I really enjoyed this breakdown by @raphaelschaad and @aaron_epstein (YC partner):https://www.youtube.com/watch?v=DBhSfROq3wU&ab_channel=YCombinatorThey walk through some cutting edge AI apps:@Vapi - voice api for devs@Retell AI - build voice agents@Gumloop - AI zapier@AnswerGrid - outbound research@Polymet - AI product designerZuni - ai in your chrome sidebarand @Argil - video clone generationThey talk about some recurring themes:- When prompts take a long time, give the user something engaging to do while they're waiting.- Canvas is a powerful UX paradigm for understanding workflows. I appreciated Rafael's insight that zooming in / out can allow for understanding a complex workflow at multiple levels of fidelity. - Voice interfaces - latency and interruptibility are the main factors in a good user experience.- Discoverability / templates. The problem with LLM prompts is they're so open - I don't know what I can do with them. Have some easy templates at your users' fingertips as a starting point. One big area I found missing is the subtleties of code assistants - In particular, I feel like the cutting edge of LLM interfaces is being explored by the likes of @Cursor , @Windsurf - and other code assistants that are being used in high-stakes production code. The other reason code assistants are interesting to me is that it appears these products are winning based on the interface, not the underlying models (since those are fungible). If you use Windsurf, Cursor, I'd love to hear what about the UX design feels new / superior.