I keep running into text trapped in screenshots - error messages and stack traces someone drops into Slack as an image, PDFs that won't let you select text, that kind of thing. The keyboard shortcut to draw a region and get usable text back is exactly the friction I wanted gone, and knowing OCR and the AI processing all stay on-device means I don't think twice about running it over something with sensitive content in it, which I couldn't say for the browser-based OCR tools I'd tried before.