I. An equation the industry stopped questioning thirty years ago
This isn't about any particular firm.
It's about an equation. One that held up the branding industry for three decades, that almost everyone still assumes is true, and that has quietly stopped working:
Revenue = headcount × hours × rate × utilization
Design didn't invent this. It came from law firms, from accounting firms, from the big agency holding companies. Every professional service business of the industrial era ran the same play: slice knowledge into hours, pack hours into people, stack people into scale.
In that equation, headcount is the only variable you can multiply. Want to earn more, hire more. So scale became the industry's common language. Headcount on the tender document. Headcount on the homepage. Headcount in the founder's introduction at every conference.
Scale used to be proof of capability. Then it became a source of comfort. Now it's an exposure.
Because the "hours" term is being pulled out from underneath it.
II. Here's the part most people miss: those headcounts aren't buying capacity
If a hundred people correspond to a hundred people's worth of work, then scale is rational.
The trouble is that in a large share of big design firms, the headcount was never sized to the work in the first place.
It was sized to bid eligibility.
Government contracts, state-owned enterprise procurement, annual framework agreements with large groups. The tender documents state the thresholds outright: number of employees on social insurance, registered capital, certification tier, comparable contract value over the past three years, count of full-time design staff. Miss any of them and you don't get to compete. You don't even get to lose.
Which produces the industry's quietest open secret: you keep a hundred people not because a hundred people are needed to do the work, but because with ninety you can't get in the room.
That sentence deserves a beat, because its implication is larger than it looks:
When a firm's size is dictated by procurement thresholds, its structure stops serving the work and starts serving the bid.
An organization grows into the shape of whatever feeds it. A firm that lives on its work grows the shape of its work. A firm that lives on its credentials grows the shape of its credentials. In the first, page one of the employee handbook is a standard of taste. In the second, it's a process chart. Both can be profitable. They are not the same species.
This also answers a question a lot of people ask quietly: why does the work so often get flatter as firms get bigger?
It isn't that they stopped caring. It's that the organization was never built for that.
This mechanism is most extreme in the Chinese market, where public-sector procurement is enormous and its criteria are unusually explicit. But it exists anywhere large-enterprise RFPs use vendor size as a proxy for reliability. The scale of the distortion varies. The distortion doesn't.
III. Three locks, closing at once
Once you see that, the rest follows.
The cost lock. Payroll, benefits, office, overhead, management salaries. None of it falls just because the pipeline thins. Cost is a straight line, revenue is a wave, and once they cross, the cash on hand decides how many months you have left.
The credential lock. This is the cruel one. Cutting staff reduces cost, but cut below the threshold and you lose bid eligibility, which is the revenue. Stay big and bleed, or shrink and disqualify yourself. That isn't an efficiency problem. It's a business model welded to a licensing requirement.
The path lock. A firm built on major tenders can't turn around and take five-thousand-dollar projects. Not out of pride. The pricing structure, the process, the client relationships, the composition of the staff — none of it supports the move. Turning a large ship requires open water, and the water is going out.
Add the macro layer. In a tightening cycle, brand budgets are always among the first cut, and public-sector brand budgets are exactly what these firms eat.
So the fatal wound isn't AI.
It's high fixed cost paired with revenue concentrated in the client type that contracts first.
AI just made the alternative look absurdly cheap and pressed fast-forward.
One counterweight worth naming, because most people miss it: a deflationary market also makes labor cheaper. Falling hiring costs will keep some of these firms alive longer than the math suggests. So the shakeout probably won't look like sudden death. It will look like slow bleeding — still solvent on paper, already dead in the work.
IV. What AI actually killed
The industry keeps repeating a line: AI is coming for designers' jobs.
I think that's wrong, or at least shallow.
AI didn't kill design. It dismantled the equation design firms lived on, by removing the "hours" term with surgical precision. Hours were the pipe that turned knowledge into revenue. The pipe is gone. The knowledge is fine.
Demand hasn't dropped. If anything it's rising, because the more products converge, the more brand becomes one of the few defensible positions left.
What changed is what clients will pay for.
They used to pay for making it, because making it was hard, slow, and expensive. Now that making it is easy, fast, and cheap, they'll only pay for deciding what to make and being right.
Scarcity migrated:
From being able to make it → to knowing what should be made → to being willing to stake your name on that call.
