Before coding assistants, I liked to give applicants a technical assessment (programming or data science exercise) that contained a fair bit of ambiguity and missing details. If they start asking clarifying questions about the assignment, that's a good sign. If they don't ask questions but make a few assumptions and then double down on building the wrong thing, that's a red flag for me. This approach means you hire people with good communication skills over 10/10 tech skills. Missing tech skills are much easier to fix in post than missing communication skills, provided you have a good mix of junior/seniors in your team.
Since LLMs are overtrained on fulfilling whatever request you throw at them, I would now also include a not-to-do or a "bad decision" in the technical assessment. Something nonsensical that the candidate should give a bit of pushback on, and a coding assistant will probably happily implement for you.
For inspiration, see BullshitBench, my favorite LLM benchmark of all time.