VO: AI-Assisted Coding / Debug Round (HackerRank IDE + AI)

New 2026 onsite round shape. Uses a HackerRank-hosted IDE that closely resembles Cursor, with an in-browser AI coding assistant available. The interviewer hands over a small full-stack app with a bug (e.g. transaction handling) and grades your ability to debug using AI as a teammate. Requirements The IDE includes a built-in AI assistant. AI use is explicitly allowed — this is the differentiator from older coding rounds. A small full-stack application (frontend + backend, often with a transaction feature) is preloaded. The interviewer states the failing scenario (e.g. "transfers between accounts sometimes produce the wrong final balance") and asks you to diagnose and fix. You may also be asked to add a new small feature on top of the same codebase. Notes The IDE supports running the app live in-browser. Spend the first 5 minutes orienting to the project layout — multiple candidates have lost the round to navigation friction. Treat the AI as a junior teammate: ask it where a function is defined, ask it to summarise a file, ask it for hypotheses on the bug. Do not paste the entire problem and accept the first suggestion — interviewers have flagged candidates who copy-pasted AI output without verifying. Effective prompting strategies reported: "Which function might be responsible for handling this transaction?" → narrows the search. "What invariants does this function assume? Are any of them violated?" → drives the bug hunt. Production-tier debugging skills (reading stack traces, adding logging, isolating the failing case) matter more than algorithmic depth. This is closer to a real on-the-job debugging session than a leetcode round. The format is new and varies. The HackerRank IDE may or may not be Cursor-shaped depending on the specific team running the round. The session transcript of your AI prompts is part of the evaluation surface. Vague prompts ("fix this bug") and accepting the first suggestion are explicitly graded down; surgical prompts that name a function or assert an invariant are graded up. Treat each prompt as a small unit-test request, not a global solve. Preparation Practice using Cursor or GitHub Copilot for code navigation in an unfamiliar small repo — clone any open-source project and force yourself to find a specific function in <2 minutes using only AI prompts. Drill debugging exercises (intentionally broken Python or Node apps) with an AI assistant. Time yourself. Pre-script a 3-step debugging mental model: reproduce the bug, hypothesise where it lives, verify with a targeted change. Use the AI to accelerate steps 2–3, never to replace step 1. Run a 30-minute self-drill against an unfamiliar small Express + React repo with a planted bug, using Cursor or a HackerRank AI-IDE trial: practice the orientation pass (entry points, routes, data layer) and the prompt cadence (locate → hypothesis → verify) before the live round.