Which companies let you use AI in coding interviews?

Before you grind another problem set, find out which of three regimes your target company runs. The AI Interview Index, a public list anyone can edit and revert, sorts employers into rounds with no AI, loops with at least one AI-allowed round, and interviews built entirely around AI. The practical move is to confirm the regime during the recruiter screen, then rehearse the version you will actually face: hand-writing code under observation, or directing and defending generated code.

Ask the recruiter which regime you are in

Treat the index as a set of questions to ask, not as proof of policy, because the wiki is crowd-edited and every change is public and revertible. An entry can go stale, so ask directly which rounds permit AI, whether the IDE ships with AI switched off, and whether you will be asked to explain tool output.
Concrete anchors help you phrase it. Applied Compute runs two 45-minute coding interviews with AI switched off in the IDE. Cognition's first step is a self-serve CoderPad assessment of roughly 1-1.5 hours with no AI allowed. Anthropic keeps live interviews AI-free unless stated, though some take-homes explicitly allow it. If the recruiter cannot say, assume AI is off by default and prepare for that.

Some loops split AI and non-AI into separate rounds

A single yes-or-no answer often misses the loop's structure. Datadog keeps live coding AI-free by default but gives some roles a designated AI-assisted round. Longlake runs one round that allows AI and a separate one that does not, treating them as two different signals. PostHog splits it into a paid project day with AI and a debugging round without it. Prepare a fast hand-coding mode and an AI-leveraged mode, because you may be scored on each.

When AI is allowed, the grading moves to defense

Canva states that AI tools are permitted and that you are expected to defend what they produced. Warp gives you an hour to build a real feature, any agent fair game. DigitalOcean replaced the phone screen and algorithm onsite with a three-hour build using any AI. The skill under review becomes reviewing, testing and explaining generated code: say why it works, where it is fragile, and what you would change if the constraints moved.

Rebuilt questions mean memorized patterns stop paying

Coinbase rebuilt its questions after the old ones proved solvable with AI switched on. When AI is permitted, a familiar LeetCode shape is no longer the point, so expect problems you reason through live. That does not retire the traditional format: OpenEvidence still runs a one-hour LeetCode-style coding interview after the recruiter call, and Ramp keeps a full-stack coding screen plus graph algorithm and HTTP questions with no AI. Rogo lets no round in the engineering loop use AI, and Cursor has no non-AI coding round in the loop at all.

AI-native loops reward shipping over trivia

GitLab's handbook says AI use is encouraged in the process. Zapier applies an AI-fluency bar from application through the executive interview. Automattic skips whiteboards for a take-home with your own tools followed by a paid trial. AngelList has candidates work in a real codebase and ship a pull request, and Sierra replaced coding and algorithm rounds with an AI-native onsite. Here, portfolio evidence and code review carry more weight than a timed puzzle.
If you are collecting AI interview questions to drill, tag each one with the regime it belongs to: hand-written, AI-allowed, or AI-native. The same prompt that helps in a Warp-style feature build is worthless in a Cognition CoderPad screen, and a polished take-home can be undercut by a no-AI debugging round. Match the practice to the loop, and confirm the rules before you pour hours into the wrong one.