Inline prompting is requesting the help of an AI system by placing a _ in your text, exactly where you want the response to land. The system reads the surrounding context, infers what is being asked, and responds in place. The idea is developed in full in the Inline Prompting essay.
The underscore is the primitive
4 * 12 _ resolves to 48; NVDA _ resolves to a live stock price; improve prompt _ re-renders a prompt you've drafted. The essay's case for _ specifically: it's typeable on every keyboard (shift-hyphen on QWERTY, present on every smartphone layout), which is what lets the same primitive port from terminal to browser to mobile. And the shape is pre-loaded: schoolchildren have filled in blanks for generations, so nobody needs the symbol explained.
It inverts the autocomplete relationship
Traditional autocomplete pops up suggestions while you type: the system chooses the moment, the position, and the shape of the suggestion, and you respond. Blanks flip it. You type _ exactly where the response should land; the system has no say over position and only has to work out what belongs there. The user initiates; the system responds.
Properties that make it composable
- Ownership lock. Once a
_resolves, no cue can overwrite the result; only editing the word releases it. - Visible failure. If nothing can resolve the blank, the
_stays. There is no silent "tried, gave up" state. - Multiple blanks. One input can carry many
_s, each dispatched independently.
The essay also names the payoff: writing contexts with no AI available at all (web forms, note apps, messaging clients) currently force a detour to a separate app and a copy-paste back. Inline prompting keeps the request inside the work. OpenCues is the implementation of the idea.
Related: What is a blank? · OpenCues vs autocomplete · AI in any text field · GitHub repo