEO entry points: "smart in the AI era," "ability to ask good questions," "how people who use ChatGPT well are different," "agenda design."
- Draft date: 2026-07-01
