Interviewing when everyone codes with AI

For twenty years the technical interview had one job: watch you produce code under pressure. That signal is now nearly worthless — an agent produces code faster than any candidate, and interviewers know it. So the loops are being rebuilt, mostly without announcement, around the question that still separates candidates: can you be trusted with the output? If your prep is still grinding algorithm puzzles, you're rehearsing for an interview that's disappearing. Here's what's replacing it.

What actually changed in the loop

Talk to the people running hiring at AI-forward teams and a pattern emerges. The take-home that banned AI tools is gone — either the ban was unenforceable or the exercise now *requires* the tools, because working with them is the job. The whiteboard puzzle is fading for the same reason. What's growing in their place are formats built to observe judgment directly: reviewing code you didn't write, debugging a system you've never seen, defending a design decision against a pushy interviewer.

The logic is simple. When production is cheap, the scarce skills are specification, verification, and ownership — so that's what gets tested. The interview didn't get easier. It got more honest about what the job actually is.

The four formats to expect

• The AI-allowed build. Tools open, screen shared, a real task. They're not watching what you generate — they're watching what you *accept*, what you push back on, and whether you verify before you trust.
• The review. A pull request with three planted problems: one obvious, one subtle, one architectural. Seniority shows in which ones you catch and how you talk about them.
• The debug. A failing system, logs, and forty-five minutes. They want your hypothesis loop — what you check first, how you narrow, when you ask for help.
• The defense. "Walk me through something you shipped. Why that design? What broke? What would you do differently?" Vague answers end candidacies here.

A two-week prep that matches the test

Week one: practice reading, not writing. Every day, take one unfamiliar pull request — an open-source repo works fine — and review it out loud for twenty minutes: what it does, what's fragile, what you'd push back on. Then have an agent generate a solution to a medium-sized problem and find its weaknesses before running it. This feels inefficient. It is exactly the muscle the loop now measures.

Week two: rehearse the defense. Pick the two or three projects you'd bring up in interviews and write the decision record for each: the constraint, the options, the choice, the failure, the fix. Then say it out loud — ideally to a person, or to an agent playing a skeptical interviewer. The candidates who stumble here don't lack experience; they lack rehearsal narrating it.

And keep using your tools during prep the way you would in the room. Interviewers consistently say the worst sessions are candidates who either refuse the tools to prove purity, or delegate everything and can't explain the result. The strong middle — direct the work, verify it, own it — is a rhythm you build ahead of time, not on the day.

In the room

Narrate your verification. When the agent hands you code, say what you're checking and why before you accept it — that running commentary *is* the assessment. Being wrong out loud and correcting course reads far better than silent, lucky success. And when you don't know something, say so and show how you'd find out; the interviewer is deciding whether to trust you with ambiguity, and honest uncertainty is evidence in your favor.

If a loop still hands you a 2015-style puzzle with the tools banned, that's information too — about how the team actually works. It's worth one polite question: "How does the team use AI tooling day to day?" The answer tells you whether the interview is a leftover or the culture.

The quiet advantage

Here's the encouraging part: this loop is much harder to game than the old one, and that favors people who've done real work. Grinding puzzle sets could beat the whiteboard. Nothing but actual practice beats "show me how you verify," and if you've been shipping with these tools honestly, the interview is mostly asking you to do your job with the sound on. Prepare the narration, sharpen the review reflex, and walk in as the person who can stand behind the work — because that's the person they're hiring.