Everyone gets a scribe
I watched a talk by Geoffrey Litt called "Dynamic Documents as Personal Software" given at the Recurse Center. Below is a short summary, and then a comparison I could not stop thinking about while watching it.
Spoilers ahead. The talk is worth the hour and this post doesn't do it justice.
The talk
Litt opens with a split. Programmers can get the machine to do whatever they want, everyone else gets whatever the programmers hand them. Apps embody that, because an app is a box of data schema, computation and views someone else packed. Maps knows locations, the weather app knows forecasts, your notes app knows text, and none of them know about each other. So when you plan a trip you become the glue, copying between apps and running half the computation in your head.
He questions the split almost immediately, though. Spreadsheets, he thinks, are a classic case of people who do not consider themselves programmers doing a lot of computational thinking.
His answer is to start from the document instead of the app, because people already write notes and a note is easy to begin. You just type. So what if the note could compute?
He shows eight years of chasing that. Margin Notes put the values from real runs beside the Ruby code that produced them, on the theory that reading code is hard because you can see the code but never the data. Embark turned a trip plan into a live outline: mention a place and the map data comes with it, add a map view and everything you mentioned appears on it, type a formula and you get the weather for the right town on the right day because it can read the text around it.
The Notion work is where it gets concrete, since that is his job now. He asked an agent to redraw a small 2D zen garden game into isometric perspective, and instead of a diff he had it write the change up as a document. It opens by explaining the game engine and which corner the coordinate system starts from, teaches isometric projection with a slider you drag to watch the tilt change, then walks the diff in stages with an explanation at each step, and ends with a quiz. He will not merge code or send it for review until he has passed his own quiz.
The detail he stresses is that it is not one big HTML file. Only the figures are embeds, and everything around them is ordinary Notion text he can edit, comment on and hand to a colleague. The document stays a document.
He is honest about the weak spot. The hard problem is getting someone from zero to their first useful thing. Templates are the usual answer and he thinks they are a bad one, because a template whose internals you cannot pry open is just a badly made app. AI does not fix it either, since if it does the work and you do not know how, you have not gained much. What he wants to copy is HyperCard and the spreadsheet, both programmable substrates disguised as documents. HyperCard let you in to browse and paint, and somewhere along the way you were authoring, then programming. A spreadsheet lets you type some numbers and already have something useful, then one day you discover SUM, then another formula. The climb is gentle enough that you never notice you started programming.
The comparison
Here is what kept nagging at me. This is the history of literacy, all over again.
A tale of two classes
Writing started as accounting. Temple inventories, tax records, debts, run by a small group of paid professionals. The scripts were hard. Cuneiform had hundreds of signs and took years to learn. That difficulty was not a problem the scribes were trying to fix, it was the reason they had jobs. The pain of a modern toolchain is not an accident anybody is in a rush to remove either.
It was never just literate versus illiterate, though. Egypt ran three scripts at once, from formal hieroglyphs down to everyday demotic. China had classical Chinese for officials who passed the exams and vernacular for everyone else. Latin sat on top of local languages in Europe. Plenty of people could read but not write, and plenty could read their own language but not the prestige one.
Which is why I think the programmer and user split is a story we tell rather than a fact about people. Ordinary people compute all the time, they just do not use the word. A shop owner running her stock and margins in a spreadsheet is doing real computational thinking. So is anyone who has ever written a filter rule for their inbox or worked out a schedule in a table. What separates them from the programmers is not ability. It is which register they were given access to.
Accessible scripts and languages win
And the scripts that spread were not the better ones, they were the more accessible ones. The alphabet has about twenty six symbols and is less precise than cuneiform or ideograms for some purposes, but a child could learn it in weeks. The same thing happened in the ports of the region where i live without anyone planning it. Bazaar Malay was the stripped down trading version of the language, affixes filed off, vocabulary picked up from Hokkien, Tamil and Portuguese as needed. It was nobody's mother tongue and nobody's idea of good Malay, but two strangers could close a deal in it, and it spread across the archipelago while court Malay stayed in court. Spreadsheets are the same trade. They are not "proper" programming, which is exactly why so many people use them.
The other thing the history tells you is that reading has always run far ahead of writing. Far more people consume than compose, in every literate society we know of, and that gap is where every literacy project gets stuck. For the dynamic document model, choosing the note is choosing the vernacular, because it is where people already are.
Accessibility doesn't guarantee adoption
Korea should bother anyone who thinks good design wins on its own. Hangul was made in the 1440s and is about as well built as a writing system gets, with letter shapes based on the shape of your mouth as you make the sound, and its preface bragged that a clever man could learn it in a morning. The scholar class wanted nothing to do with it, because classical Chinese was most of the point of being a scholar, so they left the vernacular script to women and commoners. One king banned it. It took four and a half centuries, a nationalist movement, and eventually a state that mandated it before it became the normal way to write Korean.
What moved the numbers, there and everywhere else, was compulsory schooling, because factories and armies needed people who could read instructions. Notation design has never democratized anything on its own. Litt more or less concedes the software version of this when he says the learning has to happen in the user's head and that is always hard, which is an education problem, and no product team can ship it.
Which puts his project in an awkward spot. By his own account the hardest problem at Notion is that people cannot learn the thing, and the two answers on offer, templates and AI, both work by letting you skip the learning. History's answer was compulsory schooling, slow and institutional and measured in generations. No company can run that play. Spreadsheets got there eventually, but it took about forty years, and it happened in accounting departments and evening classes rather than at Microsoft.
The educated elite did not disappear either, it moved. Once most people could read, the difficult language relocated into law, medicine, finance, engineering, and eventually code. Expect the same here. If everyone can build a plant watering tracker by asking for one, the valuable skill becomes something else, probably whoever controls the platform and the model. Which is worth saying plainly about Notion, too: it is a company selling to other companies, and the product is not open source. Litt says as much when someone asks. The printing press was a business as well, and who owns the press has always mattered as much as who can read.
Test of comprehension
Litt puts a quiz at the bottom of his code explainers because, as Andy Matuschak puts it, books are "surprisingly bad at conveying knowledge" and readers don't even realize it. Plato made almost the same complaint about writing two and a half thousand years ago: it gives you the appearance of wisdom without the real thing, because you can have the text without having the understanding. Every literacy system eventually invents a test, because looking like you understood something is easy and outruns actually understanding it.
For most of history, if you needed something written, you hired a scribe. That was never a step toward literacy, it was the alternative to it, and it was a rich man's alternative. What is new is not that the option exists, it is that everyone can afford it now. Anyone can vibe-code. Litt's demonstrated quiz is a real defense against shipping code you never understood, but it is one man's discipline which only protects the person who already wanted to understand.