AI coding interviews: what they are and how to prepare
By Shilo Drot, software engineer (ex-Meta). Updated .
An AI coding interview is a technical interview where AI is part of the round. Most often it means an AI-assisted coding interview: you solve a realistic engineering task with an AI assistant at hand, and you're assessed on how you direct it, check what it gives you and explain your decisions - not on whether you can recall an algorithm.
This guide covers both meanings of the term, which companies run these rounds, how a round usually goes, what interviewers look for, the mistakes that sink candidates, and a practice plan.
Two kinds of AI coding interview
The same phrase is used for two different things:
- AI-assisted (AI-enabled, AI-allowed) rounds. A human interviewer, and you may use an AI assistant while you code. This is the format Canva and Meta have introduced, and the one most candidates mean when they search for it.
- Interviews run by an AI interviewer. An AI gives you the problem, watches you work and asks follow-up questions. Companies use it to screen candidates at scale, and candidates use it to practice.
The skills overlap: in both, you're judged on how you think out loud, not only on the final code. A Cla64.ai practice session combines them - an AI interviewer that asks why, and an AI assistant you can use while you work.
Which companies run AI-assisted coding interviews?
Canva replaced its computer science fundamentals screen with an AI-assisted coding round in June 2025, and expects candidates to use tools such as Copilot, Cursor and Claude (Canva Engineering).
Meta has reportedly made one of its onsite coding rounds AI-enabled: about 60 minutes in CoderPad, on a multi-file project with existing code and tests, with an AI chat panel. Candidates describe it moving from fixing issues in the existing code, to implementing a new feature, to extending the system. Using the AI is optional, but you have to explain any code it writes for you (IGotAnOffer, interviewing.io). See Meta's AI-enabled coding round in detail.
Google is reported to be piloting a similar code comprehension round with Gemini for some junior and mid-level roles, and LinkedIn to run one in CoderPad where using the assistant is optional (Design Gurus, from candidate reports).
Not every company allows it. Anthropic asks candidates not to use AI in live interviews unless told otherwise (Fortune). Formats change quickly and vary by team and level, so ask your recruiter exactly which tools are allowed and what the round looks like.
How an AI-assisted coding round works
Company formats differ, but the reported ones share a shape. A typical 60-minute round:
- Orientation (about 5 minutes). You get a small project you haven't seen - several files, some tests - and a short description of the task.
- Fix what's broken. A bug or failing test in the existing code. This checks that you can read code you didn't write and find the real cause.
- Add a feature. A new requirement on top of the code. This is where the AI is most useful, and where interviewers watch whether you check what it writes.
- Extend or improve. Larger inputs, edge cases, or a design change - with questions about the trade-offs you made.
Candidates don't always reach every stage. What's assessed is the quality of the work you did and how you reasoned about it.
What interviewers evaluate when AI is allowed
- Code comprehension - finding where behavior lives in code you didn't write.
- Breaking down the task - turning a vague request into small, checkable steps.
- Verification - running tests before and after a change, and catching mistakes in AI output.
- Judgment - choosing a design and explaining the trade-off you made.
- Communication - saying what you're doing and why, the way you would with a teammate.
The AI doesn't lower the bar; it moves it. Typing speed and syntax recall matter less. Whether you stay in control of the code matters more.
AI-assisted vs. traditional coding interviews
| Traditional (LeetCode-style) | AI-assisted | |
|---|---|---|
| Starting point | An empty function and a problem statement | A realistic task: a codebase with bugs, a feature to add, or a blank project |
| AI tools | Not allowed | Allowed or expected |
| Length | Usually 45 minutes, one or two problems | Usually about 60 minutes, one project in stages |
| Main skill tested | Algorithm recall and implementation | Directing and verifying AI, debugging, designing, testing and explaining |
| Red flag | Not finding the optimal approach | Accepting AI output you can't explain |
Common mistakes in AI coding interviews
- Paste, accept, run, hope. Handing the AI the whole task and running whatever comes back. When it fails, you're debugging code you don't understand, on the clock.
- Not reading the existing code. Asking the AI to fix a bug before you know where the behavior lives. It will often "fix" the wrong place.
- Trusting a green test run. The existing tests passing doesn't mean the bug is gone - write a test that reproduces it first.
- Working in silence. The interviewer can't credit reasoning they can't hear. Say what you asked the AI for and whether you agree with the answer.
