A designer can say what they want and point at what they mean, at the same time, the way you'd talk to another person leaning over the same desk. That's the idea being tested here. Deckard is a working prototype where "make this one red" and "align them left" do what you'd expect, and where the system has to work out whether "this" means the thing under your cursor or the thing you were just talking about.
Seven videos, in order. The first is a short tour of the screen, so the rest isn't a hunt for where to look. Then three on the mechanics: how you say things, and how the system works out what you meant. Then three on why any of it matters: the distance your hand travels in a normal working day, what happens if you only have one hand, and lastly what a tool can start telling you about your own work once it can see the pieces separately.
A work in progress and a capability POC, not a design tool. Every video is a single unedited take. I didn't code it myself. Accessibility isn't a claim yet.
Start at the beginning if you have the time. If you don't, Part 4a is the one that makes the case.
Part 0 — How to read the screen · 6:19{{ watchedP0 }}
Two minutes on what each panel is for, so the rest of the series isn't a hunt for where to look.
Part 1 — Saying the first thing · 8:33{{ watchedP1 }}
Creating and changing objects by voice, and the first real test: getting the system to understand "this" and "it" as two different questions.
Part 2 — Talking about objects · 12:08{{ watchedP2 }}
Referring to things you've already made — by name, by pointing, or by what you just said — including six cards built from five spoken lines.
Part 3 — Arranging things · 17:55{{ watchedP3 }}
Selecting, grouping, aligning and rotating. Groups behave as one thing, corners stay put through a rotation, and at the end the system declines to guess when a request is genuinely ambiguous.
Part 4a — Why this is worth doing: distance · 7:10{{ watchedP4A }}
The walking. A design tool sends your hand from the canvas to a panel and back a few hundred times an afternoon, and Fitts's Law says exactly what that costs. Voice doesn't shorten the trip, it doesn't make it.
Part 4b — Why this is worth doing: access · 10:48{{ watchedP4B }}
The second reason, and the one I didn't design for: a canvas tool normally demands sustained fine motor control. It ends with a build made without touching the mouse at all.
Part 5 — Checking the work · TBC{{ watchedP5 }}
A button comes apart into a shape and a label, which is what makes it possible to ask whether anyone can read it. Contrast measured against the real standard, and a suggested colour that stays in the hue you chose.
Let Mø know what you're working on and what you need from design leadership or product innovation.