The original note, as written at the time
Imagine you could set up your applications or your phone by just using your text prompts or your voice.
Ongoing development
Imagine being able to set up your applications, or your whole phone, by using text prompts or your voice.
The reason this matters is not convenience. It is that configuration is the last part of computing that still requires you to know a machine’s vocabulary rather than your own intention.
Every settings screen is an argument the manufacturer won about how the thing should be described. You want the phone to stop waking when a message arrives; you must find a screen called Notifications, then a section called Lock Screen, then a toggle whose name nobody would use in a sentence. The distance between the wish and the switch is the tax.
Speaking collapses it. “Stop the screen lighting up when messages come in” is a complete instruction, and a system that understands intention can go and set the four things that sentence actually implies, some of which the user did not know existed.
The hard part is that setup is consequential. A misheard instruction about permissions is a security problem, and a system that confidently applies the wrong configuration is worse than a user who could not find the menu. So the design has to show what it is about to change, in plain language, before it changes it.
There is a better version than one-off commands, and it is the one worth building: setup as a continuing conversation. The machine asks why, you answer, and it proposes a configuration — and later, when something stops working, you can ask what changed and get an answer.
Preferences are not settings. They are sentences, and we have been translating them by hand for forty years.