AI-Assisted Coding Interview (2026): Which Companies Run Them and How to Prepare

An AI-assisted coding interview is a technical round where you are allowed, or expected, to use an AI assistant while you code. You are not graded on whether the AI can solve the problem. You are graded on how you direct it, whether you catch its mistakes, and whether you can explain and own the code that ends up on screen. As of October 2026, the clearest public examples are Meta, Canva and Brex, with Google running a pilot, and the format varies a lot by company.

At a glance

Question	Short answer
What is an AI-assisted coding interview?	A coding round with an AI assistant available (or required), graded on your engineering judgment and how you work with the AI
Which companies run one?	Meta, Canva and Brex have said so publicly. Google is piloting one. Others are reported but unconfirmed (see below)
Is it easier than a normal coding round?	Not reliably. Problems tend to be larger and more ambiguous, and you are still accountable for every line
What do interviewers grade?	Problem solving, verification, code quality, and communication, plus how you use the AI
Can I use any AI tool I like?	Only if the company says so. Canva lets you bring your own tools. Meta uses a built-in assistant
How do I prepare?	Practice with an AI assistant on multi-part problems, narrate your reasoning, and rehearse defending your code
Will I still get a no-AI round?	Often yes. At Meta and Google, reporting says one classic algorithm round without AI remains

Which companies run AI-assisted coding interviews (as of October 2026)

I only list companies where I could read a company statement or credible press report. Where the evidence is thinner, I say so.

Meta. Meta began an AI-enabled coding round in October 2025. CoderPad, the platform Meta uses, describes it as a pilot where candidates use AI tools like Meta engineers do, and quotes Meta's recruiting team: "It's not 'cheating,' it's how the job is evolving." Details are in our Meta AI-assisted coding interview guide.

Canva. In June 2025, Canva's engineering blog said it now expects backend, machine learning and frontend candidates to use AI tools such as Copilot, Cursor and Claude. Canva's post, "Yes, You Can Use AI in Our Interviews. In fact, we insist," says the new round replaces the computer science fundamentals screen for backend and frontend roles. See our Canva guide.

Brex. Brex's CTO told Semafor in September 2025 that Brex requires candidates to use AI in coding exams and to build an entire product in a short period of time. I could not find a Brex-published description, so treat the format details as limited to that interview.

Google (pilot). Business Insider reported in May 2026 on an internal Google document describing a pilot for junior and mid-level roles on select US teams, starting in the second half of 2026. Coverage syndicated on Yahoo Finance says candidates use Gemini in a "code comprehension" round where they read, debug and optimize an existing codebase, and that interviewers assess "AI fluency, including prompt engineering, output validation and debugging skills." This is one report that others re-published, and a Google spokesperson confirmed Gemini as the assistant. If you have a Google loop, ask your recruiter which format applies. Our Google prep page covers the classic rounds.

CodeSignal and HackerRank (assessment vendors). Both sell AI-assisted assessments that employers can switch on. CodeSignal's assistant is called Cosmo (guide). HackerRank has AI-assisted tests and an AI interviewer called Chakra (guide). Whether you get AI in a given assessment depends on the employer.

Reported but not verified. Shopify's pair-programming round is described by several third-party prep sites as allowing AI tools, but I found no Shopify source, and one Glassdoor reviewer reported being stopped from using an LLM on the problem itself. Candidates also post about AI-enabled rounds at Microsoft, Atlassian and others on Reddit; I could not verify those as official. Rippling, Stripe, Snap, LinkedIn and Microsoft appear in autocomplete for this topic, which tells you people are searching, not that the policy is confirmed.

Companies that prohibit it. Amazon reportedly bans generative AI use during interviews unless explicitly permitted, per Business Insider's reading of internal guidance. Cognition has said it does not allow AI in interviews either. The practical rule: never assume. If the invitation does not say AI is allowed, it is not.

What an AI-assisted round looks like

Formats differ, but three patterns repeat in the sources I read.

1. Built-in assistant in a shared editor. Meta's round runs in CoderPad with an AI chat panel. Candidates report a roughly 60-minute session on a multi-file project (reported, not official), with the assistant able to read the files but not edit them.
2. Bring your own tools. Canva tells candidates to use their preferred AI-integrated development environment and suggests setting up a basic project structure in advance.
3. Build something under time pressure. Canva's example prompt was building a control system for managing aircraft takeoffs and landings at a busy airport. Brex describes building an entire product quickly.

The problems are deliberately bigger and vaguer than a standard LeetCode question, because a single prompt should not be able to solve them. That is the point: the round tests the engineering around the AI.

What interviewers actually grade

Across Meta, Canva and Google reporting, the criteria cluster into four things.

Directing the AI. Can you break a vague requirement into pieces and give the assistant specific, well-contexted tasks? Canva's interviewer guide says to use AI for skeleton code, researching unfamiliar APIs and routine tasks, and to keep technical control. Hello Interview's write-up of Meta's round puts it bluntly: "Don't say 'fix this.'"

Verifying output. Run the code, test edge cases, and read what the AI wrote. Google's reported rubric names "output validation" explicitly. Meta's reported competencies, per Hello Interview, are problem solving, code quality, verification and communication.

Debugging. AI-generated code is often subtly wrong, especially across several files. Canva lists "identifying and fixing issues in AI-generated code" as an evaluated skill, and Google's pilot is built around reading and debugging existing code.

Explaining and owning it. Canva's interviewer guide says to be ready to discuss what you built, why you made specific choices, what you would improve with more time and how it would scale. Canva also says it expects candidates to have full ownership of all code.

