Lovable PromptsLovable Prompts
    Updated 6 sessions measured, 30.4 creditsDefinitions re-read 23 Oct 2026

    What is vibe coding? The definition, the record, and what it cost

    Vibe coding is building software by describing it to an AI in plain language and accepting what it writes. Collins made it Word of the Year 2025 and defines it as “the use of artificial intelligence prompted by natural language to write computer code”. Simon Willison draws the stricter line: building software with an LLM without reviewing the code it writes. I have done it 6 times on this site, each session measured in credits; the 6 cost 30.4 in total, and the studies on what goes wrong are below.

    By Marco Kohns · Sources re-read 23 October 2026

    2025Word of the Year

    Collins Dictionary, 6 November 2025.

    1.7xmore issues

    In AI-written pull requests than human ones, across 470 PRs (CodeRabbit, 17 December 2025).

    30.4credits

    For 6 measured Lovable sessions on this site, 11 questions asked first.

    The term

    Where did the term vibe coding come from?

    Andrej Karpathy named it in February 2025; within nine months it was a dictionary word of the year.

    Vibe coding timeline from the sources read 23 October 2026
    WhenWhat
    Feb 2025Karpathy names it
    Mar 2025Merriam-Webster: trending
    Mar 2025Willison narrows it
    Nov 2025Collins Word of the Year
    Dec 2025CodeRabbit measures it

    Karpathy's post, as Willison quotes it, describes giving in to the vibes and forgetting “that the code even exists”: accept every change, let the code grow past what you would read. Willison's post of 19 March 2025 keeps the term for exactly that and calls reviewed, understood AI-written code software development. Lovable's own definition is broader: “Vibe coding is about building software without worrying about syntax.”

    Measured

    What does a vibe coding session cost in Lovable?

    6 sessions on this site, credits read from the balance before and after. 3 of them asked questions before writing code.

    Every measured Lovable session on this site: date, credits, questions first, result
    SessionDateCreditsQuestionsResult
    Landing page + opt-in31 Jul 20261.80One round, all five priority features, inline success state, Cloud enabled
    Empty states + skeletons26 Sep 20269.52Asked 2 questions, built a /states gallery and a task list, verified slow load, failure and retry
    SaaS MVP with auth26 Sep 202610.44Asked 4 questions, enabled Cloud, self-tested sign-up and invites, roles checked in the database
    SEO review, no code26 Sep 20261.105 of 6 checks passing; the sitemap deferred until publish
    Installable PWA26 Sep 20263.10Manifest, icons, service worker, Add to Home Screen; verified in the preview
    SaaS landing page26 Sep 20264.55Five questions first, six minutes to a preview; server-rendered, FAQ schema, one CTA
    Lovable project chat right after a prompt was sent, showing a question card asking who will use the app and what its screens show, with an answer field, Skip all and Next
    A session that does not run on vibes alone: Lovable asking before it builds, 26 September 2026.

    The record

    What goes wrong with vibe-coded software?

    Measured, more than human code: more issues, and more security issues.

    CodeRabbit: AI-written pull requests compared with human ones, 470 PRs, 17 December 2025
    CategoryAI against human
    Issues overall1.7x
    Logic and correctness1.75x
    Security2.74x
    Readability3x

    October 2025 update

    Veracode

    “AI-generated code introduced risky security flaws in 45% of tests.” Tested on more than 100 language models across Java, JavaScript, Python and C#. Report.

    30 May 2025

    CVE-2025-48757

    Missing row-level security in Lovable-generated apps, rated 9.3. The security guide has the record and the check.

    10.83 against 6.45

    Issues per pull request

    CodeRabbit counted 320 AI and 150 human pull requests. Report.

    Judgement

    When should you vibe code, and when should you review every line?

    Vibe code what you can throw away; review what holds people's data or money.

    Vibe code

    Good fits

    • A prototype to show someone
    • A landing page that tests an offer
    • An internal tool with no outside users
    • A one-off script you will delete

    Review

    Read the code

    • Sign-in, roles and anything per user
    • Payments and the webhook
    • Data other people trust you with
    • Anything you will maintain for a year

    The middle path is what the library prompts encode: describe the outcome in plain language, and write down the rules a reviewer would check, row-level security, states, a Done When list Lovable can verify. That keeps the speed and moves the review into the prompt. The prompting guide shows the sections.

