The AI Quality Tax

When good work becomes suspicious simply because the tools got better.

We’re in a strange moment. If something is polished, thoughtful, well written, or maybe just a little too good, there’s an increasingly common reaction: “AI wrote this, didn’t it?”

Sometimes it’s a joke. Sometimes it’s genuine curiosity. And sometimes there’s an implication underneath it that if AI was involved, the work must somehow be worth less.

I’ve run into enough versions of this lately that I’ve started thinking of it as the AI Quality Tax. The better the work, the greater the chance you’ll have to explain how much of it was “really you.”

Which is kind of funny when you think about it. We’ve spent centuries inventing better tools so capable people can do better work. Better cameras. Better software. Better instruments. Better machines. Better access to information. Now we’ve created one of the most capable tools in human history, and apparently using it effectively can become evidence against you.

I think we’re looking at the relationship backward.

The Tool Was Never Supposed to Become the Author

I’ve written about AI before, particularly in “AI Can Play the Notes. But It Still Can’t Hear the Music”, “The Standing Ovation Problem”, and “The Narrow Road.” The thread running through those pieces is pretty simple: technology can dramatically expand what we’re capable of doing, but capability and judgment are not the same thing.

I use AI every day. I think it’s extraordinary. It has changed how I research, explore ideas, organize information, challenge assumptions, refine my writing, analyze problems, and generally get from point A to point B without taking quite as many scenic detours along the way.

But I don’t think of it as software that writes for me.

I think of AI more like an incredibly advanced combination of a research assistant, encyclopedia, spell-checker, editor, sounding board, and occasionally a very bright, very green intern who has somehow read almost everything. It knows an astonishing amount, but it doesn’t have experience, and more importantly, it doesn’t always know what it doesn’t know. Push back enough, and you’ll often get some version of, “Oh yeah, you’re right.” Which is useful... but also a pretty good reminder of who should be supervising whom.

That distinction matters. AI can contribute to the process, speed it up, and improve it. But if you own the input, AI should never own the output (aka, final product).

If my name is on something, I’m responsible for whether it’s accurate, whether the argument makes sense, whether the tone is right, whether the strategy holds up, and whether I actually believe what it says.

AI doesn’t have to sit across from a client and defend a recommendation when it’s wrong.

I do.

Good AI Use Requires Better Inputs and More Pushback

The question I hear most often is some variation of, “Did you use AI?” I’m increasingly convinced that’s the wrong question.

Give two people access to the same AI model, and you can still get dramatically different results. One person might type a vague prompt, take whatever comes back, copy it into a document, and move right along. Another might spend the next hour adding context, challenging the software, rejecting ideas, correcting facts, refining language, and pushing the work until it actually says something worth saying in a way that they would say it.

Both used AI. Each got very different results.

Good AI starts with good inputs. Context matters. Experience matters. Knowing which questions to ask matters. Knowing what information the machine doesn’t have matters. Most importantly, knowing what good looks like before you ask AI to help you get there matters.

Then comes the part nobody puts in the commercials: you have to push back. Constantly.

“That doesn’t sound like me. You missed the point.” “That’s too generic.” “Go deeper.” “Where did that assumption come from?” “Technically correct, strategically wrong.” “Nobody talks like that.” “Try again.”

Using AI feels like managing a very bright intern with unlimited confidence and absolutely no fear of being wrong. The value of AI isn’t merely in generating an answer. The value is knowing what to keep, what to change, what to challenge, and what to throw away entirely. That’s where IQ, EQ, experience, taste, and discernment enter the picture.

AI can give you more possibilities.

It can’t decide which ones deserve to survive.

The Quality Tax Exists for a Reason

To be fair, people didn’t become suspicious of AI-generated work out of nowhere. There’s a lot of garbage out there.

The Internet is chock full of perfectly grammatical content that somehow says next to nothing. You’ve seen it. Every challenge is an “opportunity,” every company seems to be saying some variation of the same thing, and half the emails in your inbox sound like Spock got a marketing degree. Technically fine. Completely forgettable.

Everything is transformative. Everything is being unlocked. Everything is being leveraged.

At some point, we may need to lock a few things back up.

AI has made polished mediocrity extremely cheap to produce, and audiences are learning to recognize it. That skepticism is healthy. We should absolutely become better at distinguishing substance from synthetic filler.

The problem is when we swing too far in the other direction and start treating quality itself as evidence that a human couldn’t possibly have been deeply involved.

A well-written article becomes suspicious because it’s too clean. A polished presentation must have been “made by AI.” A strong proposal gets discounted because surely the machine did most of the heavy lifting.

That’s the Quality Tax.

We’ve correctly learned that polish no longer guarantees substance, but we shouldn’t conclude that polish proves the absence of substance.

Efficiency Doesn’t Negate Value. Efficacy Proves It.

This becomes especially interesting in professional services.

Suppose an experienced strategist uses AI to accomplish in three hours what might have taken eight a few years ago. What exactly became less valuable?

The insight didn’t. The outcome didn’t. The experience required to recognize the right answer didn’t. The client certainly isn’t worse off because the work happened more efficiently.

