In this article
Before you start
Why interviewers ask this
A round that allows an AI assistant is one stage of a wider interview process, and it tests judgment, because the tool can produce plausible code in seconds. Canva says it now expects backend, machine learning and frontend candidates to use AI tools in technical interviews, and it lists what it looks for: knowing when and how to use AI, breaking down ambiguous requirements, sound technical decisions, fixing AI-generated code and meeting production standards.
Candidates report that Meta's AI-enabled round is scored on the same four competencies as its other coding rounds: problem solving, code quality, verification and communication. Google is reported to be piloting a format, starting with select junior and mid-level roles in the United States, in which interviewers evaluate "AI fluency", including prompt engineering and debugging (as of 2026, per an internal document seen by Business Insider). A weak answer treats the assistant as a solver. A strong answer treats it as a fast junior colleague whose work you check.
AI-allowed coding rounds from first principles
- AI-allowed round
- An AI-allowed round is a coding interview in which the employer states, before or at the start of the session, that you may use an AI assistant, and in which the interviewer judges how you use that assistant as well as the code you deliver.
- Prompt-review-run loop
- The prompt-review-run loop is a five-beat working rhythm for an AI-allowed round: you prompt for one small piece of code, review what comes back, run it, confirm it does what you meant, and move on to the next piece.
In plain words
An AI coding assistant is a chat tool that writes or explains code from a written instruction. Think of a calculator in a math exam where the examiner reads your working. The calculator does the arithmetic. You still choose the method, check the answer and explain it. The analogy breaks in one place: a calculator is always right, and an assistant is sometimes confidently wrong. So the skill on show is not typing prompts. It is directing a fast helper and catching its mistakes. In one precise sentence, an AI-allowed round scores your decisions about the assistant's output as much as the output itself.
Step 1: Confirm the rule and the tools
The rule for AI use belongs to the round, so confirm it before you open any tool. Silence in the instructions means ask the recruiter. Anthropic's candidate guidance says "This is all you–no AI assistance unless we indicate otherwise." and promises to be clear when AI is allowed. Policy also differs inside one employer: candidates report that Meta keeps one classic no-AI coding round next to its AI-enabled round, and Meta's careers page says only tools inside the interview environment are authorized. Once you know the rule, practice: Canva highly recommends practicing with AI tools before the interview, and Meta points candidates to a practice environment.
| Employer | What the source says | Source |
|---|---|---|
| Anthropic | No AI in live interviews or assessments unless told otherwise | Employer guidance, July 2025 |
| Canva | AI tools expected in technical interviews | Employer blog, June 2025 |
| Meta | Assistant built into the provided editor, no outside tools | Careers page and candidate reports |
| Reported pilot for some junior and mid-level roles in the United States | Press report of an internal document |
Step 2: Prompt small, then read and run every answer
The prompt-review-run loop keeps AI use visible and checked. Before the first prompt, say your plan and ask about requirements; Canva reports that its most successful candidates asked clarifying questions and used AI for well-defined subtasks while keeping control of the overall solution. Then work in the rhythm Hello Interview says Meta's interviewers want to see. The loop is prompt, review, run, confirm, move on.
Ask for one function, not the whole program. Read the reply, run it on a small test, compare the result with what you expected, and only then continue. Canva says the most successful candidates did not just prompt AI and accept whatever it generated. Hello Interview lists asking the assistant how to solve the whole problem as a weak pattern, so bring the approach yourself and use the assistant for the pieces.
Step 3: Say your choices aloud and know when to skip the AI
Narration is how the interviewer sees your judgment. Say why you asked for this piece, what you checked and what you changed. Meta's internal guidance, as quoted by Hello Interview, includes the lines "Explain the output. Test before using." Silence while the assistant works gives the interviewer nothing to score. Skipping the assistant entirely is also a risk: Canva says candidates with minimal AI experience often struggled because they lacked the judgment to guide it. The middle path is to use the assistant where it saves typing and do the thinking yourself. Hello Interview suggests debugging on your own in the first phase when the interviewer leaves the choice open. Broader habits live on the interview skills page.
