AI-native UX is still being invented. Our bet is that we should invent as little of the basic interaction model as possible.
Big Tech is spending enormous amounts of money teaching people how to interact with intelligence through text and voice. We should ride that wave rather than spend startup energy teaching users an entirely new interface.
Start with what people already know
Text and voice are becoming universal interfaces to software, but that does not mean putting an empty chat box in every product.
Blank text boxes make users wonder what they are supposed to say. We take cues from Chinese consumer products that surround an input with suggested actions, pills and contextual prompts. The input stays open ended, while the surrounding actions teach people what the product can do.
Stream the work
AI changes another old assumption about interfaces: work does not need to finish before the interface responds.
ChatGPT made streaming text familiar. Image products made progressive generation familiar. At Durable, we stream website creation while the site is being generated.
Intelligent systems often take time. Instead of hiding that latency behind a spinner, show useful progress and let the product feel alive while the work is happening.
One intelligence, many surfaces
The intelligence should not belong to the chat box. The same capabilities should be available through text, voice, mobile, desktop, SMS, calls and whatever interfaces come next.
The interface is a surface over the work the system can perform. That abstraction also lets us route between different models, tools and levels of intelligence without exposing that complexity to the customer.
They should not care which model completed the job. They should care that the job got done.
The interface is only one layer of an AI-native product. Evals, context, memory, tools and model routing make the experience reliable underneath. We cover those separately in AI native.
Deploy intelligence, not interfaces
The longer-term goal is not a better chat box. It is less interface.
AI-native products should do things for their users. They should be proactive, always available and increasingly capable of completing work without being asked for every step.
For Durable, the ideal experience might not involve opening the product at all. It might be a call or message telling you that your next booking moved to Thursday, the calendar has already been updated and the customer has already been notified.
Or that you had an opening tomorrow afternoon, three old leads were contacted and one booked.
The best interface for completed work is often a notification that the work is already done.
Do not anthropomorphize the work
We avoid inventing characters where they are not useful. Most customers do not need "Alice, your AI CMO." They need more customers.
Giving intelligence a name, face and personality can make a demo memorable, but it can also put the technology ahead of the job to be done.
We prefer capabilities with clear outcomes and traceability. Show what happened, show why when necessary and let the customer intervene when it matters. Otherwise, get out of the way.