The friction decreased.

That’s different.

We’ve always had a strange relationship with effort and value. If somebody struggles visibly for ten hours, it somehow feels easier to understand why the work costs money. If decades of experience and better tools let that same person solve the problem in three, we suddenly start doing the hourly-rate math.

But efficiency doesn’t negate value. Efficacy proves it.

The real question isn’t how many hours were consumed producing the work. It’s whether the work actually worked. Did it solve the problem? Clarify the decision? Improve the positioning? Change a mind? Produce the intended result? Move the business forward?

Expertise has always worked this way. A seasoned musician can make something difficult look effortless because you’re watching the result of thousands of hours you didn’t see. AI doesn’t erase the expertise that came before it. Used properly, it leverages it.

A lot of the time, AI doesn’t really save me time so much as let me spend it somewhere better. I may spend less time digging through and organizing research, but then use that time to sit with what it actually means. Or instead of running with the first decent idea, I can kick around several directions, figure out which one has legs, and push that one further.

Efficiency is useful.

Efficacy is what matters.

Human Doesn’t Mean Sloppy

One of the stranger side effects of all this is that we’ve apparently decided the best way to prove something was written by a human is to make it a little worse.

There’s now a whole genre of “make this sound less like AI” advice: don’t use em dashes, loosen the grammar, throw in a few mistakes, make the structure less polished, and generally rough things up enough that nobody gets suspicious.

Which is pretty silly when you stop and think about it.

We spent years trying to become better writers, and now the modern-day hack is apparently to make the writing worse so people believe we wrote it ourselves.

I understand why it happens. AI employs recognizable habits based on pattern orientation. You’ve seen the giveaway phrases, punctuation choices, and rhythms showing up everywhere. If you’re lazy with AI and let it do your thinking and expressing, it will sound like what it is: AI-generated content that’s technically correct but practically boring. Nobody wants to sound like the same machine-generated LinkedIn post they’ve already scrolled past fifteen times that morning.

But there’s a big difference between editing something because it doesn’t sound like you and intentionally lowering the quality so it appears more human.

If em dashes aren’t part of your natural writing style, don’t use them. That’s discernment. Avoiding them because you’re trying to fool some imaginary AI detector is theater.

A typo doesn’t make something human. A point of view does. So does humor, experience, instinct, and knowing when a sentence can be technically perfect and still land completely wrong.

The goal shouldn’t be to make AI-generated writing look more human.

The goal should be to make sure the human never disappears from the work in the first place.

AI Knows What Was. Humans Still Have to Imagine What’s Next.

There’s another reason I don’t think we should hand over too much responsibility to the machine: AI is fundamentally built on what already exists.

It learns from enormous amounts of human knowledge, language, behavior, patterns, history, and data. It can synthesize all of that at a scale no person ever could, spot connections we might miss, and generate new combinations with remarkable speed.

That’s incredibly useful.

But its frame of reference is still largely built from the past.

Human creativity doesn’t have to stay inside that frame.

Some of the most important ideas in business, art, music, and culture came from people who looked at the available evidence and decided the future didn’t have to resemble the past. They questioned the category. Broke the convention. Took the path that existing data couldn’t fully validate because the thing they were building didn’t exist yet.

If all we ever do is optimize based on existing data, we risk becoming incredibly efficient at reproducing yesterday.

Data can tell us what has worked, what people have done, which patterns exist, and what outcome is statistically likely. Those are enormously valuable inputs. But strategy and creativity also require imagination, intuition, conviction, and sometimes the willingness to pursue something the existing dataset would never have recommended.

AI can help us understand the box better than we ever have before.

Humans still need to know when it’s time to think outside of it.

That may be one of the most important things we bring to the relationship. AI can synthesize the past.

We still get to imagine the future.

In Sum

I suspect there will come a point when asking whether someone “used AI” sounds a little like asking whether they used Google to research an article or spell-check before sending an email.

Of course they did.

The more interesting question will be how they used it.

Access to AI is becoming nearly universal, and before long simply knowing how to prompt a model won’t be particularly differentiated either. What remains valuable are the same things that were valuable before any of this showed up: judgment, wisdom, empathy, experience, taste, imagination, character, and discernment.

The healthiest utilization of AI, at least as I see it, keeps the human firmly in charge and puts the software in its place. Feed it the right inputs. Use it to research, question, organize, and tune. Proof and push back when it’s wrong. Push ever harder when it’s shallow or superfluous. Don’t use it to outsource your thinking; use it to challenge your thinking. Never let it take ownership or responsibility for the final product. That’s on you. The pen you use shouldn’t own the letter you write. Nor should ChatGPT or Claude own your thoughts in word form.

The real benefit of AI, at least for me, is that it can clear away some of the busywork and leave more room for actual thought. It can contribute a lot along the way, but when something leaves my desk with my name on it, the judgment and responsibility are still mine.

Sometimes polished work really is AI-generated slop dressed in a nice suit.

Sometimes it’s entirely human. And increasingly, some of the best work will come from capable people using remarkably capable software with judgment, intention, and discernment.

That’s the balance worth protecting.