The skill nobody tests for

Over the last year, the best engineers have completely changed how they work. They don't write code by hand anymore, they talk to agents, interrogate them, make architectural decisions, scope projects, and verify output. Agentic engineering is real, even Torvalds is doing it.

But when these same engineers go to hire, they're still screening candidates on algorithm puzzles. We don't have anything against leetcode. It produces valuable signal for intelligence and conscientiousness. But intelligence alone isn't enough to make a great developer.

So what should we actually measure? The ability to effectively direct AI tools through complex, ambiguous problems. How someone decomposes a problem before prompting. How they evaluate AI output critically. How they iterate when the first answer isn't right. Whether they build verification infrastructure rather than trusting output blindly. How they handle the messy underspecified stuff that AI alone can't resolve. This is a real, distinct skill. Some engineers are dramatically better at it than others, and right now there's no way to screen for it.

VibeArena is here to test this. how well can your candidates drive AI, how well do they scope projects, are their architectural decisions sound? The challenges are designed to be solved with AI, what gets measured is how they use it. The full interaction artifact: every prompt, every AI response, every edit the candidate made after receiving output. As a concrete example, imagine a challenge where the candidate gets a buggy RBAC module with a test suite. They fix the obvious bug, all 15 tests pass, green across the board. A weak candidate submits. A strong candidate doesn't trust 15 tests as sufficient coverage for a combinatorial problem, they build a fuzzer, sweep the input space, and discover the edge cases the provided tests never exercised. That difference in instinct is exactly the signal we capture and no other platform attempts to.

Our core belief: the way we evaluate engineers should reflect the way engineers actually work. AI-native skills are real, they're differentiating, and the companies that figure out how to screen for them first will build better teams than everyone else.