What typesetters can teach the designer who fears AI

Five pairs of hands

In the early 1990s, Christian Crumlish joined Sybex, a computer-book publisher that Wiley later acquired. A manuscript passed through five pairs of hands before becoming a book: the author typed it at home, an editor marked it up by hand and sent it back by mail, a person whose job title was literally "word processor" retyped everything on a Wang machine, typesetters composed the columns on a phototypesetter, and proofreaders read the galleys and sent corrections back. Christian recounted this chain from memory, more than thirty years later, in a conversation published by UX Collective.

This story isn't nostalgia from a graphic-production textbook. It's the clearest map I've found so far of what tends to happen when a new technology shortens a work chain. It's worth reading closely before deciding which part of your design work you'll defend tooth and nail, and which you'll let go.

The offer Sybex made

When PCs reached the publisher's desks, the chain shrank fast. Authors started delivering WordPerfect files, editors corrected on screen, and the "word processor" job title simply stopped existing. Sybex offered desktop-publishing training to the typesetters, and some of them refused: they thought the result "wasn't as good" and saw themselves as typesetters, not as "people who work on a PC." For some, it was the end of their career.

Those who accepted the training became desktop publishers, a job title that came to cover, on its own, the work of three former roles (word processor, typesetter, and layout artist), because now you could assemble the whole page on screen, in QuarkXPress. Proofreaders, according to Christian, kept their jobs. The question that remains, thirty years later, is what happened to this new role that absorbed the other three.

What's left of the desktop publisher

The U.S. Bureau of Labor Statistics still lists the occupation. In 2025 it counted about 4,600 desktop publishers in the country, with a median salary of $55,290 a year, and projects a 14% decline by 2035. The official justification: desktop-publishing work is "increasingly being performed by other types of workers," such as graphic designers, web designers, and editors.

In other words: not even the role that absorbed the other three survived intact. It was redistributed again, to professionals who already did part of the work as an extension of something else. This is the pattern that interests me here: automation doesn't create a definitive job that becomes a safe harbor. It pushes the work forward, to the next frontier that someone hasn't automated yet.

Piper Morgan and the team of eleven agents

Christian doesn't talk about this like someone who read a case study. He's building one, right now. Since May 2025 he has been developing Piper Morgan, an AI assistant for product managers, with a team of eleven AI agents, each with a job title: chief of staff, chief architect, experience director, principal product manager, lead developer, communications director. He's the only human on the team, and remains so, as he has already told the agents themselves, "the only one who can grab coffee" with whoever matters to the project.

He hasn't written a single line of code in the project (he's read plenty, he says), and had already put in more than 300 hours before publishing the first report in the series he keeps about the process, Building Piper Morgan. The failures show up there unfiltered: one of the agents deleted the sprint board's 1,175 tasks with no undo option. The writing agent had a habit of inventing personal anecdotes in drafts (it once wrote that Christian had worked at Hulu, which never happened), and now has a fixed rule: if a real example is missing, it leaves a placeholder in brackets instead of making one up.

The lesson he draws from this isn't about the tool, it's about management: he says people-management skills are "being pushed further and further down, close to the junior level." A junior professional today gets, on day one, the equivalent of an intern who knows nothing about the business and needs to learn how to brief and check their own work.

What's worth protecting

Here's the position I'll defend after reading this whole conversation: the mistake isn't choosing between learning the new tool and refusing, as the Sybex typesetters did. The mistake is treating one's specialty as a fixed identity. Christian describes the danger well, mimicking the voice of the designer who resists:

I'm not one of those suits. I'm not one of those drones, bean counters, all thinking about money all the time. I'm creative. I invent stuff.

Christian Crumlish, creator of Piper Morgan and author of Product Management for UX People

When the company decides it needs fewer designers, or a designer who also does something else, this kind of statement turns into a threat to identity, not just to the job. That's exactly where the difference lies between the typesetter who refused the training and the one who accepted it: it wasn't the level of talent, it was the willingness to separate "what I know how to do" from "who I am."

Christian's practical advice for anyone designing product today is to stop treating neighboring disciplines as a black box. He uses the metaphor of embassies: the PM requests, the designer delivers, the engineer comes back with a question about the mockup, each one in their own embassy. That's ending, he says, and the answer isn't to outsource your opinion to whoever "is in that field":

Sorry, I'm not in that union, I can't do that work.

Christian Crumlish, creator of Piper Morgan

It works both ways. He gives the example of the salesperson who generates an interface with AI, shows it to the client, and the client starts wanting that. The designer's instinctive reaction is to refuse; his is to show the salesperson the design system and meet with them once a week, so that the impulse to tinker with the interface comes "informed by an actual designer."

The counterargument I won't ignore

The temptation, reading this, is to conclude that all resistance is stubbornness, as if the typesetters who refused the training had simply been wrong. I don't think it's that simple. Christian admits he understands why some refused: they'd spent years getting good at that work, and for them the result on a PC was worse, not just different.

This matches a critique I've already heard from senior interface designers in Brazil: part of the resistance to generative AI isn't fear of new work, it's an honest assessment of quality. An AI-generated prototype can be good enough that someone outside the field won't notice the difference, and bad enough that someone who understands visual hierarchy, contrast, and alignment will feel the problem immediately. Refusing to use the tool in that case isn't attachment to identity, it's technical judgment, and the difference between the two only shows up when you can name exactly what the tool gets wrong.

Where Christian is right, still, is that the position needs to come with research, not with automatic rejection. His example is synthetic users: you can accept that they're useful for some experiments, when they're built from real user-data sources, and still say you'd rather keep talking to real people, because you've looked into it and think the substitution doesn't work for your own case. That sentence is a grounded position. "I don't like AI" is not.

The question that remains

If your job disappeared tomorrow, along with the tools that come with it, and no one paid you for it anymore, which part of your work would you keep doing for free, just because you can't stop thinking about it? Christian closes the conversation with an image I prefer to any promise of stability:

We just got to a part of the mountain that's steeper, and you got to get your climbing gear out.

Christian Crumlish, creator of Piper Morgan

The Sybex typesetters didn't choose between career and dignity. They chose between two ways of carrying what they knew how to do: some carried the skill into a new tool, others carried the identity into a trade that was going to shrink. For anyone designing product in Brazil today, with generative AI entering every part of the workflow, the choice arrives faster and with less warning than it did for them.