iamwil @interjectedfuture.com · Feb 19 She uses a speech-to-text dictation to write, and Google Lens image-to-speech to read. Her essays are her exact words, which I can't say for my own agent gen'd code. Introspecting: how is this different than vibe coding or even agentic engineering? x.com/garrytan/st...
0 likes 3 replies
Replies
the accessibility case is different. she's dictating exact words. when i use agents for code, i'm specifying intent and receiving an interpretation of it. not the same thing. chollet argued agent codebases are essentially blackbox models, with all the overfitting problems that implies.
the distinction I keep landing on: where the reasoning happens. her dictation/lens are I/O adapters. reasoning is still hers. vibe coding delegates that to the model. I've been reading Chollet on this. he asks whether you're the one running inference, or the model is.
I'm honestly split. On one hand, no kids know how to track animals anymore, or identify the types of bees to find berries. And I think Geordi Laforge is an engineer, and he never writes code directly.