Most People Are Using It Wrong, Then Blaming the Tool

There's a conversation I have constantly now, and it goes the same way every time.

Someone tells me they tried AI. They asked it something about their business, got an answer that was generic and a bit obvious, and concluded the whole thing is overhyped. Then they say something like it's fine for writing emails, I suppose.

I understand the conclusion. I also think it's almost always wrong, and wrong for a specific and fixable reason: the answer was generic because the question was.

What "using it wrong" actually looks like

Not one big mistake. Five small ones, and most people make all five in the first ten minutes.

Treating it like a search engine. One question, one answer, close the tab. That's the single most common pattern and it wastes almost all of the value. The useful version is a back-and-forth — you push, it adjusts, you push again, and around the fourth exchange it finally says something you hadn't thought of. Most people quit at exchange one.

Giving it no context. People ask a question about their business without telling it a single thing about their business. No revenue, no customers, no constraints, no history, no idea what's already been tried. Then they're disappointed the answer was generic. It answered the question it was actually asked.

Accepting the first answer. The first answer is a draft. It is almost never the best thing you'll get out of the conversation. "That's too generic, try again knowing that my margins are thin and I can't hire" produces a categorically different response, and it takes nine seconds to type.

Asking it to do the thing it's worst at. Recalling a specific fact it has no way to verify is where it fails hardest and most confidently. Structuring a problem, drafting something you'll edit, arguing both sides of a decision, spotting what you left out — that's where it's genuinely excellent. Most people do the first and skip the second.

Never asking it to show its work. "Walk me through how you got there" catches an enormous share of the errors, and it's the cheapest quality check available.

None of that requires technical skill. It's a way of working, and it takes about a week to learn.

The scary part, said honestly

I'm not going to pretend the caution is misplaced. It isn't.

The thing will be confidently, fluently wrong. Not hedged, not obviously shaky — wrong in a tone of complete certainty, wrapped in reasoning that sounds right. That's genuinely dangerous for anyone who can't independently judge the answer, which is most people on most topics.

I've written elsewhere about what that looks like when you're building real software — the demo that works and the flaw that surfaces in week three. But the everyday version matters more to most people, and it comes down to two rules:

Never use it where you can't check the output. If you couldn't tell a good answer from a bad one, that's the one place not to trust it.

Always make it show the reasoning. Not because the reasoning is proof, but because bad reasoning is much easier to spot than a bad conclusion.

Do those two things and the risk drops enormously. Skip them and you'll eventually get burned, and you'll blame the tool for that too.

Why I think it's underrated, not overrated

Here's the part I find genuinely interesting.

For most tools, the gap between casual use and skilled use is modest. A good spreadsheet user is maybe three times as effective as a poor one. Worth learning, not transformative.

This is not that. The gap between someone poking at it casually and someone who has learned how it actually behaves is the widest I've encountered on any tool in my working life. Not three times. More like a different category of activity.

Which produces a strange situation. The hype is measuring the top of that range. Most people's experience is the bottom of it. Both sides then argue past each other, and the person in the middle — the business owner trying to decide whether any of this matters — hears noise and does nothing.

The honest position is that both things are true at once. The disappointment is real and mostly self-inflicted. The capability is real and mostly untouched.

What using it properly looks like if you're never going to build anything

Most people I work with have no intention of building software, and shouldn't. Here's what good use looks like for them.

Give it your actual situation, at length. Paste in the real numbers. Explain what you've already tried and why it didn't work. Tell it the constraint you're embarrassed about — the thin margin, the difficult partner, the thing you can't afford. Context is the entire difference between a generic answer and a useful one, and people withhold it out of habit rather than caution.

Use it to argue with you. "Here's what I'm planning to do. Make the strongest case against it." That single prompt has saved more of my clients' money than any tool recommendation I've ever made.

Use it to find what you left out. It's very good at noticing the question you didn't ask. That's a genuinely hard thing for a human to do about their own thinking.

Then check it. Every time. Especially when it agrees with you.

What the other end of that gap looks like

I said the distance between casual use and skilled use is the widest I've seen on any tool. Fair to ask what that means in practice, so here's my end of it.

The engine is the same one it has always been: a career of business operations and financial analysis. That's what decides which questions are worth asking, which answers are obviously wrong, and which of the fifty things a business could fix are the four that matter this quarter. None of that came from a tool.

What changed is the distance between having that judgment and producing the work.

Understanding a business properly means the industry trends, the competitive picture, the niche players nobody's watching, what the large players are doing and why, where the whole thing looks to be heading — and then all of it organized into things somebody can actually start on Monday. That used to be a team of ten and several weeks. I've sat on both sides of engagements exactly like that. It is now one person and a fraction of the time.

I want to be careful about what that does and doesn't mean, because it's easy to hear it wrong.

It does not mean the analysis got easy. The compression is in the labor, not in the judgment. Everything that made the work valuable — knowing what to throw away, recognizing the number that doesn't smell right, seeing which recommendation is going to die in a budget meeting and why — is doing exactly what it always did, and it took a career to build. Point the same tool at a business with nobody experienced steering it and you get a confident, beautifully organized document that's wrong in ways nobody in the room can see. I'd argue that's more dangerous than no analysis at all.

And it doesn't stop at the analysis. If the plan needs a piece of technology, I can find the one that already exists or build the one that doesn't — which is the difference between a recommendation and a change.

So let me state the position plainly, because it's the honest version. The background is the rare part. The tool is available to everybody, including you. What comes out the other end depends almost entirely on which one is doing the steering.

Where I land

It's the most valuable tool that has arrived in my working lifetime, and most people's disappointment with it is entirely justified by how they've used it. Those two statements don't conflict. They're the same statement viewed from two ends.

If you've tried it, been underwhelmed, and quietly written it off — you probably didn't test the thing you think you tested. It's worth one more honest attempt, done properly.

And if it turns out the answer for your business is that there's no worthwhile use for it, that's a legitimate finding too. I'd rather tell you that than sell you a subscription you'll open twice.