Your UX process is a 1960s ad agency relic, and AI just exposed it你的 UX 流程其實是廣告黃金年代的遺產，AI 只是把它戳破了

Most in-house UX teams did not design their own operating model. They borrowed one, wholesale, from advertising agencies: pitch meetings, client relationship management, billable deliverables, a star creative director culture, and a revolving door of talent. Patrick Neeman's piece on UX Collective calls this out directly. It is a Mad Men era inheritance, and it made sense when it was invented. Agencies serve external clients and bill by the project, so optimizing for the pitch was optimizing for the thing that actually paid the bills.

The problem is what happened when this model got transplanted into internal product teams. The incentive to make the pitch deck look good never went away. The incentive to make the thing that ships actually work, for users, after launch, was never really built in. Neeman points to McKinsey's business value of design research as the counterfactual: companies that treat design as something measured like revenue, integrated across functions, and iterated on with real users, grow revenue 32 percentage points faster over five years than their peers. That is not a case for better decks. It is a case for a different operating system entirely.

Four inherited habits, four failure modes

Neeman breaks the inheritance into four specific patterns.

Hype over practice. Teams optimize for how good the pitch looks, not what happens after launch. He calls the output "tool-shaped objects," artifacts that look like work and feel like progress but do not actually move anything.

Clients over users. Borrowing Marty Cagan's missionaries versus mercenaries framing, agencies are structurally mercenary: accountable to whoever is paying, not to the people using the product. That reflex got copied straight into internal teams that should be accountable to users instead.

Deliverables over outcomes. John Cutler's "feature factory" idea shows up here almost word for word: deliverables are billable, outcomes are not easily invoiced, so teams get rewarded for producing something that looks finished rather than proving it worked.

Auteurs over systems. Brad Frost's Atomic Design gets cited as the counter-model. Systematized components produce more consistent quality over time than any individual star designer's taste, no matter how good that taste is.

flowchart TD
    A[1960s Ad Agency Culture] --> B[Pitch Culture / Client-Centric / Billable Deliverables]
    B --> C[Inherited Wholesale by Internal UX Teams]
    C --> D1[Hype Over Practice]
    C --> D2[Clients Over Users]
    C --> D3[Deliverables Over Outcomes]
    C --> D4[Auteurs Over Systems]
    D1 --> E[AI Collapses Cost of Polished Output]
    D2 --> E
    D3 --> E
    D4 --> E
    E --> F[Gap Between Looks-Finished and Actually-Works Becomes Impossible to Hide]

Why AI makes this urgent now, not later

The agency model survived for decades because producing a polished deliverable was expensive and slow, so looking finished was a reasonably good proxy for being finished. AI generation breaks that proxy. A believable mockup, a working prototype, even a passable first draft of copy now takes minutes instead of weeks. The scarce resource was never really "can someone make this look good." It was always "does this actually work for the person using it," and that question was easy to defer when producing the alternative was so slow that nobody noticed the gap.

Once output is cheap, the gap between a good pitch and a good outcome stops being invisible. Teams that spent a decade optimizing for the wrong half of that equation are the ones who will feel this first.

What to actually do about it

Three things worth taking seriously if you are building an AI-native product or design org.

First, stop measuring the deliverable. Measure retention, conversion, task completion, whatever the actual outcome is for your product, and hold design work to that bar the same way you would hold a growth experiment to it.

Second, invest in the system, not the auteur. A shared component library and a living knowledge base compound over time. An individual designer's taste, however sharp, does not transfer when they leave, and AI-assisted generation has already made individual craft less scarce than it used to be. The moat is increasingly the system, not the person.

Third, make user contact a habit, not a project phase. Teams with high turnover and fast iteration cycles, which describes most AI-native startups, are exactly the teams most prone to losing organizational memory between cycles. A standing weekly user conversation is cheap insurance against re-learning the same lesson every quarter.