The spec is the new wireframe

The shift you didn't notice

You've prompted an AI for a UI. The first prompt was short, something like "make a card with a title and a button." The output came back, technically correct and somehow wrong. So you added a sentence. Then another. By the third or fourth try you were writing paragraphs: padding, hierarchy, states, constraints.

What you wrote at the end wasn't a prompt. It was a specification. Designers and developers who never wrote specs outside of corporate documentation are now writing them several times a day, because the quality of any AI output is gated on the quality of the input.

The wireframe was a 1990s solution to a 1990s problem. The spec is a 2026 solution to a 2026 problem. The transition happened quietly, in everyone's daily practice, before anyone gave it a name.

What a spec that works looks like

A spec is not a wireframe in text. It is a structured statement of intent and constraint, written to leave no room for guessing.

The four parts that always show up:

• Intent. What the user is trying to do here, in one sentence.
• Structure. The layout in plain words: containers, slots, hierarchy, alignment.
• Constraints. The non-negotiables: tokens, components, sizing scale, breakpoints.
• States. What changes under load, error, empty, disabled, success. Almost always skipped, almost always the reason the output looks flat.

None of that is exotic. What is new is the discipline of writing it down before you (or the AI) start drawing.

Four ways your spec fails

Most bad outputs are not the model's fault. They're the spec's fault, and the failure patterns are predictable.

1. Too vague. "A clean modern card." Every adjective in that sentence means nothing to a model. Replace adjectives with measurable constraints.
2. Mixing form and function. You wrote what it should look like (rounded, soft, friendly) before saying what it does. The output then optimizes the wrong thing.
3. No constraints. You forgot to name the tokens, the components, the sizing scale. The model invents them. They look fine in isolation and clash with the rest of the product.
4. No states. You shipped a happy path. Production is mostly the unhappy path.

The fix for all four is the same: be specific where it matters, and only there.

What this means for our craft

For a long time, the visible artifact of design was the wireframe, then the mockup. Today, the visible artifact is the spec, even when the spec is invisible to the rest of the team, stashed in a Claude Code session or pasted into a Figma plugin prompt.

This is harder than it looks. A wireframe forgives ambiguity because a human will fill the gaps. A spec doesn't forgive: every gap becomes a hallucination. Writing a spec forces you to know your design system, your tokens, your edge cases, your states, your priorities. It surfaces every handwave.

That's why writing specs is becoming the most underrated skill in design. The designers who write them well are the ones shipping AI-generated work that doesn't look AI-generated. They're not better prompters. They're better thinkers.

The wireframe isn't dead because someone killed it. It's dead because we found a better way to say the same thing. The spec is what wireframes always wanted to be: precise enough that anyone, human or model, can build from it without asking a single question.