We let job candidates use AI. It made hiring better

Warp, a software company, allows job candidates to use AI tools during interviews, finding that it improves hiring by revealing candidates' judgment and problem-solving skills rather than just their ability to prompt AI. The company argues that AI has reduced the cost of execution but increased the value of good judgment, and that hiring should focus on candidates who can direct AI effectively and combine it with deep domain expertise.

When AI https://www.fastcompany.com/section/artificial-intelligence started becoming mainstream, many companies reacted by trying to keep them out of the interview process. Particularly in engineering, candidates were warned not to use Cursor or Claude Code. Some companies looked for ways to detect AI-generated work or redesigned exercises to make AI less useful. I understood the instinct, but thought it was solving the wrong problem. If someone is going to use AI every day after joining our company, interviewing them without AI tells me very little about how they’ll actually perform. We don’t ask accountants to work without spreadsheets or designers to work without design software. AI has become part of the modern toolkit for knowledge work. Pretending otherwise during an interview creates an artificial environment that optimizes for skills people will rarely rely on once they’re hired. So at Warp http://warp.co , we took the opposite approach. Candidates are free to use whatever AI tools they normally would—Cursor, Claude Code, ChatGPT, or anything else. For engineering candidates in particular, rather than giving them a generic coding exercise, we ask them to build a real feature from our product. Candidates have an hour, the same tools they’d use at work, and complete freedom in how they approach the problem. What surprised me wasn’t that candidates worked faster with AI. Everyone does. What surprised me was how much easier it became to distinguish exceptional candidates from merely capable ones. The candidates who consistently impress us aren’t the people writing the cleverest prompts. In fact, they often spend less time interacting with AI than other candidates. They begin by taking the time to understand the problem itself. They ask thoughtful questions, clarify constraints, identify edge cases, and build a mental model before they start executing. Only then do they use AI to accelerate the work. Other candidates take the opposite approach. They immediately begin prompting, accept AI-generated output too quickly, and only later realize they misunderstood the underlying problem or have generated a subpar answer. AI lets them produce work quickly, but it also lets them move confidently in the wrong direction. That’s why I don’t think “prompt engineering” is becoming the defining skill of the AI era, but judgment is. AI has dramatically reduced the cost of execution. It has not reduced the cost of good judgment: deciding what should be built, identifying the right approach, balancing trade-offs, evaluating whether an answer is correct, or recognizing when an elegant-looking solution breaks in the real world. If anything, those abilities have become more valuable because getting the actual work done is no longer the bottleneck. This realization has also changed how we think about building a company. Warp is in the middle of doubling the size of our team. A few years ago, rapid growth often meant maximizing headcount as quickly as possible. We thought more people meant more output, and most leaders would agree. AI changes that. We’re seeing that one exceptional employee today can accomplish dramatically more than they could even a year ago. The profile of the people you hire https://www.fastcompany.com/section/hiring is changing. We’re looking for people who learn quickly, thrive in ambiguity, and know how to direct AI instead of simply consuming its output. Just as important, we’re pairing them with people who have spent years developing deep expertise. In our case, this means deep familiarity with areas like payroll, tax compliance, benefits, and HR. AI dramatically increases the speed of execution, but it doesn’t replace hard-earned domain knowledge. The combination of those two kinds of expertise is what allows us to move quickly without sacrificing quality. I suspect many companies are still hiring for a version of knowledge work that is already disappearing. Their interview processes were designed for a world where execution was the scarce skill. Increasingly, it isn’t. The scarce skill is understanding problems well enough to use AI as leverage rather than as a crutch. That’s why we’ve stopped trying to determine whether candidates can succeed without AI. It’s no longer the question that matters. The question we’re trying to answer is much simpler: when everyone has access to the same increasingly powerful tools, who consistently exercises the best judgment? So far, that’s turned out to be the best predictor of success we’ve found.