I've interviewed 200+ engineers in the last 3 years. The ones who get rejected aren't bad at code. They can't explain a tradeoff, own a decision, or debug a system they didn't build. AI made this problem 10x worse, not better.

Everyone's optimizing for LeetCode. Meanwhile, the actual job is on fire and nobody told you.
Here's the thing: I've been on both sides of the table at companies ranging from 50-person startups to Stripe. The engineers who wash out aren't the ones who can't write a binary search. They're the ones who freeze when PagerDuty fires at 2am and the runbook is three years out of date.
AI didn't create this problem. It just made it impossible to hide from.
In 2023, we hired a strong leetcode performer. Passed every round. Great Datadog dashboards explanation in the system design interview. Six months in, we had a cascading failure in our payments pipeline. $340k in delayed settlements. He had touched that code. He couldn't explain what it did. Couldn't form a hypothesis. Just stared at the traces waiting for someone else to talk.
We let him go four months later. Not because he was dumb. Because debugging a system you didn't build requires a skill nobody taught him to practice.

Don't do this: treat system design as a performance. Interviewers who've run real postmortems can smell a rehearsed answer in thirty seconds.
Do this instead. Pick one outage, real or from a public postmortem like the Cloudflare 2023 BGP incident. Understand it cold. Know the timeline, the blast radius, the fix, and the thing they should've done six months earlier. Walk into interviews able to talk about failure like someone who's lived inside one.
The engineers AI is replacing wrote code. The ones keeping their jobs own systems. There's a difference and it's not subtle.