The whole value chain shifted upward. Execution is trending toward free. Judgment is appreciating. And the layer in the middle — not cheap enough, not sharp enough — collapses first.
There's a second shift almost nobody is pricing in. The form a brand exists in has changed.
Brands used to live on packaging, in stores, inside a three-hundred-page guideline document. They now live in feeds, in recommendation algorithms, and increasingly in search results and answers mediated by AI. A static rulebook is losing its function. What replaces it is consistent behavior across an unbounded set of dynamic surfaces.
Put more bluntly: part of your brand asset now has to be legible to machines. Whether your brand can be found, cited, and accurately restated by an AI system is becoming a real asset class, and almost no one is managing it deliberately.
That isn't a loss for design firms. It's new work. It just demands a different capability than the one the industry spent thirty years building.
V. An uncomfortable sentence about rent
A line that has nothing to do with business models, and may be the most important thing here.
A design firm's creative floor is set by its fixed costs.
The higher the rent and the longer the payroll, the less able you are to turn down a bad client, to say "this direction is wrong" at round seven, to hold the position you actually believe in when the room goes quiet.
People don't go soft. The payroll run on the first of the month makes the decision for them.
The small studios we all admire hold their standards partly on talent and nerve. But a large part of it is simpler: they can afford their nerve.
Which reframes the whole question. Scale isn't only an efficiency problem. It's collateral against your creative freedom. However much you've pledged, that's how low your floor goes.
This, more than any AI argument, is why I think the best firms of the next decade will choose to be small. Shrinking isn't retreat. It's buying back your own judgment.
VI. What the right structure looks like
In one line: small core, strong periphery, high output per head.
Keep the core in the single or low double digits, and make every seat a judgment seat. Not a senior director with a stack of executors — three or four principals, each working with AI. There's a plain arithmetic reason. Communication paths scale as n(n-1)/2, so doubling headcount roughly quadruples internal friction. We tolerated that friction when capacity could only come from bodies. It can't be justified now. Every additional person has to contribute more judgment than they cost in coordination.
Convert execution from fixed cost to variable cost. AI produces the first pass. A standing network of specialists — illustration, motion, packaging engineering, category experts — plugs in when a project needs them and costs nothing when it doesn't.
Get fixed costs low enough to survive a quarter with zero revenue. This is the only real margin of safety. Not backlog, not a warm pipeline. How long you can idle.
Watch one health metric: revenue per head. Traditional firms tend to land somewhere modest (my own estimate from experience, and it varies widely by market). Firms that thrive over the next five years should be running at multiples of it. If you can't, your organization is still inside the old equation.
Build at least one revenue line that isn't project-based. Brand system stewardship, ongoing asset generation, cross-channel consistency governance, annual brand audits. Projects are peaks and troughs; retained service is a heartbeat. It's also the one category where AI earns money for you instead of taking it from you.
Longer term, own something that isn't attached to human hours. Methodology, proprietary data, evaluation frameworks, internal tooling. Otherwise the firm is permanently the shadow of a few people, and when they leave it returns to zero. That's the oldest and most lethal disease in this business.
The hardest position to occupy, in my view, is twenty to eighty people. Too big to afford only excellent people, already carrying management overhead and an office. Not credentialed enough to win the big tenders, not pure enough to be a boutique. It can't climb and it can't retreat.
VII. Staffing: from pyramid to oval, and eventually to inverted pyramid
The old shape was a pyramid — one director, two leads, eight executors. The new shape is an oval, trending toward inverted: a handful of senior people each working with AI, with most of the coordination and production layer compressed out.
Two heads, not one. A design firm needs two kinds of mind, and they very rarely grow inside the same person.
A commercial mind that understands what the client's business is actually trying to solve, that can negotiate, that will take responsibility for outcomes, that doesn't get taken on contracts and cash flow.
A critical mind with real taste, that holds the floor on the work, that can say no at the moment it costs something to say no.
With only the first, the firm drifts into being a marketing shop and the work goes flat. With only the second, the firm makes beautiful things and starves. The tension between them is the only stable structure I've seen work.
Four roles rising in importance:
Strategy and research. The ability to define the problem. This is where AI is weakest, because defining a problem means sitting with a client's silences, hesitations, and contradictions.
