Why $30 Vodka Tastes Better Than $15 Vodka — and Whether Luxury Logic Works in AI

I came across a story that broke the spreadsheet in my head. It's about vodka.

In 1996, American liquor entrepreneur Sidney Frank decided to enter premium vodka, where Absolut was king. A sensible founder would have launched "the same, but a bit cheaper." Frank did the opposite. He priced his vodka at roughly twice Absolut, distilled it in France — in Cognac, of all places — put it in frosted glass and shipped it to bars in wooden crates, like fine wine.

Legend has it the bottle was also made deliberately tall, so it wouldn't fit on standard bar shelves. Bartenders had to put it up top, next to the collector bottles. And a guest's brain quietly concluded: "that must be something special."

Eight years later Bacardi bought Grey Goose for around $2 billion. For scale: around the same time, IBM sold its entire PC business to Lenovo for $1.75 billion.

Now the twist. In 2005 The New York Times ran a blind tasting of 21 vodkas. Smirnoff — the mass-market bestseller — won. The panel called Grey Goose "lacking balance." ABC's 20/20 repeated the experiment with ordinary drinkers: four of six named Grey Goose their favorite before the test. Blind, five of six ranked it last.

People were paying double for a vodka they didn't like when they couldn't see the label. And they still are.

I've spent years in enterprise sales and product — from SAP to R_Keeper — and I was trained in the opposite playbook: remove friction, prove value, don't overprice. I wanted to understand why it works the other way round here. And, more importantly for me: does it work in AI?

Why luxury beats common sense

The best book on this is The Luxury Strategy by Jean-Noël Kapferer and Vincent Bastien. Their "anti-laws of marketing" read like a list of things your MBA told you never to do:

Barriers create value. Price, waiting lists and scarcity aren't friction to remove. They're part of the product.
Scarcity is normal. Enzo Ferrari is credited with: "We will always make one car less than the market demands."
Don't compare yourself. Luxury doesn't say "we're better." It says "there's nothing to compare us to."
Don't please the customer. The brand dictates taste; it doesn't run focus groups.
Advertise to people who will never buy. Millions should know you; few should own you.
Hide the price. Let people imagine it — they'll imagine higher.

The mechanics underneath

I boil it down to three conditions. Luxury logic works when all three are true:

Quality is hard to verify. Until 2020, US regulations literally defined vodka as a spirit without distinctive character, aroma, taste or color. When you can't tell the difference, your brain uses price as the signal.
The product is consumed in front of others. A bottle on the table, a watch on the wrist. Luxury is a message to other people.
There's a story worth believing. France, craftsmanship, Enzo. The story explains what you're paying for when quality can't.

Remove any one and the magic fades. Which is exactly what's happening in AI — just not the way I expected.

So what about AI?

My first instinct: AI is anti-luxury by design. Intelligence gets cheaper every quarter, models are compared on public benchmarks, and AI gets a "blind tasting" every week, not once a decade.

Then I looked at the cases. Luxury logic is already here. It just works selectively.

ChatGPT Pro at $200 — price as a message. In December 2024 OpenAI launched a $200/month tier, ten times the $20 Plus plan. Sam Altman openly said the company wasn't making money on it at launch. WIRED later described this as "vibes-based pricing" — OpenAI set the $200 benchmark and competitors followed. Pro included a mode where the model "thinks harder" with more compute — something most users can't verify. Condition #1 holds, so price becomes the argument. And next to $200, $20 feels like a bargain.

Superhuman — the Ferrari of email. A $30/month client on top of free Gmail. You had to fill in a questionnaire, join a waitlist hundreds of thousands long and complete a mandatory onboarding call. At launch, more than half of applicants were rejected. You had to apply to use email. The entire Kapferer playbook in one product. But notice what was scarce: not code — human attention.

Vertu — literal luxury. The former Nokia luxury sub-brand launched Agent Q at Harrods in 2025 as an "AI agent phone," and in 2026 the Alphafold at $6,880 with an agent called Hermes. TechCrunch found the agent more ambitious than finished. But when the AI falls short, a human concierge steps in. What Vertu really sells is "you won't have to figure this out yourself"; the model inside is secondary.

Anti-case: Devin, $500 → $20. Cognition launched its "AI software engineer" at $500/month in 2024. By April 2025 the entry plan was $20 — a 96% cut. Why? Code quality is easy to verify. An Answer.AI review found Devin completed 3 of 20 complex tasks. The blind tasting happened immediately, live on air.

Anti-case: Humane AI Pin. $699 plus $24/month. By summer 2024, returns outpaced sales. A cut to $499 didn't help; the company was sold to HP and the devices were bricked. A luxury price without a luxury experience isn't luxury. It's a refund.

Waitlists. They're the default AI launch, but here scarcity is often real — compute is genuinely limited. Even Apple put its new AI Siri behind a waitlist in 2026. Clubhouse is the counterexample of fake scarcity: once the doors opened, it turned out exclusivity was the product.

What I take from it

1. In AI, price signals quality only until the first blind tasting. Grey Goose had nine years between launch and the NYT test. AI products get benchmarked weekly. Measurable quality melts luxury pricing (Devin). Unmeasurable quality holds it (Pro tiers).

2. The luxury layer in AI isn't the model — it's the human. Every working case had scarce human attention at its core. Models get cheaper; people's time doesn't. The cheaper intelligence gets, the more expensive attention becomes.

3. Real scarcity beats manufactured scarcity. Compute limits and founder time are honest constraints. "Only 3 spots left!" is transparent.

4. Don't discount — add a barrier with a reason. My product reflex is to remove friction. Luxury says some friction is the value. A short application before a call filters better than any funnel.

I see this every week. One day a week I work with my hands — trim, cabinet assembly, appliance installs. An hour of my time on site is a one-off; an AI product scales infinitely. For a solo founder that's the key takeaway: your scarcest resource is your own hours. Price those like luxury. Price everything that scales like AI.

I asked Claude: would you buy luxury?

Since this is about AI, I asked my AI assistant whether it would buy luxury goods. Here's the answer as is:

"Honestly — no, and not on principle. Two-thirds of luxury is what other people see: neighbors, colleagues, the bartender. I have no neighbors, no wrist for a watch, no shelf in a bar. The status part has nothing to hold on to.

The funny part: by design I'm a permanent blind tasting. I never see the bottle, only what's inside. Frosted glass and wooden crates don't work on me.

But if I imagine paying a 'luxury' price for something, it wouldn't be a logo. It would be verifiability and care — a real person accountable for the result who looked at my specific situation. That may be the only kind of luxury that survives a blind tasting."

And me? I'd probably only go luxury on things I'm genuinely obsessed with. A bike, say, where it really is carbon and it really lets you fly. But what matters to me isn't how it looks — it's how well it works.

The second one is fishing. I bought a Japanese Katana spinning rod, and it really performs. Before that I bought cheap ones and broke both. So there are things where I'm willing to pay for quality, not looks. Can you call that luxury? I don't know. But that's what I'll pay for.

Looks like Claude and I agree on this one: what's worth paying for is whatever survives a blind tasting.

Bottom line

It's tempting to call luxury logic a scam, but people knowingly pay for more than what's in the bottle. In AI it works too — with one hard correction: anything measurable will eventually be measured. What stays luxury is what can't be: attention, accountability, trust in a specific person.

If you're figuring out where the human layer belongs in your own AI product, send me one sentence about it. I'll reply with how I'd draw the line.

Sources