The loop you're trying to escape: switch to a ChatGPT tab, re-explain your context, wait, copy the answer, switch back, paste, fix the formatting. The structural problem is that the answer arrives in the wrong window. OpenCues' fix is to move the request into the text itself: type _ where the answer belongs, and it lands there.
The same jobs, without the round trip
| You used to open a chat tab for | Now you type |
|---|---|
| Quick facts and conversions | capital of france _ · 5 miles in km = _ · 100 celsius in fahrenheit = _ |
| Rewrites of what you already wrote | make this more formal _ · fix the grammar _ · translate to spanish _ |
| Drafting from scratch | draft an email _, generated in place, signed via identity context |
| Live data | weather london _ · nvda _ · bitcoin _ |
Notice what disappears: re-explaining context (the surrounding text is the context), the copy, the tab switch back, and the reformat. In the project's 3-axis terms, the end-lag drops to zero because you never left.
Why the friction was killing the habit
The Inline Prompting essay names the real cost: when getting AI help means leaving the field, opening a separate app, and pasting back, most writing tasks that would benefit from AI assistance never get it. The friction doesn't make each use slightly slower; it makes most uses not happen. A one-keystroke primitive that resolves in place changes what you bother to ask.
What the chat tab is still for
Honestly: multi-turn work. OpenCues is deliberately single-shot per _; it is not a conversation surface, and exploring an idea across ten exchanges belongs in a chat. The split that sticks is: conversations in the chat tab, everything with a known destination (this sentence, this field, this draft) resolved inline. And when you are writing to ChatGPT, OpenCues works inside its prompt box too, improving the prompt before you spend the turn.
If the blank can't answer
The failure mode is visible and cheap: the _ just stays in your text. No error dialog, nothing lost; you can rephrase, or fall back to the chat tab for exactly the cases that earn it.
Related: What is a blank? · AI in any text field · OpenCues vs chat panels · GitHub repo