Taste Doesn't Ship. Specs Do.
Everyone Crowned Taste. Then They Handed an Agent a Vibe.
Earlier this year, a real fight broke out among tech leaders over what's scarce now that AI writes the code. Greg Brockman called taste "a new core skill." Paul Graham framed the differentiator as what you choose to make. Cloudflare's Dane Knecht echoed the same note. For a few weeks, "taste" was the consensus answer to the question every founder is quietly asking: if the machine can build anything, what's left for me?
Then the replies arrived. Vercel's Keith Messick summed up the backlash in one line — that "taste is the new core skill" was the rallying cry of men who kept Allbirds afloat. The meme did its job. It exposed something the original posts skipped over: taste is easy to claim and almost impossible to prove.
And here's the part that actually matters for anyone about to build a product with AI. Taste is upstream of nothing if you can't transmit it. The moment you hand the work to an agent, your sense of "good" has to live somewhere outside your head — or it doesn't survive the handoff at all.
The Skill Nobody's Actually Testing For
While the taste debate was getting memed, the June 2026 AI-developer hiring posts were quietly naming a different scarce skill: specification precision — can you write instructions that produce consistent, correct AI outputs? Not "do you have good judgment." Can you encode that judgment so a machine follows it.
That's a harder question, and it explains the whole fight. Linear's Nan Yu pushed back on the taste camp from the other side, pointing out that you probably don't have better taste than AI to begin with. Both observations land. The AI has plenty of generic competence. What it does not have is your specific intent — the edge cases you care about, the workflow your customers actually run, the thing that makes your product yours instead of a template.
So you have two camps that are each half-right. The taste camp is correct that judgment decides what to build. The precision camp is correct that judgment is worthless until it's written down in a form an agent can execute, fail against, and be caught by.
Taste without a spec is a vibe — and a vibe dies the instant you hand the work to an agent. You feel the loss as drift: the output is plausible, demo-ready, even pretty, and still subtly not the thing you meant. You didn't lose your taste. You just never transmitted it.
This is also why "taste" is a comfortable thing to claim in a podcast and a dangerous thing to hire for. It isn't measurable. Specification precision is. One is a personality. The other is a deliverable.
Taste Picks the Target. The Spec Makes the AI Hit It.
Reframe the meme and the argument dissolves. Taste decides what to build. The spec is the only thing that makes the AI actually build that. They aren't competitors; they're a sequence. The reason the debate felt unresolvable is that everyone was fighting over which half of one motion mattered more.
This is Codalio's wedge, stated plainly: vibe coding ships prototypes; spec-driven ships products. A prototype is what you get when taste meets an agent with no spec in between — fast, impressive, and impossible to trust in production. A product is what you get when intent is encoded first.
What does that encoding actually look like? A real spec isn't a wish. It's the smallest set of decisions that removes ambiguity from the build:
-
The business logic — the rules your product enforces, in plain language, before any code exists.
-
The inputs and outputs — what goes in, what must come out, and what "correct" means for each.
-
The edge cases you actually care about — the ones that define your product, not the generic ones.
-
The constraints — what must never happen, the lines the agent is not allowed to cross.
-
The acceptance check — how you'll know the AI built the right thing, not just a thing.
Notice what that list does. Every item is a place your taste gets written down so it can survive a handoff. That's the whole move. The spec is taste made transmissible — judgment turned into instructions an AI follows, fails against, and gets caught by.
What You Do Before You Build Anything
You don't need to become an engineer to do this. You need to refuse to start building until your intent exists outside your head.
Before you open a single AI tool, write down the one job your product has to do and the rule that makes it correct. Then write the case that would embarrass you if the AI got it wrong — that's your acceptance test, and it's worth more than any feature list. If you can't state it in a sentence, you don't have a spec yet; you have a vibe, and the agent will faithfully ship the vibe.
Then hand the agent the spec, not the mood. The difference shows up immediately: instead of judging output by gut feel and re-prompting until you're tired, you check it against the thing you wrote. Drift becomes visible. "Close enough" stops being a verdict you make under deadline pressure and starts being a measurement.
Do that once and the taste debate stops mattering to you. You already have taste. The work is making it leave your head intact.
Write the Spec Before You Build
Codalio exists for the exact moment this essay describes — the gap between knowing what good looks like and getting an AI to produce it. Its AI-powered, spec-driven workflow turns your business logic into a spec an agent follows, fails against, and is caught by, so the product you ship is the one you actually meant.
Try Codalio before you start building: walk through the spec-first workflow at codalio.com and you'll hand your next AI build a precise target instead of a vibe — and get back something you can ship, not just demo.
If you want a head start, the founder requirements worksheet that powers this workflow turns "I'll know it when I see it" into the five decisions above.
References
-
Is taste a "new core skill"? Techies debate — and get memed (AOL / Business Insider)
-
AI Developer Hiring 2026: the skills that matter — "specification precision" (Digital Applied)