← Back to blog
AiAbout 7 min read

At Zcool's Twentieth-Year Design Week, the Theme Was a Negative Sentence

Published Nov 8, 2026
At Zcool's Twentieth-Year Design Week, the Theme Was a Negative Sentence

The theme of this year's Zcool Design Week is "In the AI era, I don't design this way." It is a strange slogan for a design conference, because it defines the work by refusal. It is also the most accurate description of what a thousand designers spent a week talking about.

The event is the tenth edition, and it coincides with Zcool's twentieth year. The platform launched in 2006 as a place for designers to post work and argue about it. Two decades later the same community gathered in one hall, and the argument had changed from whether to use the tools to what is left to sell.

The numbers people repeated

Tudou_man, a digital artist who was among the first wave to build a business on AI work, described his own sequence of questions. In 2023 he was asking what the tools could do. In 2024 he was asking what his value was. This year he is asking where to go.

The concrete comparison he offered is the one that travelled. In his advertising years, a high-end illustrator could charge 50,000 to 200,000 yuan for a single piece. That market has largely evaporated. His explanation was not that the work got worse: it got too easy, and once it is easy you cannot argue for the price.

Long Jieqi, who founded TOPic & Loong after 37 years in advertising, described the client-side version. Budgets are shrinking year over year, and clients are asking the obvious question. If AI lowers the cost that much, why pay the old price for the idea.

Neither of them was making a complaint about tools. Both were describing a pricing problem, and the pricing problem has a specific structure: the part that got cheap was the part that used to be quoted.

What the five minutes were replaced with

The most useful material from the week answered a different question: which work expanded to fill the space that pricing left behind.

Mai Ju works as an AI casting director, designing characters for film and television. The briefs she receives are the kind of language that has no operational meaning: a premium face, aloof but not cold, a sense of story, intellectual. She said the words sound abstract because they are judgements, and a judgement has no standard.

Her method is to decompose a face into four dimensions, bone, skin, form and spirit, and then ask at most five binary questions. Lively or steady. Sculpted or delicate. Heavy or light. Focused or relaxed. Distant or warm.

A sharpened pencil beside blank sticky notes and pencil shavings on a worn table.

That is a requirements interview, and it is the thing that did not exist before. When a client could not describe what they wanted, the illustrator improvised and the client reacted. Now the model produces forty plausible faces instantly, so the ambiguity has to be resolved before generation rather than after. Someone has to run the interview.

The same pattern appears in the production numbers. A team that made four Club World Cup AI commercials for Hisense ran each project with three or four people over two to three weeks. Compute was about five percent of the execution budget. What consumed the rest was the selection: going through the takes and finding the ones that could be used.

Adobe's version of the same shift is a table of data in Illustrator turning into fifty variants. Fifty variants is fifty chances to make one decision, and nothing in the tool makes it for you.

The two-thirds that the boss cannot see

There is a small scene in the reporting that explains the whole market. A founder at a consumer goods company sends an AI-generated promotional image to the company's only graphic designer and asks for a quick polish, assuming the AI did eighty percent of the work.

The designer opens it and finds the packaging is wrong, the text cannot be used, and the brand colour is off. The remaining twenty percent, in the founder's estimate, is the difference between an image that ships and an image that gets pulled.

The estimates are inverted. What the model did was the part that used to take an afternoon. What remains is the part that decides whether the afternoon was wasted. Everyone who has shipped commercial work knows this, and it is remarkably hard to put on an invoice, because the visible output is a picture that already looks finished.

What clients are actually buying

The five binary questions are a small thing that describes a large one. A brief that says "premium, aloof but not cold" is not a specification, and the model will happily resolve it forty different ways. The interview exists because someone has to convert taste into constraints, and that conversion has a cost that does not shrink when generation gets cheap.

The production numbers point the same way. Three or four people for two to three weeks on a set of commercials, with compute at five percent of the execution budget, is a studio running on coordination rather than rendering. Most of the calendar is spent deciding, reviewing and redoing, and the rendering bill is noise.

Which reframes the pricing collapse. When a client says AI made it cheap, they are quoting the part of the invoice that used to carry a line item: the illustration. The rest of the invoice, the part where someone sat in a meeting and turned a vague adjective into a workable direction, was never itemised and is now the only part that still costs anything. Designers have spent two years discovering that selling that part requires explaining it, to clients who believed it was free.

Why the theme is a negation

A conference theme is a positioning statement, and this one is shaped by what has already been claimed. Every tool vendor has spent two years saying AI designs. A design community cannot credibly say "we design with AI too," because the sentence is already taken and the tools are available to everyone in the room.

So the position became the refusal: not this way. Which is a real position, and also an admission that the affirmative version of the claim is no longer available.

The economics behind the admission are visible in the room. When everyone has the same models at the same prices, tool access stops being a differentiator. What is left is taste, brief translation, selection and accountability, none of which scale the way a production line does. A studio that used to win on throughput now competes on judgment, and judgment is harder to sell by the hour.

The uncomfortable part

There is a version of this story where designers adapt and everything settles. The reporting does not support the tidy ending.

The people who described the most pressure were the successful ones. Tudou_man's line was that the anxiety exists at the top too, and that the previous cycle's advantage has already been spent. If the winners of the last round are re-asking the question, the answer is not a skill upgrade.

What the week actually documented is a market splitting. At one end, work where a specific person's judgement is the product, priced higher and sold by reputation. At the other, production where the model does the work and the price approaches the cost of the API call. The middle, where competent execution used to be enough, is where the quotes disappeared.

For an industry that spent twenty years building a platform around portfolios, the shift is awkward in a specific way. A portfolio shows output. If output is now cheap, a portfolio of output demonstrates less than it used to. Nobody at the conference had a clean answer for what replaces it, which is probably why the theme was a sentence about what not to do.

Related articles