This is a study for a 26-key keypad for AI-assisted work. It is not a replacement for a keyboard. It is a way to make a small set of recurring AI actions physical and deliberate.
Every key needs four things: a clear purpose, a target, a visible response, and a limit. The interactive map below shows the current proposal. Select any key to see what it would do.
Select a key to see its purpose, response, and limits.
Take the selected direction one level further.
The prompt could have led to a row of application shortcuts. Instead, the study asks which moves recur across writing, design, research, and coordination. The labels below name those moves in ordinary language.
Putting these commands on a physical surface creates two design problems. The keys have to be findable without rereading all 26 labels. And the same key has to know what part of the work it is acting on.
Define what is in view.
Add material without deciding what it means.
Extend or change the work.
Inspect and assess.
Package, assign, or record work.
Adjust range or reverse an action.
A press cannot go straight from a label to an action. The system first has to identify the target, show what it plans to do, and make the consequence reviewable.
Needs a target
No target is assumed. Choose what to deepen:
Nothing has run yet
Building clickable states showed where the first map was too vague. The revisions below are interface models, not observed usability results.
The revision separates the target, the range, and the control used to confirm a choice.
The revision names the output contract before anything can run.
The revision turns a vague one-press command into Prepare, Review, and explicit Start.
What changedEach revision changed what the system had to show. It did more than make the screen look cleaner.
Ongoing
The current model is still a proposal. There is no physical keypad, application bridge, or evidence from observed use.
The open questions are practical: which labels become familiar through use; how targeting works without returning to a mouse; how much state belongs on the keypad and how much belongs on screen; and what permission and reversal should mean across applications.
Stop moved from 67g back to 55g. A heavier switch made stopping feel like a commitment when it should be immediate.
The next useful step is to build a small physical prototype and watch where the labels, targeting rules, and review states break down.