I've sat on both sides. I've whiteboarded system designs while an interviewer nodded silently. I've also been the one nodding, watching a candidate design a notification system on a whiteboard.
The gap between what candidates prepare and what interviewers actually evaluate is staggering.
What Candidates Prepare For
- LeetCode hard problems
- Obscure algorithm trivia
- "Tell me about a time when..."
- Memorized system design answers
What Interviewers Actually Evaluate
How you handle ambiguity. The first thing I do when given a system design problem is ask clarifying questions. "How many users? What's the latency requirement? What's the budget?" Candidates who start drawing boxes before asking questions are a red flag. They build without understanding requirements — and they'll do the same on the job.
Trade-off awareness. There's no perfect architecture. Every choice has a cost. When a candidate says "we should use Kafka for the message queue," I ask "why not SQS?" If they can articulate the trade-off (Kafka: higher throughput, more operational overhead, better replay; SQS: simpler, managed, good enough for most cases), they understand engineering. If they say "Kafka is industry standard," they're cargo culting.
Failure mode thinking. "What happens when this service goes down?" If the answer is "it won't go down," I know they've never operated a system in production. Everything goes down. The question is whether you've designed for it.
Communication clarity. Can you explain your design to a non-technical person in the room? Senior roles involve communicating with product managers, designers, and executives. If you can only explain your system to other engineers, you've hit your ceiling.
The Questions I Ask (and What I'm Really Testing)
"Walk me through a recent project you're proud of."
I'm testing: Can you tell a coherent story? Do you mention constraints, not just technology? Do you credit your team or take all the credit? Do you mention what you'd do differently?
"You're getting 500 errors in production. Walk me through your debugging process."
I'm testing: Do you have a systematic approach, or do you guess? Do you check logs and metrics first, or do you start changing code? Do you think about blast radius?
"Design a system for [X]. You have 45 minutes."
I'm testing: Do you ask questions first? Do you start with requirements or with technology? Do you mention monitoring, error handling, and scaling — or just the happy path?
What Changed When I Started Interviewing
As a candidate, I thought the interviewer wanted the "right answer." As an interviewer, I learned there is no right answer. I'm evaluating your thought process.
The candidate who designs a simple system, acknowledges its limitations, and explains when they'd add complexity is stronger than the candidate who designs a complex system they can't explain.
My Advice (From Both Sides)
For candidates:
- Ask 3-5 clarifying questions before designing anything
- Start simple and add complexity when asked
- Mention failure modes unprompted ("if this service goes down, here's what happens")
- Explain trade-offs for every major decision
- Be honest about what you don't know — "I haven't used Kafka at scale, but I understand the throughput benefits. For this use case, I'd start with SQS and migrate if we need replay"
For interviewers:
- Don't test for specific technology knowledge — test for engineering judgment
- Ask "what would you do differently?" — the best engineers have strong opinions about their own work
- Give candidates room to recover from mistakes — how they handle being wrong tells you more than getting it right
The best interviews feel like working sessions. The worst feel like interrogations. Design for the former.