Assessment vendors have formalized this. HackerRank's AI Fluency score looks at three dimensions: context quality, critical thinking and collaboration, graded A, B or C from the full interaction history. CodeSignal says employers get a transcript of the AI conversation and a session replay.

How to prepare

You are practising two separate skills: ordinary coding fundamentals, and working with an assistant. Do not drop the first. Canva says it still assesses CS fundamentals, and Meta reportedly keeps a classic no-AI round.

A 7-day plan

• Day 1: Baseline. Solve one medium DSA problem with no AI, thinking aloud. This is the skill the no-AI rounds test. Our DSA patterns sheet is a good source of problems.
• Day 2: Learn your assistant's limits. Pick one assistant and give it a small multi-function task. Note where it is confidently wrong. Canva advises practising with AI assistants beforehand to learn their strengths and limits.
• Day 3: Plan first. Take an ambiguous product prompt (a booking system, a rate limiter, a task scheduler). Write down requirements, assumptions and a sketch before touching the AI. Then use it to implement one piece at a time.
• Day 4: Verification drill. Ask the AI for a solution, then spend ten minutes trying to break it with edge cases you write yourself. Count what it missed.
• Day 5: Debug someone else's code. Take a small open-source function or an old project and find a planted bug, with and without AI. Google's pilot centers on reading and debugging existing code.
• Day 6: Full dry run. Time a 60-minute session on a multi-part problem. Narrate everything. Finish with a two-minute summary: what you built, trade-offs, what you would do next.
• Day 7: Defend it. Have someone ask follow-ups on your code: why this data structure, what breaks at 10x scale, what if this input is empty. If you cannot answer without the AI, go back to Day 3.

If you only have 3 days

Do Days 2, 4 and 7. Learn the assistant's failure modes, build a habit of testing its output, and rehearse explaining your code.

Ask the recruiter these questions

• Is AI allowed in this round, and which tool or model?
• Is there a practice environment? (Meta reportedly provides one to candidates with scheduled interviews.)
• Do I still have a no-AI coding round?
• Can I use my own IDE and tools, or only the built-in assistant?

Common mistakes

Canva's interviewer guide names three, and they match what other sources describe.

1. Blindly accepting output. Canva calls it the "feature marathon": rushing to build many features and accepting AI output without careful review. If you paste code you cannot explain, the follow-up questions will expose it.
2. Hands-off prompting. Letting the AI generate the whole solution instead of staying engaged. Meta candidates report the standard as "Don't prompt your way out of it."
3. AI showcase. Spending the round demonstrating clever prompting rather than solving the engineering problem.
4. Skipping the plan. Jumping to code without clarifying scope and stating assumptions.
5. Silence. Typing prompts without narrating. Interviewers cannot grade reasoning they cannot hear.
6. Using tools that are not allowed. If the round provides a built-in assistant, using a separate one on the side breaks the rules. Follow the stated rules, and practise under the same conditions you will face.
7. Giving up fundamentals. Candidates at Meta report that the assistant was sometimes less helpful in the live round than in practice, so you need to be able to proceed without it.

Practising with Thita, honestly

Thita's AI mock interview at /ai-interview is not an AI pair-coding simulator. It does not give you an AI assistant to direct. What it does practise is the part every AI-assisted round ends with: explaining and defending code. It is a spoken, adaptive interviewer that asks follow-up questions, runs your code against test cases, and ends with a written report. The starter plan is free.

That makes it a fit for Day 7 of the plan above. To rehearse the AI-pairing half, use a general assistant on your own problems and apply the verification habits here. For the algorithm fundamentals that the no-AI rounds still test, use the company-wise kit and the DSA patterns sheet. See also our broader AI-allowed coding interview rounds post and how to pass an AI interview.

FAQ

What is an AI-assisted coding interview?

It is a coding round where you may use an AI assistant, and interviewers evaluate how you use it. Canva describes its version as a round that tests when and how to use AI, breaking down ambiguous requirements, and fixing issues in AI-generated code.

Which companies have AI-assisted coding interviews?

Meta, Canva and Brex have said publicly that they run them, and Google is piloting one for some junior and mid-level roles. Several others are reported by candidates but I could not verify them. Always confirm with your recruiter.

How do I prepare for an AI-assisted coding interview?

Practice multi-part problems with an assistant, write a plan before prompting, test everything the AI returns, and rehearse explaining your code out loud. Also keep your no-AI fundamentals sharp.

How do I approach an AI-assisted coding interview?

Clarify requirements, state assumptions, sketch the design, then use the AI for well-scoped pieces. Review each output, run it, and narrate your decisions. Canva advises giving your own answer to a question before turning to the AI.

What are AI-assisted coding interview questions like?

Expect larger, more open-ended tasks than a classic puzzle, such as building a small system or fixing and extending a multi-file codebase. Canva's published example is an airport takeoff and landing control system. Meta candidates report maze and path-finding style problems.

Is an AI-assisted coding interview easier?

Not necessarily. The AI saves typing, but the problems are bigger, and you are expected to verify and explain everything. Meta candidates report that finishing all phases is not required to do well.

Can I use ChatGPT or Cursor in any coding interview?

Only if the company says so. Canva lets you use your preferred tools. Meta provides a built-in assistant in its environment. Amazon reportedly bans generative AI in interviews unless explicitly permitted.

Is there a mock AI-assisted coding interview I can practice with?

Some platforms offer practice formats, and Meta reportedly gives candidates a practice environment. Thita's mock interview does not simulate AI pairing; it practises explaining and defending your code under follow-up questions.

Do AI-assisted interviews replace algorithm rounds?

Usually they replace one round, not all. Canva's replaces its CS fundamentals screen for backend and frontend roles, and Meta's reportedly replaces one of two coding rounds.