- Refusing to use the AI at all. Where it's allowed, using it well for syntax, APIs and boilerplate is part of what's assessed.
How to prepare for an AI coding interview
- Practice realistic tasks. Mix all three kinds: a bug in code you didn't write, a feature to add to an existing project, and something new built from a blank project.
- Run the tests first. Know where you stand before you change anything, and after every change.
- Give the AI small, clear instructions. Plan the step yourself, then ask for one piece at a time. Use it freely for syntax, an API or a library call.
- Verify everything it tells you. Read generated code line by line and test it before you rely on it.
- Explain out loud. Practice with someone, or something, that asks why.
- Keep your fundamentals. You still need to spot a slow approach or a wrong data structure when the AI suggests one.
- Ask your recruiter about the format. Which tools, which models, which language, and whether the AI can edit files or only chat.
The interview guide covers each of these habits in more detail, with what to do before, during and at the end of the interview. To put them into practice, try an AI mock interview.
Practicing AI coding interviews on Cla64.ai
Each Cla64.ai interview opens a real project in a cloud workspace in your browser, with a terminal. The AI in the session plays two roles: an interviewer you can ask about the task's rules, and an assistant that answers syntax, API and tooling questions directly and gives a hint when you're stuck. It won't write the solution, and it may ask why you chose an approach. When you're done, it asks you a few questions about your work. Exercises with starting code are then checked with hidden tests; blank-project exercises end with a live demo. Either way you get a written review, including how you used the AI.
Sessions cover all three kinds of task - debugging existing code, building on starting code, and designing something new from a blank project - in Python, JavaScript, TypeScript, Java and C++. One exercise is free without an account, and the weekly challenge is free with one.
Start with a debugging exercise from the exercise list, or watch a full session first.
Frequently asked questions
What is an AI coding interview?
An AI coding interview is a technical interview where AI is part of the round. Usually that means you may use an AI assistant while you solve a realistic engineering task, and you're assessed on how you direct it, check its output and explain your decisions. The term is also used for interviews run by an AI interviewer instead of a person.
Is an AI-assisted coding interview the same as an AI-enabled one?
Yes. AI-assisted, AI-enabled and AI-allowed are different companies' names for the same idea: a coding round where you may use an AI assistant, and the interviewer watches how you use it.
Which companies allow AI in coding interviews?
Canva has said publicly that it expects candidates to use AI in one of its engineering rounds. Meta reportedly made one of its onsite coding rounds AI-enabled, and Google is reported to be piloting a similar round. Others, such as Anthropic, ask candidates not to use AI in live interviews. Policies change often, so ask your recruiter.
How long is an AI-assisted coding interview?
Most reported formats are about 60 minutes, usually in a multi-file project with an AI chat panel next to the editor. Some candidates report up to 90 minutes. Your recruiter can confirm the length for your round.
What do interviewers evaluate when AI is allowed?
Whether you break an ambiguous task into steps, read and understand existing code, check AI output instead of accepting it blindly, test before you trust a change, and explain your decisions clearly. Accepting generated code you can't explain is usually treated as a red flag.
Can I use ChatGPT or Copilot in a coding interview?
Only if the company says so. In an AI-assisted round you use the tool the company provides, or the ones it names. In any other interview, using AI without permission is usually treated as cheating. If you're not sure, ask before the interview.
How should I practice for an AI coding interview?
Practice on realistic tasks rather than isolated puzzles: fixing a bug in code you didn't write, extending an existing project, and building something new from a blank project. Use AI for what it's good at, verify what it tells you, test as you go and explain your decisions out loud. A mock interview with an interviewer who asks why is the closest you can get to the real thing.
Is LeetCode still useful if AI is allowed?
For algorithm fundamentals, yes - you still need to recognize when an approach is too slow. But LeetCode is built around short algorithm puzzles, while AI-assisted rounds test different skills: understanding a codebase, debugging, designing and building, testing, and explaining your decisions while working with AI. Practice those directly.
Does Cla64.ai let me use AI?
Yes - the session's own assistant. Some companies, Meta among them, let candidates use AI to generate code in the interview; on Cla64.ai the AI won't solve the task or decide the logic, but it can write what you spell out. It answers syntax, library and tooling questions directly and gives a hint when you're stuck, and other AI tools aren't available - so the decisions, and explaining them, stay yours.