I have interviewed a lot of engineers, and I have been interviewed by a lot of companies, and I think most interviews are trying to learn the wrong things. Or rather, they are trying to learn the right things with instruments that cannot measure them.
I have come to believe a good engineering interview is trying to answer three questions, and only three.
Can they do the work
Not can they do any work, or impressive work, or work under pressure in front of strangers. Can they do the work this role actually involves, at the level the role needs.
I think the best evidence for that is a conversation about work they have done, in enough depth that it is not possible to fake. What was the problem, what did they try first, what did they change their mind about, what would they do differently. A person who did the work can go as deep as you like. A person who was nearby cannot.
The second best evidence is a small, real problem of the kind the job contains, worked through together. Small enough to finish, real enough to matter, and worked through rather than performed. I would rather watch someone reason through a problem for twenty minutes than watch them solve a puzzle in five.
How they think when stuck
Everyone gets stuck. The job is mostly getting unstuck. So the interview should contain a moment of being stuck, and what I am watching is what happens next.
Do they say what they are thinking, or go quiet. Do they ask a question, or guess. Do they try something and look at the result, or defend the first idea. None of those is disqualifying on its own. But the pattern of them is the clearest preview I know of what the person will be like on a hard week.
This is why I think the surprise question is a poor instrument. It produces being stuck, which is useful, but it produces it in a way that mostly measures composure under ambush, which the job rarely requires.
What they are like to work with
This is the question most interviews either skip or reduce to whether the interviewer liked the person, which is a different thing.
What I try to get at is how they handle disagreement, how they talk about people they have worked with, and whether they can take a suggestion mid-conversation and use it. The questions they ask about the team tell me a great deal here. So does how they describe a past team's failure, and whether anyone else in the story gets any credit.
What an interview is not for
I think an interview is not a test of stamina, so I do not run six hours of them. It is not a test of trivia, so I do not ask things that a search would answer. And it is not a place for surprise for its own sake, because surprise measures surprise.
Where this stops being true
Specialized roles sometimes need a specialized test, and a deep technical role may need a deep technical exercise that takes real time. Even then, I think the three questions are the ones being answered. The exercise is just the instrument, and it is worth asking of every instrument which of the three questions it serves.
Photo source: https://photos.robertstowe.com/imperial-sand-dunes