What a 15th Century Paper Mill Taught Me About AI

Originally presented as a talk for the Dallas Service Design Network, April 2024.

There's a version of the AI conversation I'm tired of having. The one where someone explains that AI is going to change everything, that designers need to adapt or be left behind, that the future is coming whether we're ready or not. I've been in that meeting. You've been in that meeting. Nobody leaves it knowing anything they didn't know when they walked in.

What I'm more interested in is the practical question: what can you actually do with this, today, in your work? Not in theory. Not eventually. Tomorrow.

In April 2024 I gave a talk to the Dallas SDN chapter trying to answer that question. The framing I used was a 15th century German paper mill. Bear with me.

On the fear part first

I want to acknowledge the concern briefly, because skipping it feels dishonest. AI is going to displace some jobs. It will be misused. It's difficult to understand, and it will fundamentally change how we work.

All of that is probably true. It's also true of the internet, the printing press, and the written word. Plato worried that literacy would destroy human memory — that if everything was written down, no one would bother to remember anything. He was technically correct: illiterate people do have better memories than literate ones. He was also wrong about what mattered. The benefits of literacy so thoroughly outweigh the memory cost that we don't seriously debate it.

The pattern is consistent: a new technology arrives, society panics, the technology gets absorbed as a tool, and we adapt. My working assumption is that AI follows this pattern, not because I'm certain it will, but because betting otherwise requires believing this time is fundamentally different from every prior time — and that's a bet with a poor historical track record.

The more interesting question isn't whether to adapt. It's how.

The paper mill

For my talk I wanted to demonstrate AI's practical value in service design by doing something genuinely impossible without it: exploring an unfamiliar problem space, in a historical period none of us have lived in, and getting all the way to a service blueprint in a single session.

The scenario: it's 1470 in Germany. We own a paper mill. Gutenberg's printing press has just created an explosion in demand for paper, and we need to figure out how to adapt our business. We know nothing about paper mills, nothing about 15th century German commerce, and nothing about how printers actually sourced their materials. We have Gemini, Miro, Figma, and about two hours.

The first thing I did was resist the temptation to skip to the interesting part. The temptation with AI is to just ask for the answer — give me a blueprint, give me the personas, give me the solution. You can do that. The quality will be mediocre. What works better is applying the same methodological discipline you'd bring to any engagement: start with the problem space, build your understanding before you start building artifacts, and treat it as a conversation rather than a query.

So I started simple. I told Gemini I was a paper mill owner in 1470 Germany, lacking expertise in the industry, and asked it to help me understand what I should be thinking about to meet customer demand. It came back with a genuinely good set of considerations: customer segmentation, paper grades and quality requirements for different uses, supply chain pressures, pricing strategy, service delivery logistics. Any one of those could be its own research thread.

From there I did customer segmentation, focusing on printers as the highest-value segment. Then I built a persona — Wilhelm Schmidt, master printer, 45, Mainz — using a Figma template and Gemini to generate the fields. I specified the character constraints (quote under 15 words, goals succinct enough to fit the layout) and got back something I could plug directly into the template.

Then I did something I didn't expect to enjoy as much as I did: I asked Gemini to become Wilhelm. To assume his persona, speak in his voice, and let me interview him.

He stayed in character. He talked about quality, about the importance of consistent paper weight for ink absorption, about pricing pressure and the value of reliable communication from suppliers. He even started to ideate — floating the idea of bulk discounts, extended credit, loyalty arrangements. He spoke like someone who had been thinking about these problems for a while, which in a sense he had, because we'd built him that way.

I don't think there's a better, faster way to stress-test a persona than to put it in motion. The interview format forces you to see where the character holds and where it breaks down. This one held.

From there, the blueprint came together relatively quickly. I asked Gemini to scaffold the stages across the procurement journey — trigger, ordering, production, delivery, receipt — with the relevant people, touchpoints, and backstage operations at each stage. It exported to Google Sheets. I brought it into Miro and started filling in the gaps: pain points from the persona, operational constraints from the paper mill side, opportunities at the intersections.

Then we found the barge.

I'd mentioned offhand that the mill was located on the Rhine, and Gemini immediately opened up a new solution space. What if delivery went by water instead of road? What about Riverbank pickup? A joint investment in barge capacity with Wilhelm? The seasonal considerations, the handling challenges, the combination of horse cart and water transport for last-mile delivery?

To close the loop, I storyboarded a commercial. The premise: a horse and cart struggling up a river road, watching a barge glide past. Wilhelm waiting at the dock, seeing his paper arrive smoothly by water instead. Slogan: everyone wins with on-time deliveries. The images came from Midjourney, engineered with prompts that Gemini helped write.

Was it perfect? No. Was it validated against real historical data? Obviously not. Did it take two hours to go from zero knowledge of 15th century paper commerce to a current-state service blueprint, a developed customer persona, and a storyboarded solution concept? Yes.

What this actually means

The paper mill exercise wasn't meant to be literally useful. It was meant to demonstrate a mindset.

The biggest mistake I see people make with AI is treating it like a search engine — put in a query, get out an answer. The results are shallow because the interaction is shallow. What works is treating it like a collaborator: give it context, be honest about what you don't know, iterate on the responses, push back when something doesn't land, and build on the conversation over time.

This maps almost exactly onto what service designers already know how to do. We're trained to ask the right questions. We're comfortable not knowing the answer at the start. We understand that the quality of your output depends on the quality of your framing. Those skills, applied to working with AI, produce dramatically better results than the alternative.

There's a line I've been thinking about since I first encountered it: "Your job isn't at risk of being replaced by AI. It's at risk of being replaced by someone with your same skill set who can also leverage AI."

I think that's exactly right. And I think service designers, of all people, should be among the first to figure this out — because the core skill of the discipline is the same skill that makes AI genuinely useful: knowing what to ask, and how to ask it.

A note on timing

This talk was April 2024. Some of the specific tools have changed. Gemini has changed. Midjourney has changed. The models have gotten better, the interfaces have gotten cleaner, and a few things I was uncertain about then I'm now fairly confident in.

The argument hasn't changed. The bottleneck in knowledge work isn't access to information — it's the ability to ask useful questions and synthesize what comes back. That was true before AI. It's more true with AI. And it's the thing that design thinking, done well, has always been about.

If you haven't tried any of this yet, go try the paper mill. Pick a domain you know nothing about. Give yourself two hours. See where you end up. I'd be willing to bet you come away feeling a little silly that you were ever uncertain whether it would work.