HackerRank, which companies use to test developers before hiring them, has opened its AI interviewer to all its customers. A candidate gets a task in a real code repository and works with an AI assistant beside them. The interviewer then asks why they chose an approach and what they would change if the requirements moved.
A test that allows AI measures something different
For years these tests graded the output: did the code run, did it pass the cases. That worked while producing the output was the hard part. Now anyone with an assistant can produce something that passes, so the output stops separating people. What still separates them is what they did with the assistant: what they asked for, what they refused to accept, and whether they can explain the result to someone else.
Take two candidates who hand in the same working code. One accepted the assistant's first answer. The other threw out two answers because they broke on an empty input, then wrote a test to prove it. A grader that only looks at the code cannot tell them apart. One follow-up question, "why did you rewrite this part?", can, in under a minute.
Watching candidates work with an AI assistant was already possible on HackerRank, where human interviewers could see how a candidate used the assistant. What Chakra adds is that the follow-up questions also come from AI, so every candidate gets them, not only the ones an engineer had time for.
Allowing AI also cuts the cheating
A ban on AI turns a tool that candidates will use every day at work into a secret. The ones who use it anyway are mostly trying to work the way they normally do, and the test punishes them for it. Allow the assistant, put it in the same window as the interviewer, and there is nothing left to hide. In HackerRank's own comparison, suspicious-activity flags were 70% to 80% lower than in its traditional assessments, though the size varied by country and seniority. That is the direction this mechanism predicts.
The same test works outside coding
A finance director hiring an analyst could give each finalist a real budget spreadsheet and an AI assistant, then spend ten minutes asking which number they did not trust and why. The spreadsheet shows that the candidate can get an answer. The ten minutes shows whether they would catch a wrong one before it reaches the board.
There is a time saving too. Say a team gives an engineer an hour with each of 20 candidates for one role. If an AI interview narrows that to the best five, the team gets about 15 hours back.
A score is not a decision, and the rubric is the real interviewer
Chakra scores candidates and leaves the hiring decision to people, which is the choice school districts made too when they kept humans in charge of AI hiring tools. Applying one rubric to everyone removes the interviewer's mood and tiredness, but not the rubric's blind spots. A rubric that rewards fluent, confident explanations favours people who talk well over people who think well, and that costs most for candidates interviewing in their second language. Some places already treat this as a legal risk: New York City requires an independent bias audit and notice to candidates for certain automated hiring tools.
So the question for anyone hiring with a tool like this is what its rubric actually rewards. For anyone applying, the safest preparation is to practise saying why you did something, not only doing it.