Screen questions
Half of a technical call is on somebody’s screen and none of it is in the transcript. ⌘E puts the screen into the question.
They share a dashboard and say "this spike". They put up an architecture diagram and say "the bit on the left". They scroll a stack trace faster than anybody can read it. A transcript of that conversation keeps every word and loses what each one was pointing at.
⌘E sends the current screen along with your question, so the answer is about what is on it: the numbers in that table, the shape of that diagram, the error in that trace. It is one action rather than a mode you switch into — a capability you have to go and turn on is one nobody uses in the four seconds they actually had.
- The chart they are sharing, read rather than described
- The code on screen, answered against the code on screen
- A term visible in a contract or a spec that nobody is reading aloud
Where it stops
- It does not watch continuously — a screen is sent when you ask a question with one, and not otherwise
- It reads your screen, not anybody else’s
- With optional audit consent, the encrypted screenshot may be retained for at most 30 days; without it, the pixels are not kept
- It does not decide for you whether what is on screen may be sent anywhere
The screen goes to a cloud model with the question, the same as the transcript does. If you would not paste it into a chat window, do not attach it to a question.