Google pilots Gemini in coding interviews

Google is changing a part of the classic coding interview — and that matters because the classic version was built for a world where candidates had no AI help. The company is piloting a format that lets some software-engineering applicants use Gemini during one interview round. That does not mean Google is replacing human judgment with a chatbot. It means Google is starting to test for a different skill: whether a candidate can use AI like a strong engineer instead of like a shortcut. ### What exactly is Google changing? The pilot centers on a “code comprehension” interview for some junior and mid-level software-engineering candidates in the U.S. Instead of writing everything from scratch with no assistance, candidates can use an approved AI tool — in this case Gemini — during the round. The task is less about inventing an algorithm on a blank page and more about reading existing code, spotting problems, fixing bugs, and improving performance. Google told Business Insider it is updating interviews to better match how engineering work now gets done. (businessinsider.com) ### Why that round? Because code comprehension is already closer to real work. Most engineers do not spend their day solving whiteboard puzzles from memory. They read unfamiliar code, trace behavior, debug weird edge cases, and make messy systems a little better. AI can help with that, but only if the human using it can tell when the suggestion is wrong, shallow, or dangerous. That is the skill Google seems to be probing. (businessinsider.com) ### So candidates can just ask Gemini for the answer? Not really — that is the whole point. The interview is still human-led. Interviewers are reportedly looking at how candidates prompt the model, what they accept or reject, and how they validate the output. Basically, the test shifts from pure solo production to supervised collaboration (businessinsider.com)losing judgment. (businessinsider.com) ### Why do prompting and validation matter so much? Because AI coding tools are useful in exactly the same way junior teammates are useful — they can save time, but they can also sound confident while being wrong. The hard part is not generating code. The hard part is knowing what to trust, what to inspect, and what to rewrite. In that sen(businessinsider.com)shift inside software teams, where developers increasingly manage and review AI-generated work instead of typing every line themselves. (businessinsider.com) ### Is this just about Google promoting Gemini? Partly, yes. Google obviously wants Gemini to be seen as a serious developer tool, and using it in interviews helps normalize that. But the change also reflects a real market fact: AI coding assistants are already common. Google offers Gemini Code Assist broadly to developers, including a free individual tier(businessinsider.com), an interview that bans it completely starts to look artificial. (developers.google.com) ### Does this kill the old whiteboard interview? Not yet. This is a pilot, not a full rewrite of Google hiring. The company appears to be testing one interview type with some candidate groups rather than throwing out the whole system. But the direction is clear — at least one of the biggest names in tech hiring now thinks “AI-assisted” can be a valid way to measure engineering ability, not a form of cheating. (businessinsider.com) ### What changes for candidates? Preparation gets weirder, but also more realistic. Candidates will still need fundamentals — data structures, debugging, code quality, tradeoffs. But they may also need to practice using AI under pressure: writing precise prompts, checking outputs fast, and explaining why they overruled the model. The candidate who blindly pastes suggestions will probably look worse, not better. (businessinsider.com) ### Bottom line Google is not saying raw coding skill no longer matters. Google is saying raw coding skill by itself is no longer the whole job. That is the real shift here — the interview is starting to look more like the work.