AI MADE IT POSSIBLE · YOU CAN’T PROMPT WISDOM · PART 1 OF 3

You Can’t Prompt Wisdom

Why asking AI to “act like a world-class expert” does not give you their experience.

AI can imitate the language and frameworks of expertise instantly. That is an extraordinary capability. It is not the same thing as transferring judgment to the person reading the answer.

AI can help you do things you could not do before.

It can help you write, research, design, plan, analyze, organize, build, learn and create. It can put capabilities in the hands of one person that previously required several people, specialized software, years of practice or a budget that kept many ideas from ever getting started.

I believe in that opportunity. It is one of the central reasons I built the AI Made It Possible work in the first place.

But there is a mistake that becomes easier to make as the tools get better.

It starts with a prompt.

Act as a world-class marketing strategist with 25 years of experience. You work exclusively for me. Review my business, identify the best strategy and give me the exact plan I should follow.

Change “marketing strategist” to graphic designer, publisher, product manager, music producer, business consultant, researcher or almost any other professional role.

The AI responds. The answer may be structured, confident, detailed, professional and genuinely useful.

That is exactly where the trap can begin.

You Asked for the Language of Expertise

A powerful AI system can identify patterns associated with professional expertise. It can explain frameworks, compare approaches, reproduce terminology, analyze information you give it and expose you to considerations you might never have thought about yourself.

Those capabilities are real.

But experience is not simply a collection of sentences an expert knows how to say.

Experience includes consequences.

The experienced professional has seen plans fail. They have watched customers react differently than expected. They have discovered that the thing everyone agreed would work did not work. They have dealt with the exception nobody remembered to put into the procedure. They have learned when an industry rule applies, when it does not, and when a situation deserves another question before a decision is made.

They have made mistakes. They have corrected mistakes. They have lived through changing markets, changing technology, changing customers and changing expectations.

Eventually, some of that becomes judgment.

That is much harder to reproduce with one impressive prompt.

Information Is Not the Same Thing as Judgment

Imagine two people sitting beside you.

One has access to an enormous amount of information about your industry.

The other has spent twenty years actually operating inside it.

Those people may know many of the same facts. They still may not make the same decision.

The second person may notice something that looks insignificant to everyone else. They may say:

• That customer behaviour worries me.
• That metric is giving us the wrong signal.
• That visual may be beautiful, but it does not fit this audience.
• That recommendation normally works, but your situation is not normal.
• That launch should wait.
• Or everybody else is waiting, and this may actually be the moment to move.

That kind of judgment often comes from an accumulation of situations.

AI can help reconstruct some of those considerations. But asking it to “be world-class” does not suddenly give you the ability to know whether its recommendation fits your particular situation.

And if you do not know enough about the subject to recognize the difference, the confidence and polish of the answer can become part of the problem.

There Is a Name for Part of This Problem

Researchers have studied automation bias for years: the tendency to over-rely on automated recommendations instead of independently checking them.

This problem did not begin with generative AI. But generative AI creates a particularly persuasive version of it because the system does not merely provide a number or a warning. It can explain itself. It can organize an argument. It can sound thoughtful. It can produce a response that feels like a consultation.

The Problem Is Not the Prompt

I am not telling you to stop using expert-role prompts.

I use them.

Giving AI a role can establish the kind of analysis you want. Telling it to consider a problem from the perspective of a senior brand strategist, editor, product manager or researcher can expose you to questions and frameworks you might otherwise miss.

The mistake is not: “Act as an expert.”

The mistake is: “Therefore, whatever comes next must be expert judgment.”

Those are very different things.

One is a useful way to explore a problem.

The other transfers authority to a machine before you have established whether that authority is justified in this situation.

I Had to Learn This Around Emerging Technology

Long before the current generative-AI wave, I worked around emerging technology in business environments.

One thing you learn quickly is that you cannot build a serious operation around what the technology demonstration says is possible.

You test it. You document it. You compare it. You find the edge cases. You find the thing that breaks. You determine what happens when the expected input does not arrive. You decide when a human needs to intervene.

And then, once you think the system works, you keep watching it.

Because the technology changes. The environment changes. The users change. The business changes.

That lesson matters even more with AI systems that can change rapidly through model updates, product changes, new tools and altered system behaviour.

A good AI workflow therefore cannot depend entirely on “the prompt that worked last month.” It needs checkpoints.

What I Mean When I Say I “Train” My AI

I often use the word train when describing how I work with AI.

Technically, in normal day-to-day use, I am usually not retraining the underlying model itself.

I am doing something much more practical.

I am building the context around it.

I give it my history, goals, previous decisions, brand standards, audience, examples of work I like, examples I rejected, performance information, mistakes, corrections, documents, research, rules and exceptions.

Compare these two situations.

Build me a world-class homepage.

versus a system that understands what your business sells, who it serves, which pages already exist, what customers are responding to, which design choices failed, what language does not fit your brand, what your visual standards are, what your conversion goals are and what you are trying to accomplish next.

The difference can be enormous.

But the AI did not magically become wise.

More Context Still Does Not Remove Your Responsibility

Once you have spent months feeding an AI useful information, it can become tempting to trust it even more.

Sometimes you should. A well-developed system should become more useful.

But familiarity is not proof.

The fact that an AI knows your brand does not mean every brand decision it recommends is correct.

