Production got free. Taste got expensive.

On a Sunday afternoon you can now produce a blog post, a hero image, a thirty-second video ad, a voiceover, and a backing track — start to finish, on a laptop, for less than the price of lunch. A decade ago each of those was a line item with a vendor attached: a writer, a photographer, an editing suite, a voice actor, a session musician. Now each one is a prompt.

That should feel like a superpower, and sometimes it does. But watch what happens to most of it: it disappears. It’s competent, on-brand, technically clean — and utterly forgettable. The flood of “fine” is the defining content problem of 2026, and producing more of it faster is not the way out.

When production becomes free, production stops being the thing that sets you apart. AI didn’t make content creation easier or harder — it moved the whole game from making things to choosing what’s worth making. In doing so it crashed the value of one skill and sent another through the roof.

This is just economics. When a thing becomes abundant, its price falls toward zero, and whatever was scarce around it becomes the prize. Production is now abundant. Taste, a point of view, distribution, and trust are not. Here’s where the value went — and what to do about it.

Making things stopped being the hard part

For most of history, the hard part of any creative work was the making. Words, images, sound, motion — each demanded a learned craft, hours of labor, and usually money. The skill and the output were welded together: you valued the writer because writing was hard, the photographer because lighting was hard. The scarcity of the doing was the whole basis of the value.

That weld broke. I’ve priced it out on this blog more than once. A real mixed hour of serious AI use — chat, images, voice, a video clip — came to $13.57, and one ten-second clip was 88% of the bill. A finished thirty-second vertical ad ran $4.58 and forty-six minutes. A six-shot product gallery from a phone cost sixty-three cents in API credits. The numbers only keep falling.

Sweep across the modalities and the shape is identical. A competent first-draft essay: seconds. A usable hero image: a few cents, then a few more for the variant. Studio-clean narration of an entire book: an afternoon. The team-and-a-budget version of each still exists — it’s just no longer the only door. The floor of “good enough” dropped all the way to the floor.

When everyone can make it, making it is worth nothing

Abundance has a price, and we’re paying it now. By November 2024, more than half of new articles published to the web were generated mostly by AI — a Graphite analysis of tens of thousands of CommonCrawl pages put the crossover at 50.3%. On Deezer, AI-made tracks hit 44% of everything uploaded each day by April 2026 — roughly 75,000 new songs every 24 hours, 85% of which the platform flags as fraud. Amazon had to cap self-publishing at three books a day back in 2023 just to slow the farms.

Stare at those numbers and the lesson is brutally simple. If a machine can produce a competent version of the thing in seconds, the competent version is worth roughly what it costs to make — which is approaching nothing. Economists call it the paradox of value: water is essential and nearly free, diamonds are useless and dear, because price tracks scarcity, not importance. Competent content just became water.

This is the part the “10x your output” crowd misses. Ten times more output, when everyone is holding the same button, is not an edge — it’s a contribution to the flood that buries you. Volume was a moat when volume was hard. Volume isn’t hard anymore.

Taste is the skill that survived

So what is still scarce? Start with taste — the ability to look at ten competent options and know which one is good. AI is extraordinary at generating and hopeless at judging its own work, because judgment requires a standard it doesn’t possess. It will hand you twenty fine images and rank none of them. Someone still has to.

That someone is doing the job that didn’t get automated. The model is the camera, the orchestra, the print shop. You’re the one deciding what’s worth shooting and what gets thrown away. In a world of infinite generation, the editor’s “no” is worth more than the generator’s “yes.”

Watch a film editor work and you see the whole shift in miniature. Generating footage was never the bottleneck; there is always more footage. The craft is the cut — what to keep, what to lose, the ninety percent that ends up on the floor so the ten percent can land. AI gave everyone unlimited footage. It gave no one the cut.

A point of view is the one thing a model can’t fake

Taste tells you which option is good. A point of view tells you what to make in the first place — the second scarce thing. A language model predicts the most likely next token, which means by construction it gravitates to the average of everything it has read. It is, almost by definition, the exact middle of the distribution.

The average is precisely what nobody remembers. The work that travels has a person behind it — a take, a grudge, a specific way of seeing that no average can produce. Ask a model for “a hot take on remote work” and you get a balanced summary of every take that already exists. The one thing it cannot give you is the thing not yet in the training data: your actual opinion, your lived experience, your proprietary numbers, the story only you were in the room for.

That is the un-trainable moat. Everything you have personally done, seen, measured, or screwed up is information the model never had. Lead with it and you’re making something the machine structurally cannot — not because it isn’t smart enough, but because it wasn’t there.

Distribution was always the real bottleneck

Here’s the uncomfortable one. Even with perfect taste and a sharp point of view, making the thing was never the hard part of getting paid for it. Reaching anyone was. AI drove the cost of production to nearly zero and did nothing about the cost of attention — if anything it raised attention’s price by flooding every feed you’re trying to be seen in.

The creator economy already learned this the hard way. There are an estimated 200-million-plus people making content; only around 4% earn more than $100,000 a year, and nearly half of full-time creators make under $1,000. The constraint was never the ability to publish — anyone can publish. It was the ability to be found. Make production free and you don’t dissolve that bottleneck; you make it the entire game.

Which is why distribution — an audience, an email list, a reputation, a place people already come to find you — is appreciating faster than any production skill. It’s the scarce asset AI can’t generate, because it’s built from other people’s trust and attention, and those don’t come out of a prompt.

Trust just became a premium signal

The flood has a second-order effect: it poisons trust, and trust then becomes worth paying for. When 44% of uploaded music and half of new web articles are machine-made, the default assumption flips. Readers and listeners no longer assume a human stood behind the work — they assume the opposite until shown otherwise.

You can watch the trust economy forming in real time. The Content Authenticity Initiative has passed 5,000 member organizations, and cryptographic “Content Credentials” now ship from Leica and Samsung hardware to stamp what’s real at the moment of capture. Amazon requires authors to disclose AI use. “Human-made” is becoming a label the way “organic” did — a premium signal precisely because the cheap, abundant alternative is everywhere.

For a creator that’s an opportunity, not a threat. A name people trust — a person who reliably ships things worth their time and stands behind the work — is the one asset that grows more valuable as the slop rises around it. Provenance is becoming product.

The job moved from the hands to the eyes

Put it together and the creator’s role has quietly inverted. The old job was the hands: render the thing, because rendering was hard. The new job is the eyes: decide what’s worth rendering, judge what comes back, and cut everything that’s merely fine. AI is genuinely great at the hands now. It is, and will stay, terrible at the eyes — because the eyes need taste, a point of view, and a standard, none of which it has.

So the playbook isn’t “use AI to make more.” It’s narrower and harder. Use AI for production leverage, and spend the time it frees on the parts that now carry the value: have an actual opinion, mine your own experience for what the model can’t know, build distribution like your livelihood depends on it — it does — and curate ruthlessly: ship the ten percent, kill the rest. For the same idea at ground level, the field-tested toolkits for newsletter writers and indie musicians come down to one move: let AI do everything except the judgment.

None of this is a reason to opt out. The leverage is real and it isn’t going back in the box. But the thing that pays — the thing that was always going to pay — isn’t the making. You don’t need to generate everything; you need to know what’s worth keeping. Production got free. Taste got expensive. The creators who understand which one they’re selling will be just fine.