“I stopped doing leetcode-style take-homes because AI solves them in seconds. Now I ask candidates to walk me through a real problem they solved, what alternatives they considered, and how they verified the AI-generated parts were actually correct.”
“I do not police whether a candidate used AI. I look at whether they can defend every line: trade-offs, edge cases, why this abstraction and not that one. A clean take-home that crumbles in conversation is the shortcut I actually care about catching.”
“I actually got burned by this recently: my management team was seeing candidates openly using AI interview tools and even identity fraud in screens, and I had to rethink the whole front of our funnel.”
“I’ve been writing 99% of my code with AI since late 2025, so the shortcut I look for in take-homes is the same one I watch for in my own PRs: code that looks plausible but the author cannot verify. I ask candidates to walk me through one failure mode they fixed, because anyone vibing a solution can make it run, but only the person who debugged it knows where it breaks.”
“I look for code that is clean but not lived in: perfect variable names, no scars, no comments explaining a changed mind. Then I ask the candidate to walk me through one ugly decision in the code, and AI-generated submissions usually trip there.”
“I look for candidates who use AI agents but still show their own judgment: a clean report isn't enough, I want to see the weird traces they chose to investigate and what they made of them.”
“I look for the parts that are too clean: variable names that all share the same rhythm, comments that explain the obvious, and a commit history that jumps straight to a working solution. At Applause I have seen candidates submit code that compiles but does not think, and the shortcut usually shows up in the gap between the prompt and the design.”
“I look for candidates who can explain why they picked a trade-off, not just the cleanest output. If every edge case is handled suspiciously well and they cannot walk me through the dead ends they tried, that is usually the signal.”
“I look for the tell where a candidate can explain the code they submitted but cannot change it under pressure. If they used AI, I want to see they still own the decisions, not just the output.”