The full loop has a new format now — five rounds, and I got the rejection letter a week later.
Round one, behavioral (BQ). Went pretty well — the usual how-do-you-resolve-conflict stuff, plus some talk about impactful projects, especially AI projects.
Round two, AI coding: a card game — find three cards that sum to 15. Split into three steps: find the bug, implement the algorithm, optimize the algorithm. I spent a bit too long on finding the bug, so for the last two steps I mostly gave the direction and let the AI write the code. I finished it in the end and answered all the follow-ups.
Round three, AI coding: maximize unique chars from a word list. Also three steps. This interviewer wanted me to optimize toward the direction he preferred. He asked a lot of questions throughout and didn't leave me any time to actually use the AI, so I ended up writing everything by hand.
Round four, system design: design an online game leaderboard. I did well on this one since I'd just reviewed this exact question not long before.
Round five, AI-enabled system design. This is the new format — it tests agentic system design. My question was to design an agent that automatically triages tickets. I'd spent a week beforehand looking at agent design, and I'd built a similar agent system at work, so I thought I could nail it. I'd previously built a triage agent -> investigate agent -> resolve agent pattern myself. But the interviewer seemed to be steering me toward a ReAct-loop direction instead. A lot of my answers ended up not matching what he was looking for.
Anyway, moving on.
Discussion
Loading comments…