Philip Han

beyond the text box: the bottleneck isn't the models anymore – it's the interface. our thoughts aren't linear. they're spatial, visual, emotional. sketches with arrows everywhere. half-formed plans. a melody that captures the vibe. but we're still forcing all of that through a text box. visuals, audio, code, writing – they're all just data in different forms. and ai can understand all of it now. vision models read images. audio models hear music and speech. multimodal models blend everything. so why are we still typing? imagine expressing an idea by sketching a diagram, humming a melody, dropping in a reference video, writing some throwaway code – all at once. the ai gets the whole thing: the structure, the feeling, the relationships, the intent. the interface isn't a chat box. it's where you compose thoughts the way they actually play in your head. and in teams, everyone thinks differently. designers sketch. engineers code. PMs map flows. writers draft stories. the tool should not create echo chambers but speak everyone's language – a shared space where you express ideas in your natural form and the system translates between them. agents aren't replacements. they're collaborators. the models will never nail exactly what you want without your input – not because they're not smart enough, but because your intent is yours. it's shaped by context they'll never have, by taste they can't infer, by the thousand micro-decisions that make something right instead of just correct. collaboration means continuous dialogue, not fire-and-forget. these modes aren't separate systems. you don't "switch to canvas mode" – you just express yourself however feels right. text, pixels, sound waves – it's all context. the ai doesn't care. we're no longer capped by what models can do. we're capped by how we talk to them. that's what we're building at Cursor.