Mocking fell from 70 percent of the design week to 30, and the number says which half of the job…

Mocking fell from 70 percent of the design week to 30, and the number says which half of the job survived

Jenny Wen leads design for Claude Cowork at Anthropic, and before that she led FigJam and Figma Slides at Figma, which makes her one of a small number of people who have designed the tool designers use and then designed inside the thing that is replacing half of what the tool was for. On Lenny’s Podcast, in the episode with Jenny Wen released in March 2026, she gave a number that I have not seen anyone in the design discourse take seriously enough. Asked how her week has changed, she said: “A few years ago, 60 to 70% of it was mocking and prototyping, but now I feel the mocking up part of it is 30 to 40%.”

That is one person’s estimate of her own time, offered with a hedge, and I want to treat it with the respect that honest numbers deserve rather than the reverence that gets applied to anything said by someone at a lab. It is still the most useful figure I have heard about the design job this year, because it is specific about what shrank. The mock shrank. The thing designers were trained to produce, the thing portfolios are made of, the thing that took the majority of the working week for as long as there has been a working week in software design, lost half its share. Everything that follows in this essay is an attempt to read what took its place, and to say where I think she is right about it and where I think she is conceding too much.

What the missing forty percent was for

Wen’s talk in Berlin the previous September was called Don’t Trust the Design Process, and the line from it that travelled is the one she repeated on the podcast: the research-then-diverge-then-converge loop that “we sort of treat it as gospel. That’s basically dead.” The talk drew backlash, which she reads generously as coming from people who “have invested their entire careers in learning, teaching, using this really stable design process.” She also said the talk already “feels outdated” to her, a few months after giving it, which is the sort of admission that makes me trust the rest.

Her explanation for the collapse is mechanical rather than philosophical. Engineers “can go off and spin off their seven Claudes” and produce a scrappy working version of an idea before a designer has finished the first artboard. “You as a designer actually do not have the time to make these beautiful mocks anymore,” she said, and the phrasing matters: you do not have the time, meaning the mock now arrives after the decision it was meant to inform. A mock that lands after the prototype has been tried is a receipt, and nobody needs a beautiful receipt.

I want to be precise about what the mock was for, because I think the discourse has muddled two things. Some of the mock was specification: the exact type sizes, the spacing, the states, the behaviour of a control at the edges. Most of it was persuasion. A polished mock existed to get a room of people to agree that a thing should exist, and its polish was doing rhetorical work, the way a rendering of an unbuilt building does. When a working prototype can be produced in an afternoon by someone who is not a designer, the persuasion job transfers to the prototype, and the mock is left holding only the specification job, which was always the smaller share. That is the forty percent. It was the rhetoric.

The specification share has a further problem that Wen names directly. With non-deterministic products, “you can’t mock up all the states,” because the states are produced by a model whose outputs you cannot enumerate. You can draw the loading state and the empty state. You cannot draw the state where the model returns a 400-word table when the user asked a yes-or-no question, because you do not know it will until someone tries. For that class of product, the mock cannot even do the specification job well. The prototype running on the real model can.

What filled the gap

The half of the week that opened up did not go to leisure. Wen says “there’s that other 30 to 40% there that is now jamming and pairing directly with engineers,” plus a slice she could not size that is now building, meaning she is shipping changes herself. Read those together and the shape of the new week is a designer sitting next to the engineers who are sitting next to the agents, reviewing what comes out and steering it, occasionally committing a change herself.

She calls this execution support, and I have watched designers hear that phrase as a demotion. It sounds like the designer has been moved from author to assistant. I do not read it that way, and the reason is what she says about where design work now splits. In her account “design work is becoming really stratified in this new world,” into support for execution on one side and direction on the other. The execution side is where the volume is, and volume is where slop comes from. Seven agents producing seven scrappy versions of a feature each week is a machine for generating plausible interface at a rate no team can review by eye, and the designer in the pairing seat is the person who decides which of the seven ships, which get merged and which get thrown out. That is a reviewing job, and reviewing is what a trained eye is for. The mock was one way of exercising that eye. Sitting in the diff is another.

There is an observation about the work itself here that I think matters more than it sounds. Wen still uses Figma for exploring options and for fine interaction detail. So the drawing tool survives for exactly the two jobs a prototype does badly: laying out six alternatives side by side so the difference is visible, and getting a transition or a control to feel right at the level of pixels and milliseconds. Those two jobs are the specification remnant of the old forty percent, and they are the part I would fight to keep on a team’s calendar, because they are where the eye is trained and where the difference between competent and good gets made.

Direction is now a season long

The second stratum in her account is direction, and it has shrunk in a different dimension. Design visions used to run two years out, or ten, and were delivered as decks. Now, she says, it “usually becomes a vision that’s three to six months out,” and the artefact is often “just creating a prototype that points people in the right direction.”

