A chat panel makes you leave your work to use AI: switch surface, describe the task, wait, copy the answer, switch back, paste, edit. Every round trip is two context switches. OpenCues answers inside the text you are already writing, so there is nothing to switch to and nothing to paste back.
The cost model
Borrowing the 3-axis lens from the HCI essay: a chat panel has heavy start-up frames (navigate, focus, re-describe context), a blocking active window (you wait, then read), and heavy end-lag (return, re-find your place, paste, adapt). A blank is one keystroke to start, lets you keep typing while it resolves, and has zero end-lag because the answer lands where you put the _.
Context comes free
In a panel you re-explain what you are doing. Inline, the surrounding text is the context: make this more formal _ needs no explanation of what “this” is.
A different safety boundary
Agentic chat tools act: they run commands, call APIs, send things. OpenCues only ever edits your unsent draft; you are the one who presses Enter. Worst case, a bad response is text you see and delete before anyone else does.
Related: AI in any text field · the Inline Prompting essay · OpenCues vs Raycast AI · Stop copy-pasting from ChatGPT · GitHub repo