Skip to content
Lab4 entries

Interaction studies and prototypes, built to answer the questions a static frame can't.

Some shipped, some didn't. One of them runs right here.

01 / 04

Common Ground

sep 2026prototype

The actual problem: put three people and three AIs on one project and the default answer is a shared chat, which flattens everything anyone believes into a transcript nobody rereads. This builds the other thing, a typed model of what the team knows, assumes, disputes and has decided, where every card carries whether a person said it, a model inferred it, or the team accepted it. Private threads stay private; moving something out of one is a deliberate act with four different destinations, and proposing it as truth opens a review instead of writing it. Disagreements sit side by side rather than getting summarized away, and the model's confidence number is shown last, under the evidence, because a number raises trust in a wrong answer even when it means nothing.

open live ↗
Esc closes
A private AI thread open over a shared knowledge model, offering four ways to move an insight out of the private context