In this article
What vibe coding is
What is vibe coding?
Vibe coding is making software by describing what you want in plain language and steering an LLM with follow-up prompts instead of reading every line. Wikipedia says it "may involve accepting AI-generated code without thorough review of the output". You describe what you want, the model writes code, you run it, and you feed errors back as prompts. Vibe coding skips the review step.
Text description of the diagram
- You describe what you want, the model writes code, you run it, and you feed errors back as prompts.
- Vibe coding skips the review step.
Why it is asked: it checks that you can define the term precisely instead of using it as a synonym for any AI help. Source: Wikipedia, Vibe coding.
Who coined vibe coding, and what did he mean?
Andrej Karpathy coined the term in early February 2025. Wikipedia reports that he described a form of coding where you "fully give in to the vibes, embrace exponentials, and forget that the code even exists". He later said that in practice he rarely goes fully vibe: he more often still looks at the code and adds complexity slowly (O'Reilly Radar, May 2025).
Why it is asked: knowing the origin shows you understand the original, casual framing and not only the later hype. Source: O'Reilly Radar, Vibing at home.
Is all AI-assisted coding vibe coding?
No. AI-assisted programming is using tools such as large language models to help write code, and vibe coding is "the less responsible subset of this", in Simon Willison's tag page wording. Willison's boundary, in his March 2025 post: if the model wrote every line but you reviewed, tested and understood all of it, that is using it as a typing assistant, not vibe coding.
Why it is asked: it separates candidates who can draw the line between assisted and unreviewed work. Source: Simon Willison, AI-assisted programming.
Why do interviewers ask vibe coding questions?
They are usually a way to probe judgement about AI tools, not trivia about a buzzword. A strong answer shows that you use AI to move faster, that you still own what ships, and that you know where unreviewed code breaks. Expect the topic to sit beside ordinary coding questions, so keep the fundamentals in the coding interview guide fresh.
Why it is asked: it tests whether your habits with AI tools are safe for a team. Read the full explanation: coding interview guide.
When vibe coding fits and when it does not
When is vibe coding appropriate?
Vibe coding suits prototypes and throwaway projects, where being wrong costs little. Karpathy framed it for throwaway weekend projects (Wikipedia). Willison draws the other end firmly: "Vibe coding your way to a production codebase is clearly a terrible idea" (Simon Willison, March 2025). A good answer names the stakes, who maintains the code and what a failure would cost.
Why it is asked: it tests whether you match the method to the risk. Read the full explanation: Simon Willison, Will the future of software development run on vibes?.
What are the risks of vibe coding in production?
Wikipedia summarises the critics' concerns as a lack of accountability, poor maintainability and a higher risk of security vulnerabilities, and notes that it can make code harder to maintain over time, leading to technical debt. Willison puts it briefly: "Where vibe coding fails is in producing maintainable code for production settings" (March 2025).
Why it is asked: interviewers want you to name concrete failure modes, not just say "it is risky". Source: Wikipedia, Vibe coding.
How do you take accountability for AI-written code?
Treat the code as yours. Willison writes that "if you're going to put your name to it you need to be confident that you understand how and why it works" (March 2025), ideally well enough to explain it to someone else. In practice: read the diff, run it, write or run tests, and be ready to explain every line in review.
Why it is asked: ownership is what separates an engineer using AI from a person pasting output. Source: Simon Willison, March 2025.
How would you review AI-generated code in an interview?
Verify in a fixed order: restate what the code should do, read it for the idea, run it on a normal input, then try edge cases such as empty input, a single element and repeated values. Say each check aloud and fix what you find. A strong answer shows verification and clear communication, not only a working result.
Why it is asked: it shows whether you check output or trust it. Read the full explanation: coding interview guide.
What would a prompt-and-verify walk-through look like?
This is an illustrative scenario. You ask an assistant for the longest substring without repeating characters. Before accepting the answer, you trace it: an empty string should return zero, a string of one repeated character should return one, the right edge should advance on every character, and the left edge should jump past the earlier copy only when a repeat appears; for "abba" the answer is 2. If a case fails, you say so and correct it. The method is explained in longest substring without repeating characters.
Why it is asked: it shows you can test code you did not write. Read the full explanation: longest substring without repeating characters.
Does AI make developers faster?
Not always. In a July 2025 randomised controlled trial (RCT) of experienced open-source developers working on their own repositories, those allowed to use early-2025 AI tools took 19% longer. They had expected a 24% speedup, and afterwards still believed AI had sped them up by 20%. It is one study of early-2025 tools, so treat it as a caution, not a law. The study's authors, Model Evaluation and Threat Research (METR), also note that in February 2026 they published new data on the productivity impact of late-2025 AI tools, so the 2025 result is not the latest state.
