Problem
Currently, I have this rule:
When asking the user, always enumerate your questions with q1, q2, … and provide most probable answers enumerated a1, a2, … where the most probable answer is a1. Expect the user to reply with q2a3 meaning that the answer of q2 is a3 or with text or with both q2a3 and clarifying text.
It would be faster and easier to comprehend, if for agent asking the user (me) multiple choice questions there is a dedicated UI.
(e.g. the tasks checklist UI you’ve recently implemented in Chat is very nice.)
Why is this important
I think it may be a part of a dominant workflow where the user gives the necessary minimum of context and leaves the agent to ask him questions for clarifications. If answering them is easy to comprehend and click with the mouse - it will be very efficient process. Cursor is all about UI efficiency - the rest is provided by the AI models and tools.
A step further to make it even more useful feature
The user to give less than the necessary minimum - just the bare bones and the rest is given by the user by answering multiple-choice questions. That way - the agent does the specs writing (via questions and answers), the structure of the assignment, the check for completeness of the info in the assignment - while the user mostly just reads questions and clicks with the mouse to answer (far less typing, far less voice-typing).
The big picture
I think the trend is for the user to guide the model, rather than instruct it in details. We need better tool(s) for guiding that will replace most of prompting (typing/voice-typing) in the future.