Last week, while obsessing over an ultra-slim serif font in a vintage quote image, I did something many of us have done in the AI age: I asked ChatGPT what font it was. The answer came back quickly — Caslon something. Not wrong. But also, not right.
That’s when things got interesting.
See, most people might’ve shrugged and moved on with a "close enough." But the real joy of working with AI, the part where you actually learn something, is not in the first reply. It’s in the back-and-forth, the unexpected detours, the arguments over the “z” terminal, and eventually, the discovery of something you didn’t even know you were looking for.
This is a short story about that moment. About the value of the human-in-the-loop. About taste, intuition, niche expertise, and the importance of knowing when to deliberately stray from the original prompt.
Expertise in the Long Tail
It started with a pretty straightforward task: font identification. I had a quote image. The font was serifed, bold, compact. Looked kind of like Caslon, but slimmer. I uploaded it, and the machine went to work. ITC Caslon 224? Clarendon? Lubalin Graph?
Wrong, wrong, and interesting but still wrong.
But here's the thing: these answers weren’t random. They were reasonable guesses, based on high-probability matches in the design ecosystem. But I wasn’t looking for high probability — I was hunting down aesthetic truth. I knew this was a boutique, maybe even custom font. I knew the "z" terminal mattered. I knew this wasn’t just any bold serif. This was a feeling, not a classification.
And that’s where human niche expertise comes in. The machine had good priors, but not my trained eye. I’ve been knee-deep in type catalogs and indie foundries. I can smell the difference between Garamond and Plantin in two glyphs or less. This was about long-tail intuition, the kind of knowledge that isn’t well-represented in tokenized training corpora.
AI as a Collaborator, Not an Oracle
I didn’t discard the AI’s input. On the contrary, each suggestion nudged me closer. Clarendon was too slabby. Lubalin was too geometric. And then, a spark: I remembered Klim Type Foundry. That led me to Martina Plantijn.
It was almost a match. But not quite.
And here’s where the magic happened: instead of just asking "is it this?" again and again, we started talking about why it wasn’t. About spacing. About the "z". About editorial fonts used by Monocle Magazine. Suddenly, we weren’t hunting fonts — we were mapping the ecosystem of taste in modern publishing.
We forked the prompt. We let go of the target and followed the thread. And we ended up in a far more valuable place: a list of top independent foundries, an understanding of the editorial serif aesthetic, and a workflow I can reuse.
This is AI at its best: not solving problems, but helping us navigate better ones.
Knowing When to Divert
This is the meta-skill of the AI era: knowing when to let go of the original question.
I started by trying to identify a font. I ended up learning more about:
- Why certain editorial fonts signal modernity
- Which foundries shape the aesthetics of The Gentlewoman, Monocle, and Kinfolk
- And how to spot a boutique cut that never made it to Google Fonts
That’s bonus round learning.
You can think of it like jazz. The prompt is the sheet music, but the real story is in the improvisation. And great AI use is a kind of duet — one where you're not afraid to riff.
The Human Workflow Advantage
So what did I, the human, bring to this AI interaction?
- Long-tail expertise: the kind that doesn't show up on the first page of Google
- Intuitive feedback: “this feels off because the z is too curved”
- Taste: trained over years of visual culture exposure
- Willingness to wander: the most underappreciated research skill in the world
And what did the AI bring?
- Pattern-matching horsepower
- Knowledge scaffolding: links, references, context
- Speed: I could iterate through entire font families in seconds
Bonus: Foundries to Watch
If you’ve made it this far, here’s your reward: a shortlist of foundries that define modern editorial design.
- Klim Type Foundry – Home of Tiempos, Founders Grotesk, and yes, Martina Plantijn
- Commercial Type – Canela, Portrait, Atlas Grotesk
- Grilli Type – GT America, GT Super
- Schick Toikka – Noe Display, Lyyra
- Swiss Typefaces – Suisse Int’l, NewParis
- Lineto – Brown, Circular
- Optimo – The classic Granger, Apax
- Typotheque – Fedra, Lava, November
Takeaways for Your AI Workflow
- Don’t settle for first hits. High-probability isn’t high-fidelity.
- Treat AI as a research partner. Not an oracle, not a servant.
- Let your intuition lead. If something feels off, follow that.
- Divert freely. Side quests are where the real insights live.
Further Surfing 🏄♀️
- Typography as a Window to Culture – AIGA Eye on Design