Generative UI doesn't make sense for startups
This video popped up in my YouTube feed yesterday and it feels roughly right to me. You should watch the video, but to keep it moving, here’s an AI summary:
This video examines the evolution of user interfaces, arguing that the future of computing lies not in replacing screens with pure AI agents, but in generative UI (9:41-10:32). While traditional graphical user interfaces (GUI) excel at simple tasks through direct manipulation, they falter under extreme complexity, where agentic AI proves more efficient (8:32-9:16). The emerging solution—generative UI—creates dynamic, context-aware interfaces in real-time, blending the ease of visual interaction with the power of automation to create a hybrid, highly adaptable user experience (13:31-14:36).
“UIs are over” is clearly a case of throwing the baby out with the bathwater. The end state Enrico lands on seems plausible1, but I think “generative UI” is only a practical solution for established companies.
All of the companies mentioned in the video (Google, Airbnb, Spotify, etc) found product-market fit decade(s) ago. As such, one has to imagine that they’re pretty good at defining the kinds of jobs their customers want done and the kinds of UI components their customers need to do those jobs.
If I’m starting a company and don’t yet have evidence that I know what customers want and will pay for, then I’m not going to know what components and interactions to encode in rules. I could try, but I would be very likely to encode the wrong things. And if every user is theoretically getting a different interface, generated on the fly, to solve their specific immediate use case, it seems much harder for me to build intuitions about how I can best serve them.
Call me old-fashioned, but I think building a specific, tailored solution for a customer’s problem is still a good way to zero in on product-market fit. Once you start having evidence that your solution is working, then you can experiment with encoding the DNA of your solution into rules for generative UI elements. Wait until the gold’s been found before building the mine.
I can already hear some of you saying in your heads “but you can monitor each user’s generated UI and have an LLM distill these recorded sessions into patterns and rules to feed the generative UI!”
I guess so? It just seems like an overinvestment in AI for the sake of it. And likely to manifest in so many degrees of freedom in the overall product surface that it would be hard to hone and deliver on a consistent product vision.
And with so much of this stuff, it feels like we’re too eager to automate away parts of startup-building that help you develop intuitions about customers’ motivations. I have a hard time believing that customers will be served well by products built by people who don’t want to be in the details.
Links
🎶 [Jazz] Chet Baker Trio – Chet’s Choice
🎶 [Ambient] Lusine – Language Barrier
🎶 [Electronic] Dave Stapleton – Portrait Mode III
🎶 [Ambient Country] Chuck Johnson – Balsams
🎶 [Italo Disco] Two recent Herb Sundays mixes
Or at least more so than the idea of normal people using computers exclusively by writing prompts instead of tapping buttons.