“I don't run a fully algorithmic loop, but I do bias toward signal over theater. At Eightfold I used two-degree validation for senior sourcing: a strong employee plus their strongest referral from the same company pre-validates culture, so the interview can focus on tech fit.”
“I’ve interviewed more than 300 engineers, and the ones who thrived were rarely the ones who memorized algorithms best. They could articulate their thinking, admit what they didn’t know, and show they cared about the team’s success, not just their own code.”
“I used to run interviews like a hazing ritual, whiteboard reverse linked lists and all, until I realized I was mostly testing who had time to practice LeetCode and who could perform under a power trip. Now I design every section around the actual job, with a rubric written before we ever see a resume, and I have pulled interviewers off panels for enjoying the judge dynamic too much.”
“I stopped doing live coding screens a while ago. The signal I actually wanted was how someone thinks through ambiguity, so I moved most loops to a paid take-home project and a structured debrief where I ask them to explain their own tradeoffs.”
“I once phone-screened five candidates in 24 hours, and the speed made me notice how easily culture fit becomes a proxy for comfort. I now treat structured rubrics as the main signal, because the loudest bias in hiring is the one the interviewer thinks they don't have.”
“I stopped doing LeetCode-style screens at ProSights. We give candidates a real problem, let them use any tool they want, and watch how they break it down and recover when it breaks.”
“I think the fairest coding interviews test the actual work the person will do, not how fast they can solve a puzzle. At Rosie and the companies where I ran loops, I pushed for take-homes scoped to our stack and behavioral questions tied to our values, because a candidate nervous in a live whiteboard can still be the engineer you want.”
“I rebuilt Productboard’s hiring loops after a failed North America expansion, and the thing I got wrong first was over-indexing on polished performance in live coding. We now bias toward work-sample problems candidates can reason through aloud, with the same rubric and interviewer calibration, because that is where we actually remove noise.”
“I’ve run hiring loops at ZZAZZ and sat through enough interviews to know the fairest coding round is one where the candidate can show how they actually work, not how well they rehearse a puzzle. When AI is allowed, I watch the prompt log and the reasoning, not just the final code. The signal is in what they challenge, accept, and explain out loud.”