AI Is Splitting Coding Interviews Into Two Completely Different Formats

I’ve been reading a lot of 2025–2026 SWE interview reports lately, and it feels like technical interviews are moving in two completely opposite directions.

Direction #1: Make interviews harder to use AI in

More proctoring. More screen monitoring. More identity verification. And in some cases, more in-person interviews.

Google, Cisco, and McKinsey are among the companies that have increased in-person interviewing partly because of concerns around AI-assisted cheating and candidate fraud.

Direction #2: Just let candidates use AI

This one is more interesting.

Meta has piloted AI-enabled coding interviews through CoderPad.

CoderPad itself now has built-in access to GPT, Claude, Gemini, and Llama for interviews.

Instead of asking:

“Can you write this code without help?”

the question starts becoming:

“Can you use AI without blindly trusting it?”

That means reading generated code, catching mistakes, testing it, explaining the tradeoffs, and knowing when the model is taking you in the wrong direction.

Anthropic has an even stranger version of this problem.

They publicly wrote that they had to redesign one of their performance-engineering take-homes multiple times because newer Claude models kept getting too good at it. Under the same time constraints, Opus 4.5 eventually matched their strongest human candidates.

So now you basically have two kinds of interviews emerging:

AI-free rounds: prove that you still understand the fundamentals.

AI-assisted rounds: prove that you can supervise AI like an engineer instead of using it like autocomplete.

And I think this also explains another pattern I keep seeing in recent interview experiences:

• more debugging
• more existing-codebase work
• more “implement this feature” questions
• more project deep dives
• more emphasis on tradeoffs and verification

Traditional LeetCode obviously isn’t disappearing, but memorizing 300 patterns feels less defensible as the entire interview strategy than it did a few years ago.

My takeaway for 2026 prep would be:

Practice coding both with and without AI.

If AI writes something, make sure you can explain every important decision and verify that it actually works.

And spend more time on debugging, code reading, and talking deeply about projects you actually built.

It’s interesting to compare companies because the interview formats are becoming much less standardized.

Curious what people think: If engineers are expected to use AI every day at work, should coding interviews allow AI too?