The Most Important AI Skill Isn’t Prompting. It’s Knowing What Problem You’re Solving.

THE OWNER’S ALMANAC

September 11, 2026

Imagine an acupuncturist opens ChatGPT and types:

Build me a website for my acupuncture practice.

A capable AI can do a remarkably good job.

It can create a polished homepage, write the copy, choose the sections, suggest calls to action, produce an FAQ, optimize the page for search and even generate the code.

Within minutes, the practitioner could have something that would have taken an agency weeks to produce a few years ago.

There is only one problem.

Nobody established what the website was supposed to accomplish.

Does she want more acupuncture patients?

Does she actually want fewer acupuncture appointments and more herbal medicine patients?

Is her problem traffic, conversion, positioning or patient quality?

Does the market misunderstand what makes her different?

Is the website supposed to fill an associate’s schedule, support higher fees, attract more complicated cases or help her work fewer hours?

The AI may have created an excellent website.

It may also have solved the wrong problem beautifully.

The quality of the answer begins before the prompt.

We’ve spent several years talking about prompt engineering as though the main advantage in AI would belong to people who learn how to ask machines for things more effectively.

That matters.

But I think a more important skill is emerging.

Problem framing.

The person who understands the problem at the deepest level can give AI a dramatically better problem to solve.

Consider three requests.

Build me a website.

Build me a high-converting website for an acupuncture practice.

Before designing anything, determine what economic job this website needs to perform, which patients the practice most wants to attract, why those patients currently fail to understand the practitioner’s value, and whether redesigning the website is actually the highest-leverage intervention.

Same AI.

Three very different levels of thinking.

The third request doesn’t merely give AI better instructions.

It changes the problem.

AI is extraordinarily good at operating inside a frame.

This is one of its greatest strengths.

Give an AI a goal, some context and enough constraints, and it can explore that territory faster than almost any human team could.

But that strength creates a peculiar risk.

If the frame itself is weak, AI can become extraordinarily sophisticated inside a weak frame.

A business owner can ask how to get more leads when the real problem is that nobody wants the offer.

A company can ask AI to write better sales emails when the real problem is that they’re contacting people with no reason to buy.

A consultant can ask for 30 days of social content when publishing more content has almost nothing to do with the bottleneck in the business.

An entrepreneur can spend an afternoon perfecting a landing page for a product that should never have been created.

And AI will help with all of it.

Enthusiastically.

This may make expertise more valuable, not less.

One of the early assumptions about AI was that expertise would matter less because everyone would suddenly have access to expert-level information.

I think reality may be more interesting.

Expertise tells you what deserves attention.

An experienced clinician sees something in a patient’s history that a beginner overlooks.

An experienced investor notices the assumption buried inside a financial projection.

An experienced marketer recognizes that improving an ad won’t fix an offer nobody wants.

An experienced designer understands that making a website prettier can actually make the underlying positioning problem harder to see.

AI can dramatically accelerate the work after those observations are made.

But someone still has to recognize what matters.

The scarce resource may increasingly be judgment.

There is another level beyond better prompting.

You can ask AI to question you.

That sounds simple, but it changes the relationship.

Instead of saying:

Here’s what I’ve decided. Help me execute it.

try:

Here’s what I’m trying to accomplish. Before helping me execute my plan, tell me whether I’m solving the right problem. Identify the assumptions I’m making, what information could change the decision, and whether there is a higher-leverage approach I’m overlooking.

Now AI isn’t merely an execution engine.

You’ve given it permission to challenge the frame.

I suspect this will become an increasingly important distinction between people who get enormous leverage from AI and people who simply produce more things with it.

Owners should use AI one step earlier.

Before asking AI to write the email, build the page, create the campaign or generate the strategy, ask:

What problem am I actually trying to solve?

Then ask AI:

• What assumptions am I making?
• What would have to be true for this approach to work?
• What evidence would suggest I’m solving the wrong problem?
• What problem might sit underneath the one I’ve described?
• What would someone with deep expertise in this area notice that I haven’t mentioned?
• Is there a simpler intervention that would produce the outcome I want?

Then make the thing.

The forecast

AI will continue getting better at execution.

That makes deciding what deserves to be executed increasingly important.

When almost anyone can create a competent website, campaign, business plan, presentation, sales sequence or piece of software, the advantage moves upstream.

Toward the person who can see the problem more accurately.

The future probably belongs less to people who know the perfect words to type into an AI.

It belongs to people who know when the question they’re typing isn’t yet the question worth answering.