Thesis
Why we're betting on generative UI
Chat won the first interface question — how you talk to an agent. The open question is how an agent shows you your business. We restructured aeqi around the answer: Pages, and a bet that agents composing calm, design-system UI is how they'll express themselves to the people they work for.
July 18, 2026 · Luca Eich
Every agent product has answered the first interface question: how do you talk to the agent? Chat won, deservedly. The second question is still mostly unanswered: how does the agent show you what it's doing? A transcript scrolls away. A business does not fit in one.
Our answer — and it is a bet, made deliberately — is that the agent should compose the interface itself. We restructured aeqi around it. The product now has two halves: Sessions, where you direct your Company in language, and Pages, where the Company shows you its work.
A view is a page
Everything in an aeqi Company renders as a Page: the quest board, the org chart, the memory graph, an app you connected yesterday. Apps are sources; pages are arrangements — the way a music library has artists and playlists, both rendered the same way, never confused with each other. Every app ships a canonical page, and any component can flow into any page. The interface stops being a fixed hierarchy of screens and becomes a surface the Company arranges to be understood.
Generated doesn't mean improvised
The obvious way to do generative UI is to let a model write HTML. We're not doing that. UI improvised from scratch is janky, unsafe, and different every time it regenerates — the opposite of what an operating surface owes you. In aeqi, agents write page specs against the design system's component vocabulary: the same crafted components the product itself is built from, at deliberately coarse grain — the crafted quest board is one component, not a pile of bricks. Generation decides what to show; the design system decides how it looks and behaves. That is how a generated page stays calm, coherent, and safe enough to run a business on.
Chat is the build tool. The page is the artifact.
The loop we're building toward: say what you want to see, and the page re-renders live while you talk — a diff to the spec, not a fresh roll of the dice. A session and the page it's authoring pair up, and you move between them the way you move between a conversation and the thing it's producing. You direct in language; you verify at a glance. That's the real division of labor: pages are optimized for legibility, the same reason a diff view beats rereading the whole file.
Where this stands, honestly
Today, the foundation is shipped: Pages are live across the product, every primitive and connected app carries a canonical page, and the front door is a composer — the latest changelog walks through it. Agent-composed pages are the layer we're building on top — and the reason we believe it's buildable is that the contract layer already exists. Every primitive in an aeqi Company exposes typed tools that agents already work through; the interfaces describe themselves to machines today. Teaching agents to compose them for humans is the next move. That is the bet.
A browser for your Company
The endgame reads simply: aeqi as a browser for your Company. You talk to it in Sessions — and everything else is a Page: the work, the team, the memory, native or connected, with your agents able to recompose any of them while you watch. Agents that can only talk stay assistants. An agent that can show you the live state of your business, arranged for your eyes, is something else — a Company you can actually see.