It's actually been a while since I interviewed, but I was recently looking back on my job switch this year and it came to mind. It was a pretty distinctive interview experience, so I'm writing it down to give back to the forum.
I'm an MLE. Every other company I interviewed with leaned toward models; this was the only one that leaned toward agents, and the title was very on-trend: "AI Engineer". Their Cortex Code seems to have a decent reputation in the industry, it's growing fast, and the stock has been shooting up, so when I was choosing between offers I did waver a little.
Before the onsite there was one perfectly normal coding round (just a normal LeetCode problem; I honestly can't remember which one anymore) and one perfectly normal ML Fundamentals round (they call it ML Expertise). Then came their signature onsite.
First, it's a genuinely in-person onsite (of all the non-startup companies I interviewed with, they were the only one doing it in person). You have to go to their office. You get a QR code ahead of time and can park right in the parking garage across the street, which is pretty convenient.
Second, they have a distinctive "AI Augmented coding" round. You're asked to bring your own laptop and you can preinstall any AI you want. I've been on double Max plans for Codex and Claude for ages, so this is my comfort zone. Later, though, it turned out you could also use the interviewer's laptop, and after thinking about it I decided there was no need to waste my own tokens, so I used the interviewer's.
Now for the actual interview. The onsite was three rounds in total:
The "AI Augmented coding" round I just mentioned. There seem to be two question types, "refactor and debug" and "design and build", because the recruiter first told me it would be the first type and then changed it to the second. The question was the classic design and implement a rate limiter. It's probably been covered in lots of places. You can have the AI write the code, but you have to clearly explain which algorithm you chose, why you chose it, what other algorithms there are, what the tradeoffs are, and so on. Then, with the AI-written code, you have to show live how you communicate with the AI, how you review the code, and how you test it. Overall, this round should be pretty easy and enjoyable for people who are AI-native in their day-to-day work.
Another "Expertise" round. This one is a bit like a system design round tailored to the team's actual work (Snowflake hires directly by team, not the general-hire kind). All the questions I got were about agents. First they asked how to design a pipeline that can automatically extract/distill agent skills for enterprise customers, and then they asked about the common components of agent memory (there's a video on something similar).
Behavior. Just a chat with the Hiring Manager, a normal behavioral round.
The most, most, MOST annoying thing about them is the reference check before they send the offer. You have to provide two contacts, who can be former coworkers or a former boss. I gave them a coworker who already knew I was job hunting, plus one who had already jumped ship before me. Otherwise, letting my current boss know I was getting ready to bail would have been kind of awkward.
Overall their onsite is pretty distinctive and keeps up with the times. The money was pretty generous too (but they don't do formula-based refreshers, so each year's refresh depends on your boss's mood lol). The product looks like it's growing well, but I heard the culture is really toxic, and on top of that I'd still rather do model-related work, so I didn't take it. Folks who lean toward agents should still go interview with them~
Discussion
Loading comments…