2 min lesson
Strong driving, concretely
Use "Strong driving, concretely" to tell the cases apart, then choose a response for each one.
Step 1 of 2
Strong driving, concretelythe moves the panel can see
"Use AI well" is too vague to rehearse. The people who actually use Cursor daily drive it with four habits that read as authentic on a shared screen. Practice each until it's reflex, because the interviewer is watching how you operate the tool, not just whether the tests pass.
State the plan before you let AI write a line
The flagship loop: plan with the strongest reasoning model, hand the to-dos to a cheaper, faster model to implement
Letting the expensive reasoning model write all your code burns the budget; you don't need it once the plan exists
Tag the precise file instead of asking vaguely and making the tool hunt the whole repo
@-mentioning guarantees the file lands in context rather than relying on semantic search
Night-and-day difference in both speed and accuracy of the output
"Make it better" on the most expensive model is the canonical anti-pattern
With no clear context the model makes blind edits, then burns more tokens reading and fixing them
Scope it: name the surface, the exact change and the constraints
Model output ships with generic names, verbose comments and needless abstraction
Cursor engineers run a de-slop pass at generation time, before code reaches a PR
Clean it to the standard you'd defend, then accept - never paste raw
The vague-vs-scoped contrast is the one to internalize; here it is side by side.
// WEAK: no surface, no change, no constraint - blind edits, compounding cost. make the app better // STRONG: surface + exact change + constraints, on a model matched to the task. Under the shop page, add an in-stock filter next to the existing tag and sort controls. Preserve the URL query params and update tests if needed.
Learn more
Full explanation
Full explanation
“I'll plan this with the reasoning model first, then hand the to-dos to a faster model to implement - no point spending the expensive one on boilerplate. I'll @-mention the exact file so it doesn't have to guess, and I'll run a quick de-slop pass on whatever it writes before I accept it.”