Step 4: Own the final code and protect your time
Whatever is on the screen at the end is yours. Canva writes that it expects engineers to "take full ownership of any code they produce", whether they wrote it or an assistant did, so be ready to defend every line. Finish a correct core before you optimize. Hello Interview says not completing every phase is fine and reports several candidates who missed the last phase still received offers, which is anecdotal.
Candidates also report that the in-interview assistant can be less helpful than the practice one, so keep your core algorithm skills sharp. In the last minutes, reread the final code from top to bottom, run your tests once more, and prepare a two-sentence summary of what the program does and what you would improve next.
Step-by-step description
- Ask the recruiter which tools are allowed.
- Prompt small, then review and run every answer.
- Explain what you asked for and what you changed.
- Defend every line that is on screen.
- Silence in the instructions means ask the recruiter.
- The loop is prompt, review, run, confirm, move on.
Worked example: one loop on a maze task
The task is illustrative: find the number of moves from the start to the exit in a small maze, using breadth-first search. You say your plan, then prompt: "Write bfs(grid) that returns the number of moves from S to E, or -1 if there is no path." In this illustrative example the first reply returns the number of cells on the path, which gives 9 on the test maze where the correct answer is 8.1 You catch it by running the test, not by trusting the reply. You say what was wrong, ask for moves instead of cells, and confirm the test now passes.
from collections import deque
def bfs(grid):
rows, cols = len(grid), len(grid[0])
start = next((r, c) for r in range(rows) for c in range(cols) if grid[r][c] == "S")
queue, seen = deque([(start, 0)]), {start}
while queue:
(r, c), moves = queue.popleft()
if grid[r][c] == "E":
return moves
for dr, dc in ((1, 0), (-1, 0), (0, 1), (0, -1)):
nr, nc = r + dr, c + dc
if 0 <= nr < rows and 0 <= nc < cols and grid[nr][nc] != "#" and (nr, nc) not in seen:
seen.add((nr, nc))
queue.append(((nr, nc), moves + 1))
return -1The queue holds each cell with the number of moves taken to reach it. A cell goes into seen when it is queued, so no cell is queued twice. Say that aloud: it shows you understand the code you kept.
Worked example: a sixty-minute budget
Hello Interview reports that Meta's round runs 60 minutes in a shared editor, that the interviewer spends five to six minutes orienting you, and that, after the bug-fixing phase, roughly 30 to 40 minutes remain for the implementation and optimization phases. An illustrative split follows from that.1 In this illustrative budget, orientation takes 5 minutes, reading the code takes 7 and bug fixing takes 18.
Core build takes 18 minutes and optimization takes 12 minutes. That totals 60 minutes, with half the hour, 30 minutes, left for building. The 30 minutes spent before building leave 30 minutes for the core build and optimization, which matches Hello Interview's estimate of roughly 30 minutes of effective working time for implementation and optimization. Reaching the build phase on time matters more than polishing earlier phases.
Text description of the diagram
- In this illustrative budget, orientation takes 5 minutes, reading the code takes 7 and bug fixing takes 18.
- Core build takes 18 minutes and optimization takes 12 minutes.
Knowing it is half the job. Saying it is the other half.
Interview Clozo is an AI interview copilot for real interview rounds: it hears the question and puts a strong answer on your screen, and you say it in your own words. See the features or how it works.
Start Free Trial →The Staff-level answer
How would you approach a coding round where an AI assistant is allowed?
First I confirm with the recruiter that AI is allowed and which tools, because the rule differs by round. Then I restate the problem, ask about edge cases and say my plan before I prompt anything. I give the assistant small, specific tasks, such as one helper function, then read the output, run a test and explain what I kept. If it is wrong, I say why and fix it myself or reprompt with the failing case. I keep ownership of the design, so I can defend every line on screen. If time is short, I finish a correct core before I optimize.
What makes it Staff-level is that the answer names a verification habit and a time trade-off, not a list of tool features.
