AI Debugging Challenge
A coding test that assumes what's already true: your candidate has AI. So do all your candidates — the same one, at the same time, on the record. You get pass/fail, plus an evidence-backed read of how they used it.
Debugging Workspace · AI Debugging Assessment
Fix the code
Auto-savedfunction mergeSorted(a, b) {
let i = 0, j = 0;
const out = [];
while (i <= a.length && j <= b.length) {
if (a[i] < b[j]) out.push(a[i++]);
else out.push(b[j++]);
}
return out;
}Your edits are saved automatically. The code is reviewed after you submit — there is nothing to run here.
AI assistant
a[i] return past the end?How confident are you in this fix?
Optional — you can leave this blank.
A candidate's debugging session — buggy code in the editor, the platform AI assistant alongside it, one submit that locks both. Sample content shown.
One bug. One evidence-backed read.
What you give it
A code snippet with an intentional bug, the expected behavior, and a rubric. Write it yourself, or start from a Question Studio draft and edit it before it's published.
What the candidate does
Debugs the code in a real editor while chatting with the same standard AI assistant every candidate gets. No run button — the work is reviewed, not executed. No bringing a personal AI tool.
What you get back
Pass/fail against your rubric, plus a categorical read of how they worked with the AI — never a score. Click any claim to see the exact chat moment it came from.
How it works
Author the challenge
Write the buggy code, the expected behavior, and the rubric — or generate a draft and edit it. It goes through the same Draft → Review → Publish approval every question on the platform already uses; nothing reaches a candidate unreviewed.
The candidate debugs, out loud
They work in the editor and can ask the AI assistant for help at any point. Every message and every paste is part of the record — and the candidate is told that, before they start.
Submit locks it
Final code and the full conversation are frozen the moment they submit — the same irreversible, one-shot submit every assessment on the platform already uses.
A dedicated grading step reads the collaboration
Code correctness is graded against your rubric. Separately, a Meta-Grader classifies how the candidate worked with the AI — validated it, leaned on it uncritically, barely used it — and attaches the transcript evidence behind each read.
A read, not a ruling
AI Debugging Challenge tells you what happened in the session. It never decides what happens to the candidate.
Recommendation, not a ruling
Nothing here auto-fails, auto-advances, or auto-eliminates a candidate. The report is evidence for a person on your team to weigh.
A category, never a number
The collaboration read comes back as a label — “validated the AI’s output,” “accepted it without checking,” “used it lightly” — not a 0–100 score standing in for certainty we don’t have yet.
The same AI, for everyone
Every candidate gets the identical platform assistant. Nobody gets a better model, a personal ChatGPT account, or no AI at all — it's the same tool, every time.
No code execution
Candidates write in a real editor; there's no run button and no sandbox. Correctness is graded against your rubric, not a hidden test suite.
Told upfront, not discovered later — before a candidate starts, they see exactly what's collected: the full chat transcript and paste events, and why.
See it on one of your own questions.
We're onboarding our first hiring teams into AI Debugging Challenge. Bring a role that involves real debugging work, and we'll show you the candidate side and the report together.