How to Prepare for an AI-Assisted Coding Interview
A few companies now run a coding round where you’re allowed, or expected, to use an AI assistant. You may hear it called an AI-assisted, AI-enabled or vibe coding interview. You’ll need practice reading code, deciding what to ask the AI to do, and checking its work. LeetCode practice alone won’t cover that.
What it is
You get a realistic programming task and an AI tool. Depending on the company, the tool is a chat assistant built into the interview editor or your own setup, such as Cursor, Copilot or Claude. The interviewer watches how you work and asks questions as you go.
The task is bigger than an algorithm question, and one prompt won’t solve it. You have to plan the work and decide what to hand to the AI.
Who runs them
In June 2025, Canva announced that backend, frontend and machine learning candidates are expected to use AI tools in technical interviews. Its AI-assisted round replaced the computer science fundamentals screen for backend and frontend roles.
Meta confirmed in July 2025 that it was testing a coding interview with an AI assistant. Candidates describe a 60-minute round in a shared editor with a built-in AI chat, which takes the place of one of the two coding rounds.
Candidates report similar rounds at other companies, but most of those companies haven’t announced anything publicly. Formats and rules differ between companies and even between rounds at the same company, so ask your recruiter which of yours are AI-assisted.
What the round looks like
The details vary, but candidate reports describe the same outline. You start from existing code: several files, some tests and a short description of what the system does. The work comes in stages. Typically you read the code, then fix a bug that a failing test points to, then add a feature. The requirements are incomplete on purpose, and you’re expected to ask about them. Throughout, the interviewer asks you to explain things, including code the AI wrote.
Most rounds run 45 to 90 minutes.
What they’re grading
A Canva interviewer described three things they assess:
- Problem solving: how you break down an unclear task, what you build first, and whether you think about edge cases.
- Ownership: whether you understand your solution, including the code the AI wrote, and can explain your design decisions.
- Working with AI: where you use it to save time and how you check the result.
How to work during the round
Read before you prompt. Spend the first few minutes on the task and the code, and say what you understand so far. Ask the questions the task leaves open, such as who uses this, what should happen with invalid input and how much data there is. Then tell the interviewer your plan before you write or generate anything.
Give the AI small, well-defined jobs: one function with a clear input and output, a test for a specific case, an explanation of a library you don’t know. Read every line that comes back. Check the logic and the edge cases, fix what’s wrong, and tell the interviewer what you changed and why. Run the tests after each change.
Explain your decisions as you go, including what you’re asking the AI to do. The interviewer needs to follow your reasoning while you’re prompting as well as while you’re coding.
Your prompts show the interviewer how well you understand the problem:
| Weak prompt | Better prompt |
|---|---|
| The whole task description pasted in | “Write a function that merges overlapping time ranges. Input is a list of start and end pairs, sorted by start.” |
| “Fix the failing test” | “This test expects an empty list for no bookings, but the function returns null. Show me where null comes from.” |
| “Make this faster” | “This loop calls the database once per user. Rewrite it to load all users in one query.” |
What fails people
The same Canva interviewer described candidates who spent too much time showing off prompting techniques and tool features. Others raced through features, accepting output as fast as it arrived, or handed the whole problem over without refining what came back.
Be ready to explain any function in your solution. If you can’t say what it does, go back and read it. Run the tests too; generated code can look right and still fail on an edge case.
Refusing to use the AI can hurt too. At Canva, using it is part of the evaluation. Ask your recruiter whether it’s required or optional.
How to prepare
Use the same kind of tool before the interview. Canva’s post says the candidates who struggled were often good programmers with little AI experience. If the company provides the assistant, ask what it can do. If you bring your own, set up your editor and an empty project in advance.
Practice on code you didn’t write. Pick a small open source project, choose an open issue, and give yourself 45 minutes to fix it with an AI assistant while explaining your decisions out loud.
Practice reviewing AI code as well. Ask an assistant to write something with a few traps in it, like a rate limiter or a cache with expiry. Then look for the bugs, the missing edge cases and anything you wouldn’t accept in a code review.
Keep practicing your fundamentals. You need them to judge whether generated code is correct or fast enough, and Canva says its round still assesses them.
Finally, ask your recruiter:
- Which rounds allow AI, and which don’t?
- Is using it required or optional?
- Do I use my own tools or one you provide?
- Which languages can I use?
- Is there a practice environment?
If there is a practice environment, spend some time in it before the interview so you’re comfortable with the editor and assistant.