The rubric
A strong answer directs the assistant on every dimension below; this is advice on what to show, not a published company rubric.
| Dimension | Weak answer | Strong answer | Staff answer |
|---|---|---|---|
| Rule awareness | Assumes AI is fine or banned without asking | Confirms the rule and tools with the recruiter | Knows the rule can differ between rounds of one loop |
| Planning | Prompts with the whole problem at once | States an approach and asks requirement questions first | Splits the task and names which pieces suit the assistant |
| Prompting | Vague, large requests | Small, specific prompts with inputs and outputs | Adds the failing case to the next prompt and knows when to stop prompting |
| Verification | Pastes output and moves on | Reads the output, runs a test, compares with expected | Writes the test first and states what it does not cover |
| Communication | Silent while the assistant works | Narrates choices and explains the output | Explains trade-offs and what was rejected and why |
| Ownership | Cannot explain parts of the code | Defends each line | Defends the design and names the remaining risks |
Three variants of this question
What if the instructions do not say whether AI is allowed?
Ask the recruiter in writing before the round. Anthropic says it will be clear when AI is allowed, so treat silence as not allowed until someone confirms.
What if the assistant gives a wrong or weaker answer?
Say what is wrong, then fix it yourself or reprompt with the failing test. Candidates report the live assistant can be less helpful than the practice one, so a correct fix by hand is a good result.
What if I run out of time before finishing?
Make sure the core works and explain what you would do next. Hello Interview reports that candidates who did not finish every phase still received offers, though that is anecdotal.
Common mistakes
- "The assistant wrote it, so it should pass." (wrong) Unrun output is the error Canva's best candidates avoided. (right) "I'll run a small test before I keep this."
- "How should I solve this problem?" typed straight into the assistant. (wrong) It hands over the reasoning. (right) "My approach is a breadth-first search; I'll ask for the queue loop."
- "Every company has one AI policy." (wrong) Anthropic bans AI in live rounds unless stated, while Canva expects it. (right) "Can you confirm the rule for this round?"
- Using a favorite outside tool. (wrong) Meta says only in-environment tools are authorized. (right) "Which assistant is built into the editor?"
- Prompting in silence. (wrong) The interviewer cannot score your judgment. (right) "I'm asking for the helper only, then I'll test it."
Follow-up questions
Can you use AI in a coding interview?
Only when the employer says so for that round. Anthropic says no AI assistance unless it indicates otherwise, while Canva expects candidates to use AI tools, so ask the recruiter.
Does Meta allow AI in coding interviews?
Candidates report an AI-enabled coding round that started rolling out in October 2025, and Meta's careers page describes a built-in assistant with no outside tools. For how that round is graded, read what an AI-assisted coding interview is.
Do you have to finish every phase?
No source says you must. Hello Interview reports offers after unfinished last phases, but it is anecdotal, so aim for a correct, explained core.
How do you practice for an AI-enabled coding round?
Practice in the same editor and assistant when the employer provides one, and rehearse the loop on small tasks. The AI-allowed build exercise covers a longer format.
Can you use AI to prepare for an interview that bans it?
Anthropic's guidance welcomes AI for preparation such as research and practicing answers, while banning it during the assessment; see the Anthropic interview process.
What about take-home coding tasks?
Anthropic asks candidates to complete take-homes without Claude unless told otherwise, so check the instructions. The online assessment page covers the format, and job interview tips covers what interviewers look for overall.
- Confirm the AI rule for each round; silence means ask, not assume.
- Plan aloud, then prompt for one small piece at a time.
- Prompt, review, run, confirm, move on.
- Explain the output and keep ownership of the final code.
- Finish a correct core before you optimize.
Sources
- How to collaborate with Claude during our hiring process (Guidance on Candidates' AI Usage) — Anthropic; accessed 2026-10-26, 10 Jul 2025
- Yes, You Can Use AI in Our Interviews. In fact, we insist — Canva Engineering Blog (Simon Newton); accessed 2026-10-26, 11 Jun 2025
- Meta Jobs Hiring Process — Meta Careers; accessed 2026-10-26, 26 Oct 2026
- Meta's AI-Enabled Coding Interview: How to Prepare — Hello Interview (Evan King), candidate reports; accessed 2026-10-26, 26 Oct 2026
- Google is testing a new rule to transform job interviews — Entrepreneur, reporting Business Insider; accessed 2026-10-26, 26 Oct 2026
