Updated by OpenAI candidates
Software Engineer (New Grad) Interview Experience
I expected the AI safety interview to be super guarded, but the engineer was actually really open about his views on regulation and what AI safety should mean. It ended up feeling more like a lunch conversation with a very opinionated engineer than a normal behavioral.
Interview process
I applied for a new grad software engineer role, did the OA, and then got nudged forward by a recruiter after I let my original link expire. Because I was slow to schedule, the interview slots were really backed up, so my live rounds did not happen until January. The process itself was pretty short once it started: one easy live coding round, then a final that was a coding interview plus a behavioral across two consecutive days. Everything was virtual and honestly a lot less intimidating than the in-person finance interviews I had done before. I got the offer in February, but I already had another offer lined up, so I stopped at team matching and never actually met a hiring manager.
- Online assessment
- Technical interview
- Final round
Interview tips
I would just prep pretty normally for coding screens up front, because the earlier rounds were the more algorithm-heavy ones. For the final coding round, I would not over-index on LeetCode tricks because mine was mostly basic OOP and whether I could build clean classes and keep state straight. On the behavioral side, I would actually be ready to have opinions, especially if you get someone from a more mission-heavy part of the company. That round went best when it stopped sounding rehearsed and turned into a real conversation.
Company culture
They seemed pretty interested in moving candidates fast once they cared, because the recruiter personally reached out when my OA link expired and sent me a new one. At the same time, scheduling could get backed up a lot if you were not quick, because by the time I replied most of the availability had slipped way out. The engineers I talked to were pretty direct: one basically skipped intros and went straight to code, and another clearly had a behavioral checklist but was still willing to turn it into a genuine discussion. At least for this new grad process, it felt like they were screening for general SWE ability first and only doing team fit later, since I got to offer and team matching without meeting a hiring manager.
Questions asked
Overview
My final was two back-to-back interviews across consecutive days: one coding round and one behavioral. The coding one was way more object-oriented than algorithmic, and the behavioral one ended up being the most interesting part of the whole process. I expected the behavioral to be pretty guarded because the interviewer was from the AI safety side, but it turned into a very open conversation.
Question types asked
Specific questions asked
Add support for a command like 'meet <name>' so the system tracks that the person is in a meeting.
Add support for an 'out <name>' command so the system responds that the person is out of the office.
Add a taco emoji feature that counts tacos sent in the thread and lets you query that count.
This felt like they were testing OOP way more than algorithms. I built classes for the chat interface and message handling, and the interviewer kept layering on functions. The main idea was to keep track of message state, then add commands like meeting status and out-of-office status, and finally a taco counter you could query later. I remember thinking this round was mostly about whether I could structure code cleanly with classes, inheritance, and polymorphism, not whether I knew some tricky algorithm.
What's one thing on your resume you want to talk about?
I picked a trading project and talked through that. I kind of yapped a lot, so he didn't really need many follow-ups. After that he started sharing some of his own background, including stories about mining Bitcoin in college, so it felt less like a strict interrogation and more like a real conversation.
What are your opinions on AI safety?
I answered that a big part of my interest was the AI safety side, and that basically opened up the whole rest of the interview. I started asking him questions too, and he was way more open than I expected. We got into OpenAI's mission, what AI safety should even mean, and how regulation fits in. It stopped feeling like a checkbox behavioral and felt more like the kind of conversation I'd have over lunch with technical friends.
Get full access with a membership, or share your experience to try it free.
