Architecture has an old concept called parti — the ability to shrink a whole building onto a single napkin. The AI era needs the same training. Today we unpack this principle through a tool called Napkin AI.
Most people who sit in front of AI to make a presentation have the same moment. The text is already piled up — a ten-page explanation from ChatGPT, meeting notes, lecture memos. But when it's time to make a single slide, your hand stops. Why? The writing is plenty, but you were never trained in how to shrink it into one picture.
This essay is about that shrinking power. If you've never used an AI tool before, you can still follow along — we'll go slowly. Today's example happens to be a service called Napkin AI, but the principle applies to any visualization tool, in any era. Five years from now, even when "Napkin" is gone, the spine of this essay still holds.
Let's start with the principle. Architecture has a concept called parti, borrowed from French. It means shrinking a whole building down to a napkin-sized sketch. A twenty-story hospital, a three-building museum campus — in the end, all of it has to fit inside a sketch smaller than your palm. If it doesn't fit, the architect hasn't understood it yet.
This isn't a new idea. Louis Kahn and Frank Lloyd Wright would ask for a napkin at restaurants and draw the core of a project on it. That single napkin became the seed of a ten-year design. In other words — the depth of your understanding shows up as the density of your compression. Knowing a lot isn't understanding. Fitting it into one page is.
And yet this common sense disappears the moment we face AI. Why? Because ChatGPT spits out a ten-page document in a minute, and we skip the step of shrinking it back down. The more text floods in, the foggier our own heads get.
Let's look at one tool to make the principle concrete. Napkin AI — the name itself inherits the parti philosophy. It's a US web service where you paste text and the AI reads the core and draws a diagram.
I tried it myself. I asked ChatGPT "how do you visualize text?" and got a list of seven methods. I pasted that into Napkin and hit the lightning icon on the left. Three seconds later, five diagram options popped up. I picked a style, and it was cleaned up to A4. I could export as PNG, SVG, PDF, or share by URL. The Pro plan was completely free during the beta.
Canva has something similar. So do Microsoft and Claude. The core feature is the same — text → automatic diagram. No matter what company ships what name next, this feature will repeat. The principle demands it.
To make it easier, picture a kitchen. A cooking YouTuber posts a twenty-minute video. The script is 40 pages. But what the home cook actually needs is a single sticky note taped to the fridge. Five ingredients, four steps, one timing — that's it. Shrinking 40 pages into one sticky note is the real cooking knowledge.
Napkin AI does exactly that. It shrinks a ten-page ChatGPT answer down to fridge-note size.
Reading long writing as long writing is like hauling the 40-page script into the kitchen — greasy hands, scattered eyes.
I got curious about how dense a napkin sketch actually is. The video shows a collection of 17 architect parti sketches. Each about 10 cm across, holding buildings dozens of stories tall. Measured by surface, the compression is under 1 to 10,000 against the real thing.
Now the other side. The AI-generated text we receive every day is 2,000 to 5,000 characters at a time. Many people paste it straight onto slides. That's over 300 characters per slide. The audience's eyes close within three seconds.
Here's the first aha.
The problem isn't that AI writes too long. It's that we are not training ourselves to shrink it.
In an era where AI produces text in one minute, the human job shifts — from "producing lots of text" to "shrinking text to one page." This is parti sense. Not an architect's skill, but basic literacy for everyone in the AI era.
So, how do you actually train? Just one question to wear on your body.
"If I shrink this writing to an A4 diagram, what's left?"
Ask it once a day. Got a 20-page meeting doc — what's left if it's one page? Read a paper — what's left if it's one page? Watched a YouTube video — what's left if it's one page? Three months of this and your head changes. When you read sentences, a diagram is already forming in your mind.
Let's do one concretely. Task: shrink a one-hour team meeting's 20 pages into one parti.
Step 1. Paste the full meeting doc into Napkin AI.
Select the whole meeting text and click the lightning icon on the left (Auto Spark).
Step 2. Pick one of 5~7 diagram options.
The AI usually offers 5~7 candidates — hierarchy, flow, comparison, circular, timeline. For meeting notes, flow (decision path) or hierarchy (topic → action items) usually fits best.
Step 3. Polish by hand.
Fix just three things on the AI draft. (1) Owner (subject) of each action item (2) Deadline (date) (3) Remove duplicates. Five minutes, done. A 20-page meeting becomes an A4 parti.
When my team actually used this, meeting follow-up time dropped from 30 minutes to 5 minutes. That's one-sixth. Because we no longer had to re-read 20 pages.
Now, how do you actually start. No complicated setup.
1. Go to napkin.ai (log in with Google)
2. New Napkin → Blank Napkin
3. Paste your text with Ctrl/Cmd + V
4. Click the lightning icon (Auto Spark) on the left
5. Pick a diagram you like → Share or Download PDF
Five steps. From first visit to finished output, three minutes. Try it for a week. Any long writing you receive, run it through Napkin once. It feels weird at first, but within days your reading eye changes to "how would this shrink to one page?"
One honest limit. AI-made diagrams are good for "information-organizing" level compression. The artistic insight of Louis Kahn's parti is still a human job — AI can't imitate that depth yet. But AI handles 90% of the organizing. Spend the saved time on the 10% that needs real insight.
Let's wrap up.
Architects compress a whole building into a single napkin. That act is called parti. I explained this principle through Napkin AI, but the same applies to any visualization tool, any future AI. Specific names change. The compression principle repeats.
Just one question to wear on your body — "If I shrink this writing to an A4 diagram, what's left?" This single question changes how you read, listen, and think. When long writing lands, run it through a tool like Napkin. Five minutes, and you have a parti. Read the writing again with the parti in hand, and the structure becomes visible.
The people who last won't be those who produce lots of text, but those who can shrink text to one page. This sense doesn't come from AI. It comes from an architectural tradition — parti. Five years from now, if Napkin AI is gone, the principle you learned today still works. Tools change. Compression doesn't.
Text. One page. What remains.