We invented an avatar that does not exist yet. If you type for a living — and everyone types for a living now — it might be you.
Marketing people build imaginary customers and give them names. We did it too, and we are telling you, which marketing people do not do. Ours is called the Prompt Artist, and we are fairly sure he does not exist — yet. What doesn't exist is perhaps futurum est: still to come. Digital nomads did not exist until the phrase did. Then thousands of people read it and recognized themselves in advance.
The name is a little ironic, and the irony is the point. Anyone doing office work in 2026 will get it. Your output increasingly IS your typing: prompts, messages, specs, replies to a machine that answers faster than you can ask. The people winning with AI are not the ones with the cleverest tricks. They are the ones whose fingers keep up with their heads.
Typing tests measure words per minute, which quietly assumes continuous typing — a player piano unrolling. Nobody who thinks for a living types like that. Real typing ebbs and flows the way thinking does: you compose a phrase, hold it, render it in one burst, and rest while you read what came back. Composing prompts is closer to music than to transcription — phrases and rests, not a metronome.
So the honest metric is not words per minute. It is completed units per session: how many times in 25 minutes a whole thought became whole text. The burst matters, not the average. What you want to train is the attack — the speed at which a sentence you are holding in your head becomes a sentence on the screen, entire, before it decays.
Here is what a dictation rep actually is: you hear a full sentence of real human speech, hold it, and type it — the whole unit, from memory, against a clock made of your own forgetting. Then a character-level diff tells you exactly what your ears and fingers dropped. That is the perceive → hold → burst → judged cycle, stripped of everything else, with unlimited real material behind it.
The founder never took typing seriously — self-taught fingers, a natural and probably suboptimal method. It was good enough for years of Romanian dictation, and somewhere in those years the thing happened that this post is about: once you recognize words as you hear them, your fingers move very quickly to get the whole units. Recognition, not finger technique, was the speed. The unit arrives whole and the hands just print it.
Now run that inward. Your inner monologue is a voice too. If it can whisper a full sentence to you — silently, but whole, the way a native speaker's sentence arrives whole — then you type from having already heard the entire thing mentally. That is what composition is: dictation from your own inner voice. Train the outer circuit and the inner one comes along, because they are the same circuit wearing different input.
Honesty clause. If you are, say, a non-native solopreneur deciding where 25 minutes goes: many things would benefit you, and your own business is probably the right pick. You could also just walk for 25 minutes, and some days the walk wins. Entrepreneurs invent rituals the way other people buy gadgets, and this could become one more ritual. We would rather say that out loud than pretend every minute here compounds.
What we will claim is narrow: this trains a circuit you use all day, it happens to install a language while doing it, and it pays per rep rather than someday. Either the appetites are all the same and we are shouting into the same crowded market as everyone else — or you build something that creates a new appetite. This page is us betting a name on the second.
Related: Understanding Is Compression · Their Training Data Is Your Training Data