A designer I know spent five days in 2020 producing a patient-intake flow for a telehealth startup. Thirty-one artboards in Figma, two rounds of red-lines, a Loom walkthrough, a spec doc, and a Notion page of open questions. At the end of the week, the flow got a thumbs-up from the PM and entered the engineering backlog, where it sat for another three weeks before a single pixel of it shipped.

Last month I watched the same designer (now three years older, same role) demo her workflow over Zoom. She typed a paragraph into Claude Design, wired the output into Claude Code against her team's React repo, and twenty-two minutes later she was poking at a running prototype on a localhost URL, typing into the actual form fields, watching the actual validation errors fire. She had not opened Figma once.

"I think I just don't make mockups anymore," she said. "Like, at all. I haven't drawn a rectangle in three weeks."

This is the thing I want to sit with this week. The artifact that the entire product-design profession has organized itself around for twenty years is becoming vestigial. Not overnight, not uniformly, but fast enough that if you are still producing polished comps as your primary output, you are working at the wrong altitude.

Why the mockup existed at all

The mockup is an artifact of friction. It exists because three gaps used to be enormous, and now all three are closing.

The first gap was between idea and running software. In 2005, spinning up a working prototype of a new flow meant committing an engineer for a week. That was expensive enough that nobody did it speculatively. You drew the thing in OmniGraffle or Axure or eventually Figma, and the drawing stood in for the build because the build was too costly to do twice. The mockup was a stunt double for the real product, and the stunt double was cheap.

The second gap was stakeholder literacy. Engineers could read code. Designers could read screens. Nobody else could read either. A PM reviewing a PR has always been theater. A PM reviewing a comp is an actual review. So the comp became the lingua franca of the cross-functional product meeting, not because it was the best representation of the work but because it was the only one everyone in the room could parse.

The third gap was the designer's own deliverable problem. If you do knowledge work that's mostly judgment, you need something tangible to point at in your one-on-one. "I thought about this for four days" doesn't land. "Here are fourteen artboards" does. Comps were as much a political artifact as a design artifact. They were how designers proved the work existed.

All three of those conditions are softening. Claude Design and v0 and Lovable and Bolt can produce running interfaces in the time it used to take to name a Figma file. Stakeholders can poke at a live prototype on a URL just as easily as they can click through a Figma prototype, and the live version is more honest. And the tangible-deliverable problem is solved better by a link to a working thing than by a screenshot of a fake one.

When all three reasons for an artifact go away, the artifact goes away too. Not tomorrow. But it goes.

What replaces it

Here's where I think a lot of the conversation is miscalibrated. People keep looking for the next artifact. They ask what "the new Figma" is going to be, or whether Framer will eat it, or whether design-to-code pipelines will finally make Zeplin-style handoff obsolete. Wrong question. The replacement for the mockup isn't another artifact. It's a different skill set.

In the world I'm describing, the working designer produces three things, none of which are comps.

The first is a design system expressive enough that a model can extend it without asking. This is the dividend I wrote about last issue. If your tokens are documented in Design Tokens Community Group format, if your components have machine-readable usage rules, if your accessibility constraints are encoded rather than folklore, then a model can generate work that ships against your system. If any of those are only in someone's head, the model has to invent, and what it invents will feel like a different product. Design systems used to be a nice-to-have documentation layer. They are now the API surface your team exposes to the generative layer. No system, no generative upside.

The second is prompts and constraints that shape what the model generates. This sounds trivial. It is not. The difference between "make me a settings page" and a prompt that actually produces usable output is eight hundred words of accumulated context: who the user is, what their mental model is, which pattern library you're drawing from, what error states you care about, what the tone of microcopy should be, what the page this settles against already does. Writing that prompt is the design work. The image the model returns is the artifact of the prompt, the way a rendered Figma file used to be the artifact of a sketch. Designers who can write a rigorous, specific, constraint-loaded brief are about to eat very well.

The third is editorial judgment. When a model will give you ten options in forty seconds, the bottleneck stops being production and starts being selection. Which of these ten should ship? Which five don't even count as valid responses to the prompt? What does "good" look like in this context, specifically, with evidence? This is the muscle that gets thinner the less you exercise it. Most designers I know undertrained it because production was where the day went. It's where the day should go now.

The mockup was the work. The decisions are the work. The deliverable used to be the drawing. The deliverable is now the encoded judgment.

If you've been in this job long enough to have a Figma hygiene setup you're proud of, the thing you're proud of is about to matter less than the thing you were doing while you built it.

The uncomfortable part

Not every designer is going to like this, and I want to be honest about why.

If what you trained on was the craft of making a beautiful comp, the thing you are good at just got cheap. That's a real loss. The satisfaction of pushing a button component through fourteen iterations until the focus state is exactly right, the pride of a portfolio case study full of polished screens, the identity of "I am the person on this team who makes things look right", all of that is being unbundled from the job at some speed. You can feel the vertigo in the design-Twitter discourse right now. People are worried, and they should be.

But here's the distinction that matters. The designers who were using the comp as a communication tool for deeper work, who were always really doing systems thinking, information architecture, strategy, taste, user advocacy, and who treated the Figma file as a byproduct, those designers are about to compound. The value they were adding was never the comp. It was the thinking the comp represented, and the thinking is exactly what a model can't do yet.

The designers who confused the comp for the craft are in trouble. And there are more of them than the profession likes to admit. A lot of portfolios I've reviewed over the last decade were beautiful execution against a design language somebody else defined, and nothing else. That work was always commoditizable. The commodification just arrived faster than anyone expected.

I'm not saying this gleefully. Several of my friends are in that bucket. The honest advice is: figure out which bucket you're in, fast, and if it's the second one, start climbing toward the first.

What to do Monday morning

Four concrete moves.

Export your design system to Markdown and JSON. If your tokens only live in Figma variables, get them into Style Dictionary or a similar format and into your codebase. A model can read the codebase. It can't read a panel you haven't shared.

Spend a week working in Claude Design, Cursor, or v0 as your first move for a new problem, not your last. Figma becomes the place you go when you need pixel-level control after the shape is settled. Not the place you start.

Practice writing design briefs that a model could execute. Pretend you are dispatching a contractor who is fast, literal, and has no context. Write that brief. Run it through Claude Design. See what comes back. Edit the brief, not the output. Repeat until the brief is tight enough to produce a shippable result.

Time-box your Figma work. If you spent thirty hours in Figma last week, try to spend ten this week. Use the twenty you get back on the three things above.

Signoff

Next week: what happens to agencies whose entire billable model was comp production. The hour-for-hour math stops working when the comp gets cheap, and I've been talking to a few founders about what the replacement business looks like. Preview: it's less like a studio, more like a law firm.

Hit reply and tell me what's on your mind. I read all of it.

Jameson