17x the Code, 1.3x the Releases. Design's Job Is the Spec Now.

This week Figma held its first Config in India and shipped something I did not expect from a canvas company. A text editor.

It is called Specs. Per Ivan Mehta's report at TechCrunch, it is a Markdown editor where a team defines its design language, and it is meant to be read by people and by agents. It can sit on the canvas next to the designs and link to them. Alongside it, Figma pulled design libraries, code libraries, and guidelines into one place agents can follow, and moved code layers into public beta.

Then Loredana Crisan, Figma's chief design officer, gave Mehta the number that explains why. There are 17 times more lines of code being produced today than in the past. Releases are up 1.3 times. And a lot of that code, she said, "is not necessarily built to ship, but really built to discover what the product is."

Read those two numbers together and you have the whole series.

This is part 1 of The Design Transition. It runs twice a week, each part written against that week's findings, for two readers at once: designers and design leaders living through this, and the product, engineering, and executive leaders who have to reorganize around it. I've led product and design, research included, for more than ten years. This is the shift I'd want both groups reading the same page about.

The short version

AI now generates the screens, the flows, and the first prototype, and design did not shrink. It got promoted. The evidence this week is Figma, a company that sells the canvas, shipping a spec editor for agents while its design chief reports 17x more code and only 1.3x more releases. Generation is not the bottleneck anymore. Deciding what is good enough to ship is, and that decision has always been design's. The job moves from producing artifacts to specifying behavior: the constraints a surface must never break, the acceptable range of output, the rubric that scores it, and the wrong path. That makes design the other half of the Product Builder collapse. The Product Builder makes more things. Design decides which ones deserve to exist.

The gap between 17 and 1.3

The obvious reading of Crisan's numbers is waste. All that code, so few releases. I read it the other way. Most of that code is doing what sketches used to do. It's exploration that happens to compile.

Which means the expensive step moved. It used to be making the thing. Now it's choosing among the things, and saying out loud why one is good enough and the others aren't.

The same pattern shows up inside Anthropic, per a Designer Fund profile of its design team published October 6. Joel Lewenstein, who leads product design there, estimates engineering is moving about 10x faster and design 3 to 5x. The piece is careful to call that an illustrative estimate, not measured data, and I'll treat it the same way. But the direction matches Figma's. Engineering can build the feature in 48 hours, so a week of design exploration gets hard to defend.

The tempting response is to make design faster at drawing. Generate the mockups too, close the gap, keep the old job at the new speed.

That's the wrong fix. If the screen is free, a faster screen is worth nothing.

Look at how Meaghan Choi, who leads design for Claude Code and Cowork, works in the same profile. She asks for five variations, has Claude recommend the strongest with its reasoning, and then makes her own call. The generation took minutes. The call is the part nobody else on the team was going to make.

That call is the job. It was always the job. It just used to be hidden inside the hours it took to draw the screen, so nobody had to name it. Now the drawing takes minutes and the call is exposed, sitting there on its own, and the question design leaders keep getting from their CEOs is whether that call is worth a salary.

It is. But only if it's written down.

The middle drained, and the spec moved ends

Anamol Rajbhandari made a sharp version of the structural argument in UX Collective on October 13. For fifteen years, he writes, the design process kept growing while the time left for making the product kept shrinking. Personas, journey maps, workshops, wireframes, handoff specs. That middle is draining now, and the weight is moving to the two ends, deciding and making. An hourglass.

I agree with the shape. I'd move one thing.

He lists the handoff specification with the middle, and the old one was. It described a screen someone else would build: redlines, spacing, states. Nobody needs that document when the agent builds from the design system directly.

The spec that matters now is a different object, and it sits at the top of the hourglass, with the deciding. It doesn't describe a screen. It describes behavior across every screen the system might generate. What the surface must never do. The acceptable range, because a generated interface has a distribution, not a final state, and you cannot mock a distribution. The rubric that scores output without the designer in the room, which is why the rubric is the spec. And the wrong path, the shape of uncertainty, refusal, and recovery, owned as surfaces in their own right.

That's what Figma just built an editor for. A spec in Markdown, readable by agents, living next to the canvas. When the company that sells the canvas ships a text file as the headline feature, the market is telling you where the design work went.

I argued the frame for this in the first chapter of the Design handbook, Design Just Got Promoted. Drawing got cheap. Deciding didn't, and there is more to decide now than at any point in the last twenty years. This series is that frame meeting each week's news.

The other half of the Product Builder

On the product side I've been writing about a collapse. The PM who wrote specs and waited is being replaced by the Product Builder, who ships a working prototype because a prototype costs a day. The FDE series made the same argument from the services side: the FDE is the Product Builder wearing a visitor badge.

That collapse leaves a hole, and design is what fills it.

When the PM can generate four directions before kickoff and the engineer can generate an interface that looks designed, nobody in that seat is structurally responsible for whether the thing is good. More output, same people, nobody owning the bar. That's how you get 17x the code and 1.3x the releases, and my worry is that what does ship drifts toward the median of everything the model has seen.

So the Product Builder needs a counterpart. Someone who decides what good means for this product, writes it down so the agents and the builders can hit it, and makes the ship call against it. That's design's half. The Product Builder makes more things. Design decides which of them deserve to exist, and puts the reasons in a file other people can run.

If that sounds like a smaller job, look at the hiring data. In Figma's design hiring study, published in February, Andrew Hogan reported that 47% of design leaders saw their need for designers increase, 35% saw it hold, and 18% saw it decline. And 79% of hiring managers said the need for people who can design AI products is growing. It's a survey of design managers, so read it as their view, not a census. But it doesn't read like a funeral. It reads like a job description changing.

It's also the same structure I described in Stop Picking Winners. Design the Tournament.: when variants are cheap, the work is the scoring, not the picking.

Where this goes next

The series takes the promotion apart one relationship at a time. Next is design and the PM, and who owns the spec when the prototype is free. Then the handoff to engineering, where a rubric and an eval replace the mockup. Then the design system becoming the context agents read, which is exactly the ground Figma moved onto this week. Then designing for output you can't predict, research at generation speed, critique when the first draft is the model's, what to measure, how juniors learn when AI does the production work, hiring for judgment, what the tools actually changed, the exec room, and what breaks in a design org at 50 and at 200.

Each part will be written against what happened that week. When nothing happened, I'll say so.

One thing to try this week

Pick one surface in your product that generates output. A summary, a draft, a recommendation, a generated layout. Write its spec on one page of Markdown. Four sections: what it must never do, the acceptable range from best case to floor, four or five rubric dimensions with what a pass looks like on each, and what the user sees when it's wrong.

Then hand that page, and no mockup, to whatever agent your team uses. Generate five versions. Score each one against your rubric, and have one other person score them too, separately.

Wherever the two of you can't agree on a score, the spec is missing a line. Write it. That page is the job now.

Sources: Ivan Mehta, "Figma announces new features to define how agents work on its platform," TechCrunch, October 14, 2026, with Loredana Crisan quoted. Designer Fund, "When code is no longer the constraint: Inside Anthropic's design team," Designer Fund, October 6, 2026, with Joel Lewenstein and Meaghan Choi. Anamol Rajbhandari, "AI is hollowing out the middle of design, leaving an hourglass behind," UX Collective, October 13, 2026. Andrew Hogan, "Why Demand for Designers Is on the Rise," Figma, February 10, 2026.

Part of the running argument on AI Product Management: AI collapsed the cost of the parts product teams were trained to be good at, and on the design side that means the screen got cheap while the spec, the rubric, and the ship call became the job.

Related answer: What is Figma Specs, and what does it change for design teams?