Marcel Proust wrote In Search of Lost Time by hand. 13 years. 7 volumes. 1,500,000 words.
I'm dictating 1,500,000 prompts to Claude through Genspark Speakly. Each spoken word produces ~15 words of code. ~2 lines. Target output: 3,200,000 lines of code.
He had a pen. I have a microphone.
→ Live tracker: https://grimonprezedition-art.github.io/microphone-vs-pen/
| Volume | Words | Year |
|---|---|---|
| Du côté de chez Swann | 224,000 | 1913 |
| À l'ombre des jeunes filles en fleurs | 213,000 | 1919 |
| Le côté de Guermantes | 311,000 | 1920–21 |
| Sodome et Gomorrhe | 240,000 | 1921–22 |
| La Prisonnière | 142,000 | 1923 † |
| Albertine disparue | 119,000 | 1925 † |
| Le Temps retrouvé | 213,000 | 1927 † |
| Total | ~1,500,000 | 1909–1922 |
| François | Proust | |
|---|---|---|
| Method | Voice + AI | Hand + pen |
| Daily output | ~8,100 words | ~316 words |
| Ratio | ×26 faster | reference |
Proust took 13 years because writing is slow. Voice + AI is a phase shift, not an improvement. The bottleneck moved from hand to mouth, then from mouth to thinking.
This tracker is one runner posting their splits.
- Single-file HTML/CSS/JS, no dependencies
- Genspark Speakly for dictation
- Claude (Opus 4.7 / Sonnet 4.6) for code generation
localStoragefor the running counter- No backend, no tracking, no analytics
MIT.