Nailing the right UI for AI

We're still figuring out the best way to 'work with' AI

Chat interface was v1 —> where is UI going from here?

One of the most exciting and under discussed opportunities in AI right now is to develop new UI models for AI-first products. While the chat interface has emerged as the dominant UI today thanks to the success of chatGPT, this UI is extremely limited. The reasons why (does not natively have context, exists in a separate window from your work, etc.) are fairly self evident.

What’s less evident is exactly what new UI patterns will emerge and win in the next chapter of software. One product I’ve been enjoying lately that inspired this post is Granola, an AI-notetaking app. However, unlike most note takers which operate as fully autonomous solutions, Granola is has a much more collaborative UI model.

I have taken my own notes on calls for years and, as I’ve written previously, taking notes helps me recall information. The Granola UI-pattern allows me to still take my own notes but not worry about transcribing the information word-for-word. I can focus on substance. Even starting with just my own sparse set of notes, I find the Granola AI-generated notes are much more in line with what matters to me than a fully automated model.

Rather than expecting the model to infer exactly what I want to capture, my input helps guide the model towards my goals. My input changes the goal of the note taking model from a very broad task to a narrow one. Further, my input could help train future versions of the product to better identify what is most important from a given transcript. 

This UI pattern is not one size fits all but to me illustrated that we’ve only just scratched the surface of how we can design AI-native products.  

Designers have always been an important (and sometimes under appreciated) part of a startup’s success. In the era of AI, being mindful of design and UI from day one is even more critical. The job of founders and early designers of AI companies will be even more reliant on creativity and original thinking than for traditional software where so many best practices already exist. Some may scoff at design as a ‘soft moat’ but all too often design can make all the difference.