AI Made "Average" Free. That's Why Taste Just Got Expensive.

There's a line you hear in every conversation about AI and creative work: "It gets you 70% of the way there, and a person finishes the rest." It's usually said with a shrug, as if the 70% is the interesting part.

It isn't. The interesting part is the stretch AI can't reach — the part that decides whether your work is worth paying for. And once you look closely at what actually lives in that stretch, the entire economics of creative work reorganizes around it.

AI collapsed the cost of "average"

Generative AI is extraordinary at one thing: producing plausible, competent, middle-of-the-distribution output, fast and nearly free. Ask it for a launch email, a tagline, a brand story, and you'll get something that reads fine. That's genuinely useful. It's also the problem.

"Competent and average" carries a hidden cost, and it's not hypothetical. A 2024 study in Science Advances found that generative AI raises individual creativity but reduces the collective diversity of what gets produced — people using AI wrote stories judged more creative on their own, yet those stories were markedly more similar to each other. The tool lifts the floor and, at the same time, quietly pulls everything toward a shared center of gravity.

Translated out of the lab: when AI writes everyone's first draft, everyone's first draft starts to rhyme. The baseline goes up and the distinctiveness goes down. If your work sits at that baseline, it now looks like everyone else's — and anything that looks like everyone else's is, by definition, hard to charge for.

Why the dice get expensive in creative work specifically

Here's the distinction most people miss. Engineering, data, anything with a checkable right answer: in that kind of work, AI is cheap to push to "done," because there's a ground truth you can test against and iterate toward. The cost of getting it right stays roughly linear.

Creative work has no such ground truth. "Correct" is a matter of taste, and taste is high-dimensional and subjective. So getting AI from good enough to exactly right means rolling the dice again and again against a target you can only recognize, not specify — and the hit-rate falls with every roll. Worse, you're now working against the model's strongest instinct, which is to regress to the mean. Distinctiveness is precisely the thing generative AI is built not to do.

That's the real cost curve. The more distinctive and finished you need the work to be, the steeper the marginal cost of squeezing it out of a machine that wants to give you the average. The last increment isn't a little more expensive than the first. It's the whole game.

And there's a second tax: the disclosure penalty

Suppose you get there anyway. There's still a cost waiting that has nothing to do with quality.

A growing body of research on AI disclosure (the "transparency dilemma") finds that simply telling people AI was involved reduces their trust in the work, even when the output is equal or better. The penalty isn't about quality; it's about authenticity. And it's strongest in exactly the categories where authenticity is the point: creative, emotional, relational, brand.

So the more visibly a piece of work reads as AI-made, the less it lands — regardless of how good it actually is. For a creative business, that's not a footnote. It's a structural headwind on anything that feels machine-produced.

The human at the end is doing two jobs, not one

Put those two forces together and the role of the human "finisher" looks completely different from polishing.

The person at the end of the process is doing two irreplaceable things at once. First, they close the taste gap the model structurally can't — the specificity, the point of view, the willingness to throw away the competent version for the right one. Second, they restore the authorship that makes people trust and engage with the work in the first place. One move recovers two things: the quality AI couldn't finish, and the trust AI's involvement quietly cost you.

That's why "a human does the last 20%" undersells it so badly. The human isn't contributing the final slice of the work. They're contributing nearly all of the differentiated value and all of the trust. The 70% the machine did is real and useful — but because anyone can now get that 70% for free, it's worth almost nothing on its own.

Where this actually goes (the honest part)

It would be easy to stop there and sell "AI draft, human polish" as a magic formula. It isn't one, for a simple reason: everyone is about to run that exact playbook. Within a year or two, "we use AI and then a human refines it" will describe every studio, every in-house team, every freelancer. The workflow will not be a differentiator, because everyone will have it.

What can't be commoditized is the taste itself — the judgment at the end, the point of view, the refusal to ship the average version. That's not a tool you buy or a process you copy. It's a capability you build, and it's the only thing left that's genuinely scarce once the machine makes "fine" free.

What we're betting on at slfemp.studio

This is the bet behind how we work. We use AI where it earns its place — getting to the starting line faster, clearing the blank-page tax, handling the commodity 70% — and we spend our real effort exactly where the value migrated: the distinctive last mile most people skip, and the human signature that makes work trusted instead of tuned out.

Because in a world where "average" is free and everyone's first draft rhymes, the job isn't producing more. It's producing the version that doesn't sound like anyone else — and standing behind it as a human who made a choice. That's the whole business now. It always kind of was.

If you want creative work that doesn't blur into everyone else's AI-flavored baseline, that's the conversation we like having. Let's talk.