If AI can write code now, what are coding interviews actually testing?

Genuine question: With AI coding tools getting better, why are coding interviews still so focused on solving problems under pressure without much real-world context?

I get that companies still need to test fundamentals. But in actual work, engineers now use AI tools, docs, search, linters, tests, code review, and existing codebases. The job is increasingly about understanding messy requirements, debugging, evaluating tradeoffs, and shipping reliable systems.

So what should coding interviews actually test in 2026?

• Classic LeetCode under time pressure?
• Practical coding with tests?
• Debugging broken code?
• Code review?
• System design follow-ups?
• AI-assisted coding, but with harder evaluation?
• Something else?

I’m not saying coding interviews should disappear. I’m just wondering whether the format is starting to feel outdated compared with how software engineering actually works now.

Curious what people are seeing in recent interviews.