AI-native interfaces & products

Products designed from the ground up for human-AI collaboration need more than chat. While the chat interface broke AI into the mainstream by being familiar—familiar to humans, familiar to AI—it's fundamentally limiting for real work. True AI-native products define a clear principle: anything a human can do in the product, an AI system should be able to do too, backed by specialized UIs that map to how humans actually think and organize information.

The chat breakthrough—and its limits

The familiar chat interface was the breakthrough. ▶ 7:47 "That's what made AI explode into the mainstream over the winter—that it's so familiar. Like humans became familiar to AI and AI became familiar to humans." But familiarity is not enough. ▶ 8:52 Once people encountered ChatGPT and the chat interface, the immediate question became "So this is cool and all, but what do I do with this?"

▶ 9:18 Seeing generative AI only as "putting text in text boxes" is limiting because ▶ 10:01 it obscures the reality that specialized UIs—Figma for design, Photoshop for pixels, Excel for spreadsheets—have been developed over decades for specific modalities and data structures.

Beyond chat: the necessity of specialized UIs

The primary interface for talking to AI is chat, but it can't be the only one. Products need to adapt their structure to what humans and AI can both meaningfully manipulate. With Multiply, ▶ 14:16 the UI is specialized: "Yes, you can write text, but..." the key innovation is the data structure underneath. ▶ 14:41 "Data lives on a graph and you can form relationships between data items and you can build hierarchies and trees and star-shaped data in that graph. And that maps to how the human brain works."

Defining AI-native products

The definition is deceptively simple but profound: ▶ 27:25 "Anything in the product can be done by both AI and people." This is the core principle. It means every function, every feature, every interaction must be accessible to both humans and AI agents. It's not about making AI do human work—it's about building products where both can participate symmetrically in the same actions.

Rasmus Adler Wahlberg, CEO of Multiply, frames this alongside a larger AI opportunity landscape: ▶ 28:12 "There are roughly 3 categories of opportunities within the AI space. The first is the foundational layer—foundational models and infrastructure. Then you have the current products that already have existing distribution. But the third one is that actually a lot of these products can be reimagined." AI-native products fall into that third category: reimagined products where the architecture assumes human-AI collaboration from the start.

Design principles: right-brain over left-brain

Product design for AI systems shouldn't default to hierarchical, spreadsheet-like (left-brain) structures. Rasmus advocates for prioritizing ▶ 25:46 associative patterns—right-brain thinking. "I think we need to really prioritize the right brain. I wonder what that will be like, like if you just prioritize the associative aspects of it." Graph structures, relationship mapping, and non-linear organization are more natural for both human creativity and AI reasoning than rigid hierarchies.

The frontier: multi-party voice interaction

Voice AI remains one-to-one. The challenge of multi-party voice conversation is profound and still unsolved. ▶ 28:46 "In my view, there's still no product that can do that, especially if you—an AI talking to a group of people is such a chaotic environment to be in that there's no product that even comes close or even have launched."

The sub-problems are deceptively hard: telling speakers apart and counting how many people are in the room, judging whose turn it is, deciding when it's relevant to speak without rudely interrupting, and reading the layered social protocols of meetings—who's the facilitator, who are the experts, who are the learners. This stack of unsolved problems explains why every voice product so far, including Kindship's own work, has stayed strictly one-to-one. Group-capable voice products are likely still 6–12 months away from realistic product form.