Production-Grade Pair Programming

After the screen, Coinbase puts you in front of one or two pair-programming rounds, and they look like ordinary work. You code in your own editor and your own language, the interviewer shares the prompt, and your code runs against real inputs. The problem is sized to overflow the time box on purpose. You are not expected to finish it, and reaching the end with a tangled answer scores worse than stopping early on a clean, well-structured slice.

The grade comes down to whether the code looks like something you would merge. Clear components with names that say what they hold. Input validation and error handling you wrote rather than assumed. Small methods that each do one thing, and tests that prove the parts work. Coinbase is explicit that it is not hunting for the minimal optimal solution a contest would reward. The two problems below are sized and shaped the way this round runs. Each one has more in it than you can finish. Both pay off the engineer who structures first and handles the mess instead of racing to a fragile end.

The Coinbase angle

Coinbase says plainly that more candidates lose the loop in this round than in any other. An algorithm screen and a manual coding round measure different things, and people prepare for the first one. The interviewer watches how you build. They see how you name a component, where you guard against bad input, and whether you stop to write a test before piling on the next feature. Optimize for code you would be comfortable shipping. A clean half is a pass; a rushed, unmaintainable sprint to the finish is the failure mode this round is built to catch.