When AI began changing into mainstream, many corporations reacted by making an attempt to maintain them out of the interview course of. Notably in engineering, candidates have been warned to not use Cursor or Claude Code. Some corporations appeared for methods to detect AI-generated work or redesigned workout routines to make AI much less helpful.
I understood the intuition, however thought it was fixing the unsuitable drawback.
If somebody goes to make use of AI day-after-day after becoming a member of our firm, interviewing them with out AI tells me little or no about how they’ll truly carry out. We don’t ask accountants to work with out spreadsheets or designers to work with out design software program. AI has change into a part of the fashionable toolkit for information work. Pretending in any other case throughout an interview creates a synthetic setting that optimizes for abilities individuals will hardly ever depend on as soon as they’re employed.
So at Warp, we took the alternative strategy. Candidates are free to make use of no matter AI instruments they usually would—Cursor, Claude Code, ChatGPT, or the rest. For engineering candidates specifically, relatively than giving them a generic coding train, we ask them to construct an actual function from our product. Candidates have an hour, the identical instruments they’d use at work, and full freedom in how they strategy the issue.
What stunned me wasn’t that candidates labored sooner with AI. Everybody does. What stunned me was how a lot simpler it grew to become to differentiate distinctive candidates from merely succesful ones.
The candidates who constantly impress us aren’t the individuals writing the cleverest prompts. The truth is, they usually spend much less time interacting with AI than different candidates. They start by taking the time to know the issue itself. They ask considerate questions, make clear constraints, determine edge circumstances, and construct a psychological mannequin earlier than they begin executing. Solely then do they use AI to speed up the work.
Different candidates take the alternative strategy. They instantly start prompting, settle for AI-generated output too rapidly, and solely later understand they misunderstood the underlying drawback or have generated a subpar reply. AI lets them produce work rapidly, but it surely additionally lets them transfer confidently within the unsuitable path.
That’s why I don’t assume “immediate engineering” is changing into the defining talent of the AI period, however judgment is.
AI has dramatically lowered the price of execution. It has not lowered the price of logic: deciding what ought to be constructed, figuring out the precise strategy, balancing trade-offs, evaluating whether or not a solution is appropriate, or recognizing when an elegant-looking answer breaks in the true world. If something, these talents have change into extra priceless as a result of getting the precise work accomplished is not the bottleneck.
This realization has additionally modified how we take into consideration constructing an organization. Warp is in the midst of doubling the dimensions of our crew. Just a few years in the past, fast progress usually meant maximizing headcount as rapidly as attainable. We thought extra individuals meant extra output, and most leaders would agree.
AI modifications that. We’re seeing that one distinctive worker right this moment can accomplish dramatically greater than they may even a yr in the past. The profile of the individuals you hire is altering. We’re in search of individuals who study rapidly, thrive in ambiguity, and know the right way to direct AI as an alternative of merely consuming its output. Simply as vital, we’re pairing them with individuals who have spent years growing deep experience. In our case, this implies deep familiarity with areas like payroll, tax compliance, advantages, and HR. AI dramatically will increase the velocity of execution, but it surely doesn’t exchange hard-earned area information. The mix of these two varieties of experience is what permits us to maneuver rapidly with out sacrificing high quality.
I believe many corporations are nonetheless hiring for a model of information work that’s already disappearing. Their interview processes have been designed for a world the place execution was the scarce talent. More and more, it isn’t. The scarce talent is knowing issues effectively sufficient to make use of AI as leverage relatively than as a crutch.
That’s why we’ve stopped making an attempt to find out whether or not candidates can succeed with out AI. It’s not the query that issues. The query we’re making an attempt to reply is far less complicated: when everybody has entry to the identical more and more highly effective instruments, who constantly workout routines one of the best judgment?
To date, that’s turned out to be one of the best predictor of success we’ve discovered.