Client relationship. Trust doesn't automate. What a client is ultimately buying is the belief that this person won't let them look foolish.
Content and publishing. This is the new business development function, not a marketing afterthought. Consistent, public, retrievable expertise is the cheapest durable channel available today.
AI workflow architecture. Encoding the firm's methodology into reusable systems. Almost nobody staffs this today. In five years it will be standard.
Falling: pure production design, descriptive copy, most project coordination.
Three qualities that define the next generation of designer:
Judgment over craft. The number of people who can make something is exploding. The number willing to decide is not.
Directing AI like a team, not using it like a tool. The difference: with a tool you specify the output. With a team you specify the standard, then review, reject, and send it back. That's a taste-and-management skill, not a prompting skill.
One thing the model can't learn. Deep knowledge of a specific industry, a particular manual sensibility, an unusual body of lived experience. Generalists are being priced flat. Specialists are appreciating.
VIII. The risk nobody wants to discuss: the apprenticeship broke
This section is for the industry rather than any one firm.
The roles AI replaces most completely are junior production roles. Junior production roles are exactly how designers have been made for twenty years. Three hundred icons, eighty layout revisions, a hundred rounds of feedback — the craft and the judgment grew in that grind.
That path is now cut.
Short term, firms save money and output per head looks excellent. Five years from now the industry will collectively discover it has no senior designers left.
So I think firms with any long view should keep hiring juniors, but the apprenticeship has to be rewritten. Stop having them practice production. Have them do research, orchestrate solutions, and curate and critique AI output — and grow by being sent back.
From training the hand to training the eye.
The short-term math on this is negative. But no industry ever maintained its talent supply by doing short-term math.
IX. What we're building at Xinmingshe
Everything above is structural. Here's what we've done with it.
Xinmingshe is not trying to become a hundred-person firm. That's not a limitation. It's the deliberate conclusion of the reasoning above.
Three things we believe:
A firm should be small enough to hold its judgment, and serious enough to carry responsibility. We keep fixed costs low enough that we can decline the wrong client, and low enough that at round seven we can say "go back to version two, that one was right." We don't want a payroll deadline making our aesthetic decisions.
We sell judgment and outcomes, not headcount and deliverable volume. Execution is trending toward free. Resisting that is pointless. So we've moved upstream: we help clients define the problem rather than executing a brief that was wrong before it reached us.
Brands now have to be designed for machines as well as people. More and more first impressions happen inside search results and AI answers. Whether a brand can be found, cited, and restated accurately is a genuine asset, and almost nobody is managing it on purpose. We intend to be three years early on this.
The people we're looking for:
People with judgment who will put their name on it. We don't need executors; AI is faster and cheaper. We need someone who can say "this direction is wrong, and here are the three reasons."
People who direct AI like a colleague. Not tool proficiency. The ability to set a standard, review against it, reject, and keep asking until it reaches the number in your head.
People with one deep specialty. You understand an industry better than the client does, or you have a sensibility that can't be imitated. Generalists get cheaper. Specialists get scarcer.
People willing to publish. In this era, someone who consistently articulates a point of view becomes a source in their own right. We want everyone here compounding value on their own name, not only on the firm's.
Designers who aren't allergic to commerce, and commercial people who respect craft. We want both heads working.
Our vision: to be a small, sharp firm that redefines what a brand advisor is in the AI era. Clients should come to us not for how much we can produce, but for our judgment about what's worth producing at all.
Three goals for the next five years:
- Revenue per head among the highest in the industry — very few people, carrying very heavy decisions.
- A reusable brand methodology and AI workflow so the firm's capability lives in a system rather than in a handful of heads.
- Become the name you can't route around in a small number of verticals we genuinely understand. Through depth, not size.
X. Closing
Thirty years ago, scale was this industry's trophy.
Today it's its most expensive liability.
The demand for design never went anywhere, and the market still needs design firms. It just doesn't need this many of them, and it definitely doesn't need them this large.
What's being eliminated isn't design. It's a way of living: packing design into hours, hours into people, and people into a moat.
The firms that survive will be much smaller, much sharper, and considerably more free.
Because when execution costs nothing, the only thing still worth paying for is someone willing to say: I think it should be this.
Xinmingshe — we're looking for people with judgment.
发表评论