The fact that it knows your publishing goals does not make it your lawyer.

The fact that it has analyzed a large body of your content does not mean the first campaign concept it generates will connect with your audience.

Context improves the decision environment.

It does not eliminate the need to make the decision.

Execution and Judgment Are Not the Same Thing

People talk constantly about AI replacing human work.

Some paid execution work is going to be compressed or eliminated. In some areas that is already happening.

There are tasks people previously had to pay another human to perform because they did not possess the software, technical skill, time or production ability to execute them themselves.

AI changes that equation.

We should be willing to say that plainly.

But there is a major difference between replacing execution and replacing judgment.

If AI can perform more execution for you, your responsibility increasingly moves upstream:

• What are we actually trying to accomplish?
• What information should the system have?
• What assumptions is this answer making?
• What might be wrong?
• What should we test?
• What requires outside verification?
• What result would cause us to change direction?
• Where does human review belong?
• How serious are the consequences if we get this wrong?

Those are not merely prompting questions.

They are control and governance questions.

The Amount of Checking Should Depend on the Consequence

This is where I think AI education often becomes unrealistic.

People are told either to verify absolutely everything or to automate almost everything.

Neither position is useful by itself.

If I ask AI for ten possible names for a fictional character, I am not commissioning an independent audit of every suggestion.

If I use AI to help with a major legal, financial, medical, rights, safety or business decision, the standard should obviously be different.

The amount of control should rise with the consequence of being wrong.

Before trusting an AI output, one of the most useful questions you can ask is:

What happens if this answer is wrong?

Maybe almost nothing happens. Move quickly.

Maybe you waste a little time or money. Check more carefully.

Maybe you publish false information under your own name. Slow down.

Maybe you change an important part of your business. Test it.

Maybe the decision involves someone’s rights, livelihood, safety or significant money. Now you need a stronger validation process and, where appropriate, qualified human expertise.

This is not anti-AI.

It is how you use AI without surrendering responsibility to it.

You Do Not Need to Become an Expert Before You Start

There is another extreme I want to avoid.

I am not saying you should refuse to use AI until you have twenty years of professional expertise.

That would destroy much of what makes this technology important.

The extraordinary opportunity is that people can now attempt things before they possess every skill traditionally required to begin.

The studio does not always have to be the price of entry. The agency does not always have to be the price of entry. The publishing team does not always have to be the price of entry. The consultant does not always have to be the first step.

You can begin.

But beginning without expertise creates a new responsibility.

You need a system for dealing with what you do not know.

That is the missing piece.

The solution to missing expertise is not pretending expertise appeared because you wrote a better prompt.

The solution is building a process that helps expose the gaps.

Instead of One Answer, Create a Decision

Suppose you need a new branding direction and you know almost nothing about professional design.

The simplest AI workflow is:

You are a world-class graphic designer. Design the best possible campaign for my company.

You receive something beautiful. You like it. You use it.

Maybe it works.

But consider a different approach.

I am not a professional designer, so do not assume I will recognize the strongest solution simply by looking at it. Give me three materially different approaches. For each one, explain the design principle, the audience it may serve, the assumptions you are making, what could make it fail and what an experienced designer would check before approving it.

Now something different is happening.

You are not simply generating design.

You are building a decision process around design.

Then reality gets a vote.

You publish. You measure. You observe. You get feedback. You revise. You run another campaign.

And something important starts happening.

The AI receives better information.

But so do you.

The System Should Train the Human Too

This may be the most important point in this entire series.

The goal should not be to create an AI that becomes increasingly useful while the human becomes increasingly dependent.

A good system should make both sides better.

You should gradually learn why one headline worked better than another, why customers misunderstood an offer, why one visual performed differently, why a workflow broke, why a recommendation failed, which metrics actually matter, what questions should have been asked earlier and which AI answers deserve more scrutiny.

That is how judgment starts to develop.

You do not download it.

You accumulate it.

AI can accelerate that process enormously, but only if you use the tool to help you learn rather than using it to avoid learning.

AI Made It Possible. It Did Not Make Judgment Optional.

This is the tension at the heart of how I think we should use these tools.

AI has made an extraordinary range of capabilities accessible. I do not want to surrender that opportunity or rebuild old barriers around tools that can help ordinary people create, learn and compete.

But access without responsibility creates another problem.

If we are going to use powerful systems, we need to learn when to trust, when to test, when to verify, when to compare, when to bring in someone who knows more than we do and when the stakes are low enough to simply try something and learn.

That process starts before government regulation.

Before corporate policy.

Before somebody else tells you what responsible AI use is supposed to look like.

Regulation starts with the controls you put around your own work.

Once you understand why your own system needs rules, checkpoints, evidence and accountability, you are in a better position to understand the larger arguments about what AI regulation should address.

But first we need to understand what is actually being replaced.

Because AI can absolutely replace parts of jobs.

That still does not mean it replaced everything the professional knew.

Where This Fits

This mini-series belongs inside the larger AI Made It Possible conversation.

If you want the earlier foundation, read Before You Create It: Why the ABCs Matter in the AI Era.

For the companion lesson on why polished output is not automatically finished work, read AI Is Not a Shortcut.

And for the larger philosophy behind keeping human direction at the centre, continue with AI Should Make You More Valuable. Not Easier to Replace.