What we learned hiring our first 20 engineers

Our early interviews leaned hard on algorithm puzzles, because that's what we'd seen everywhere else. In hindsight, almost none of our best engineers were the strongest at those, and a few of our worst hires were.

What actually correlated with good hires

The clearest signal turned out to be how someone handled an ambiguous, underspecified problem, closer to real client work than a leetcode-style question. We rebuilt our process around a paid trial task instead, and hiring got noticeably more predictable.

It's slower per hire. It's also the reason we haven't had a bad senior hire since we switched, which more than pays for the extra time.