ZipHQ Software Engineer Interview Experience — AI-Assisted Onsite Coding on an Expense Reimbursement Codebase

It was roughly three small problems. They gave us a codebase, but it had so many files that there was no way to read through all of it, so basically you have the AI read and write everything. It was a simple system for managing company expense reimbursements, with features like monthly expense and budget reports. The first problem was that the monthly report data was wrong, for example someone filed a reimbursement under February, but it was actually for a January bill. We had to debug it. I had the AI find 2 bugs and fix them. They followed up asking whether I thought the AI's fix was correct; I looked at it and thought it made sense. The second problem was adding a new feature to the report. I just had the AI write the whole thing, and it passed the tests. The third problem was: suppose you were going to move this to production, brainstorm some new features and then have the AI write them. Basically the whole time was just having the AI read the codebase and write the code. The final follow-up was that they picked a few functions and asked whether I understood what the AI-written code did, and how I would optimize it.