    Starting

    How do you start vibe coding without losing control of the code?

    Five habits from the sessions above and Lovable's own guides.

    1. 01

      Write the brief first

      Who it is for, the one job, the channel. Lovable's guide calls a good knowledge file a one-page brief.
    2. 02

      Let it ask

      End the prompt by asking for questions; 3 of my 6 sessions drew them before any code.
    3. 03

      One change per prompt

      Check each change in the preview before the next one, as Lovable's idea-to-app guide puts it.
    4. 04

      Put the rules in the prompt

      Row-level security, states and Done When checks: the review written down before the build.
    5. 05

      Scan before you publish

      The Security view's scans are free; the chat's security review is billed.

    Vibe code with the rules written in

    Library prompts carry states, access rules and checks Lovable can run, so a fast build is also a reviewable one.

    Disclosure: if you create a Lovable account through links on this page, we may earn a commission. It costs you nothing and does not change what we recommend. Create a free Lovable account

    Frequently asked

    What is vibe coding?

    Building software by describing it to an AI in plain language and accepting what it writes. Collins, which made it Word of the Year 2025, defines it as the use of artificial intelligence prompted by natural language to write computer code.

    Who coined vibe coding?

    Andrej Karpathy, in February 2025. Merriam-Webster listed it as trending in March 2025, Simon Willison narrowed its meaning the same month, and Collins named it Word of the Year in November 2025.

    Is vibe coding safe?

    For prototypes, mostly. For anything with users' data, review it: CodeRabbit found 2.74x more security issues in AI-written pull requests, Veracode found risky flaws in 45% of AI code tests, and a CVE rated 9.3 covers missing row-level security in generated Lovable apps.

    What does vibe coding cost in Lovable?

    My 6 measured sessions cost 30.4 credits in total, from 1.1 for an SEO review to 10.4 for a SaaS foundation with sign-in and roles.

    Is vibe coding the same as AI-assisted programming?

    Simon Willison separates them: vibe coding is building with an LLM without reviewing the code it writes. If you review and understand the code, he calls it software development with an AI helper.

    Do you need to know how to code to vibe code?

    Not to start: Lovable and similar builders turn plain language into a working app. To ship something other people rely on, someone has to check what the code does with data and money, either by reading it or by writing those rules into the prompt and testing them.

    Which tools are used for vibe coding?

    Prompt-to-app builders such as Lovable, Bolt, v0 and Replit, and coding agents inside an editor such as Cursor or Claude Code. The builders suit people who do not read code; the editors suit people who do. The comparisons on this site set them side by side.

    What does Lovable say vibe coding is?

    Lovable's own blog, published 3 March 2025, defines it as building software without worrying about syntax. That is the broadest of the three definitions on this page; Collins and Willison describe the same practice from the dictionary's and the developer's side.

    What we did not publish

    • A study of my own: the CodeRabbit and Veracode figures are theirs, read from their reports.
    • A session in Bolt, Cursor or Replit.
    • A review of the code my sessions produced, line by line.

    Sources

    1. Simon Willison, 19 March 2025, re-read 23 October 2026
    2. Collins Word of the Year 2025, 6 November 2025, re-read 23 October 2026
    3. CodeRabbit: AI vs human code, 17 December 2025, re-read 23 October 2026
    4. Veracode 2025 GenAI code security report, re-read 23 October 2026
    5. Lovable blog: What is vibe coding, 3 March 2025, re-read 23 October 2026
    6. Wikipedia: Vibe coding for the timeline

    Written by

    Marco Kohns

    Marco Kohns

    Founder of ProtoBites - Venture Growth Studio

    Growth PM at a Silicon Valley scale-up (a16z and General Catalyst backed), ex-Techstars where he consulted 13 early-stage startups, Reforge-trained. Every prompt on this site comes out of shipping ProtoBites' own portfolio products.

    WebsiteLinkedInX