Why it is asked: it tests whether you rely on evidence instead of feelings about productivity. Source: Measuring the Impact of Early-2025 AI.
What did Andrew Ng say about the term vibe coding?
In June 2025 Andrew Ng objected that the term misleads people into assuming software engineers just "go with the vibes" when they use AI tools to build applications (Wikipedia). A useful answer agrees that skilled AI-assisted work involves design, review and testing, and that "vibe coding" describes only the unreviewed end of the range.
Why it is asked: it checks that you can discuss a contested term fairly. Source: Wikipedia, Vibe coding.
Are real companies shipping AI-generated code?
Some report heavy use. In March 2025, Y Combinator reported that 25% of startup companies in its Winter 2025 batch had codebases that were 95% AI-generated (Wikipedia). That is a share of one batch, relayed second-hand, and it does not say the code was unreviewed. Use it as context, not as proof that vibe coding is safe.
Why it is asked: it tests whether you can cite a number with its limits. Source: Wikipedia, Vibe coding.
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 →AI-enabled coding rounds
Does Meta let candidates use AI in coding interviews?
Meta told employees it would allow some coding candidates to use an AI assistant during the interview, according to internal communications seen by 404 Media in July 2025. A Meta spokesperson confirmed to Business Insider that "we're testing how to provide these tools to applicants during interviews". Meta calls it an AI-enabled interview.
Why it is asked: candidates want to know which companies allow AI, and this is the best-documented case. Source: 404 Media, July 2025.
What does an AI-enabled coding interview look like?
Meta has not published a format or rubric for it. What is public is that candidates in some rounds have an AI assistant available: 404 Media reported this from internal Meta communications, and Meta confirmed to Business Insider that it is testing these tools for applicants. Details such as duration, tooling and how it fits the loop are not confirmed, so ask your recruiter.
Why it is asked: knowing what is confirmed and what is not helps you plan without relying on rumour. Source: 404 Media, July 2025.
How is an AI-enabled round scored?
Meta has not published a rubric, so no one outside can say exactly how it is scored. As advice, a strong answer shows planning before prompting, reading and testing what comes back, and clear narration of your choices. Prompting speed on its own shows little.
Why it is asked: it tells you that judgement, not prompt speed, is the skill worth showing. Related: coding interview guide.
What are OpenAI vibe coding interview questions?
No public primary source describes OpenAI's questions, so this page does not claim any. Candidate-report pages exist but are low-reliability. Formats vary by company, team and role, and they change. Ask your recruiter whether AI tools are allowed and in what form, and prepare the same fundamentals: reading, testing and explaining code, as in the coding interview guide.
Why it is asked: candidates search for lists tied to one employer, and the honest answer is to rely on the recruiter's rules. Read the full explanation: coding interview guide.
Can you use an AI tool in any coding interview?
No. Use an assistant only where the interviewer allows it, and only the one provided in the environment. Follow your recruiter's instructions over any general advice. Policies differ between companies and can change, so confirm the rules before the round.
Why it is asked: it tests whether you respect the stated rules of a process. Related: coding interview guide.
Follow-up questions
How should I practise for a coding round with an AI assistant?
Practise the routine, not the prompt: state the problem, plan, let the assistant draft, then read, run and test the result aloud. Rehearse on classic problems such as merge intervals so you can spot a wrong answer quickly.
Do I still need to know algorithms if an AI writes the code?
Yes, because you must judge whether the output is correct and efficient. Knowing patterns such as the prefix-sum and hash-map solution or the heap approach to closest points lets you spot a slow or wrong draft.
How do I check AI-written code in a language I use less?
Read it slowly, run small inputs and compare against what you know the result should be. Brush up on idioms first, for example with Python interview questions, so unfamiliar syntax does not hide a bug.
- Vibe coding means prompting, accepting code and feeding errors back, with little or no review.
- AI-assisted programming is the wider category; reviewing and understanding every line moves you out of vibe coding.
- It fits throwaway projects, not production code that someone must maintain.
- One randomised controlled trial found experienced developers 19% slower with early-2025 AI tools, while believing they were faster.
- In AI-enabled rounds, show planning, verification and clear communication; Meta has published no rubric, so follow your recruiter's rules.
Sources
- Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity — METR, 10 Jul 2025
- Vibe coding (Wikipedia) — Wikipedia, 26 Oct 2026
- Will the future of software development run on vibes? — Simon Willison's Weblog, 6 Mar 2025
- Simon Willison on ai-assisted-programming (tag page) — Simon Willison's Weblog, 26 Oct 2026
- Vibing at home — O'Reilly Radar (Mike Loukides), 13 May 2025
- Meta Is Going to Let Job Candidates Use AI During Coding Tests — 404 Media, 29 Jul 2025
- Meta will let some job candidates use AI in their coding interviews — Business Insider (via AOL), 29 Jul 2025
