microseyuyu

It's not about the test format; it's about relevance. The real test should be tossing a live issue at someone to see if they can fix it.

And if they fix it? Pay them. Simple.

Stop fooling around with abstract puzzles that have zero relevance to a battle-tested production environment.

I wouldn't call it "burning tokens"—I'd ask if those tokens are getting the job done. That is literally what AI is for.

Ultimately, hiring is a transaction: Can this candidate fix your issue?

Sure, you need to filter for red flags, but come on—the current interview meta is broken/dumb.

It really boils down to this: What value does A bring to B, and vice versa?