I find this the most interesting part of the conversation, because a season-long vision is a different kind of object from a decade-long one. A ten-year vision deck was a work of fiction that a company agreed to believe, and its design quality was mostly the quality of its storytelling. A six-month prototype is a proposal you can hold in your hand and argue with, and its quality is whether people who try it want more of it. The prototype does the pointing job better because it can be wrong in a way a deck cannot. A deck is never wrong; it is just not yet true. A prototype that fails in the hand is wrong on the spot, which is information.

Wen’s argument for why direction still matters is about coherence, and it is the same argument every design leader makes when they are being honest about what the job is for. When anyone can produce any feature in any direction, “you need to point them towards something,” or the product becomes a pile of features that each made sense to whoever built them. I would add only that the pointing is now cheap to do and expensive to do well, because the prototype that points can be produced by anyone, and the one that points somewhere the company should actually go requires the person making it to have decided that, which brings us to the part of the episode where I part ways with her.

Where I think she concedes too much

Lenny asked whether AI will get very good at taste and judgement, and she answered that it will: “I think it will get better at taste and judgment and design. Yeah, I think we might be holding onto that a little bit too much.” I understand why someone who watches these models daily says that, and I think she is right and wrong in a way that is worth separating.

She is right that the models will get better at the taste that most design work requires, which is the taste of the competent average. Choosing a type scale that does not fight itself, spacing that breathes, a palette with one accent, a layout that reads top-left to bottom-right without tricks: this is taste in the sense that a good design school produces it in most graduates, and a model trained on the accumulated output of those graduates will produce it on demand. I concede that more fully than most designers I know, and I have written elsewhere that this is why the middle of the design market is in trouble.

Where I disagree is on the taste that has commercial value, which is the deviation from the average that a particular product needs at a particular moment. The value of a design decision is often that it is unlike what everyone else is doing, and a model whose competence comes from having absorbed what everyone else is doing is structurally poor at that, for the same reason a survey of what people already like is a poor guide to what they will like next. It can propose deviations. It cannot know which one is right, because “right” here means right for this company, this quarter, this audience, this price point, and that judgement is not in the training data because it has not happened yet. Wen’s own qualification gets at this. She says the judgement that matters is “judgment around what to do next,” and distinguishes it from aesthetic taste. My point is that the aesthetic taste that pays is a species of judgement about what to do next, and the two cannot be pulled apart as cleanly as her answer implies.

There is a second reason I would hold onto taste more tightly than she suggests, and it is unglamorous. Taste is a discipline that decays when it is not exercised. A team that lets the model choose the type scale for a year will not be able to tell, in the second year, whether the type scale is any good. If the designers in the pairing seat are reviewing without still occasionally making, the eye that does the reviewing dulls, and the review becomes a rubber stamp with a job title. This is my argument for keeping the surviving thirty percent, the exploring and the fine detail, on the calendar as a deliberate practice rather than as leftover.

The half that survived is deciding

Where Wen and I fully agree is on the thing the number does not shrink. “Someone has to decide what is actually going to get built and what actually matters,” she said, and the hard parts of building software “are actually not building it.” The hardest days at work, in her description, are the days when you and someone else disagree about what goes into a feature, and AI “can’t necessarily solve this dispute between you and somebody else.”

I have come to think that this is the job, and always was. The mock was the evidence a designer brought to the dispute. The prototype is better evidence. Neither settles anything on its own, because a dispute about what to build is a dispute about what the company is for, and that gets settled by a person who is accountable for the answer. A model can weigh in, as she says. It cannot be fired for the outcome, and until it can, the decision belongs to someone who can be.

If that is right, then the surviving half of the design job is direction plus adjudication, and the practical consequence for a working designer is a reallocation of the week that most teams have not made on purpose. Less time producing the artefact that used to carry the argument. More time in the room where the argument happens, with the prototype as your exhibit. More time in the diff, deciding which of the seven versions is the one. And a protected slice, smaller than it was, for the drawing that keeps the eye sharp enough to make those calls.

The uncomfortable implication is for portfolios, which still overwhelmingly show mocks, and for design education, which still overwhelmingly teaches the loop Wen called dead. Both are optimised for the seventy percent that became thirty. Neither has a good format for showing that a designer can hold a room and decide. I do not have a fix for that. I only notice that the people who will be hired for the surviving half of the job are currently being assessed on the half that went away, and that a number as plain as Wen’s should be enough to start changing what we ask to see.

Mocking fell from 70 percent of the design week to 30, and the number says which half of the job… was originally published in Bootcamp on Medium, where people are continuing the conversation by highlighting and responding to this story.