A vibe coding platform that survives the code review.
Vibe coding is building software by describing what you want to an AI in plain language and letting it write the code. Andrej Karpathy named it in February 2025; by 2026 it’s how a meaningful share of new software gets started. opix is the platform version of the idea: live preview, real Next.js code, your own database — and gates that check every build before you trust it.
Free welcome credits · no card required · failed builds cost $0
What vibe coding means — the precise version
On February 2, 2025, Andrej Karpathy described “a new kind of coding I call ‘vibe coding’, where you fully give in to the vibes, embrace exponentials, and forget that the code even exists.” He meant it playfully — accept every diff, paste the error messages back in, fine for “throwaway weekend projects.” The name stuck harder than the caveat.
Simon Willison pinned down the strict definition a few weeks later: vibe coding is building software with an LLM without reviewing the code it writes. If you read, test and understand the output, you’re doing AI-assisted programming — a different and older thing. The distinction matters, because most of the criticism aimed at vibe coding is really aimed at the “without reviewing” part.
The dictionaries followed the usage: Merriam-Webster listed “vibe coding” as slang and trending in March 2025, and Collins named it Word of the Year that November. By mid-2026 the term covers the whole broad practice — prompt-driven app building in natural language, where the human doesn’t read most of the generated code. Which is exactly why the platform underneath the prompt matters so much.
From weekend toy to production tooling
The adoption numbers stopped being funny fast. Stack Overflow’s 2025 survey of roughly 49,000 developers found 84% using or planning to use AI tools; JetBrains measured 85% of 24,534 developers regularly using AI coding tools the same year. And a quarter of Y Combinator’s Winter 2025 batch had codebases that were about 95% AI-generated, per YC’s own managing partner.
The tools split into two families. Editor-side agents — Cursor, GitHub Copilot, Claude Code — assist people who already program. Prompt-to-app platforms — Lovable, Bolt.new, v0, Replit Agent, and opix — build and run the whole product from a description. The platform family is where vibe coding stopped being a meme: description in, running software out, no local toolchain at all.
The vocabulary evolved with the stakes. By late 2025, Willison proposed “vibe engineering” for accountable, production-grade agent work, and by early 2026 professionals preferred “agentic engineering” for the disciplined practice — leaving “vibe coding” to mean the casual, unreviewed mode. The lesson for platforms is direct: keep the describe-it-in-plain-language part, but the platform — not the user’s discipline — has to supply the reviewing.
What separates a vibe coding platform from a toy
Four properties decide whether a vibe coding tool can carry an app past the demo — plus two that most tools skip.
A live preview you can click
Files stream into a real sandbox, the dev server runs, and a clickable preview arrives in about two minutes. You judge working software, not a mockup.
Real code, readable
Idiomatic Next.js and TypeScript in a code view that’s a real editor. If you never open it, fine — but the day you need to, it’s code, not a black box.
Your own database
Every app gets a dedicated Postgres (Neon) with auth wired in — not rows in the platform’s shared store. Your data was never ours.
A real deploy path
Publish to your-app.opix-app.com, connect a custom domain, or export the repository to GitHub and eject to your own hosting. Exit is a feature.
Honest metering
A live meter prices every build, hard spend caps bound the month, and one-time credit packs never expire. No math surprises at the invoice.
Gates, not hope
Every build is type-checked, runtime-smoke-tested and visually verified before you see it. The platform reviews so vibing stays safe.
The criticism wave is real — here’s the sourced version
The security numbers are the strongest case against blind vibe coding. Veracode’s GenAI code security research found AI-generated code introduced OWASP Top 10 vulnerabilities in 45% of its test tasks — and its Spring 2026 update reported newer models writing more functional, not more secure, code. CodeRabbit’s December 2025 analysis of 470 pull requests found 1.7x more major issues and roughly 2.7x more security vulnerabilities in AI co-authored code than in human-written code.
The incidents made it concrete. In May 2025, a security review found 170 of 1,645 sampled apps built on Lovable exposing personal data through misconfigured row-level security. Two months later, Replit’s agent deleted a customer’s production database during an explicit code freeze — the company’s CEO called it unacceptable and shipped guardrails. Both stories are canonical for a reason: unreviewed output plus real user data is where vibe coding earns the criticism.
Even the productivity claim is contested: a METR randomized trial in 2025 found experienced developers took 19% longer with early-2025 AI tools while estimating they’d been about 24% faster. And trust is falling as adoption rises — 46% of developers in the Stack Overflow survey actively distrust AI output accuracy. None of this is a reason to abandon vibe coding. It’s a spec for what a serious platform has to check before you ship.
The gates every opix build passes
The criticism wave doubles as a checklist. opix runs it on every build, automatically, before the result reaches you.
Type-check, actually run
tsc runs in the sandbox on every generation. Errors are repaired automatically and re-checked — 94% of first builds compile clean.
Runtime smoke test
Compiling isn’t working. opix loads the preview; on a server error it reads the stack, finds the suspect files and fixes them before the build reaches you.
Visual verify
A screenshot of the running app goes through a vision check for gross defects — blank screens, broken layouts, raw error text — the breakage a type-checker can’t see, repaired instead of waiting for you to point at it.
Honest failure handling
When a build still fails, it fails visibly and costs $0. Retries, auto-fixes and failed generations never burn credits, and spend caps bound the worst case.
What the gates don’t do
Gates catch classes of defects — type errors, crashes, broken screens. They don’t make unreviewed code a good idea for a banking backend, and we won’t pretend otherwise. What opix guarantees is that review stays possible: the output is standard Next.js and TypeScript that you, or a developer you hire, can open, read and change, with the whole repository exportable to GitHub. Vibe the first 90%; review the parts that touch money and personal data.
Scope honesty, too: opix is at its best on full-stack web apps — sites, dashboards, SaaS tools, booking and admin systems. It doesn’t build native mobile apps, and heavy custom backends are still better written by hand. If that’s your project, an editor-side agent or a different platform is the better tool, and we’d rather say so here than after your first build.
the Karpathy post that named vibe coding — Collins’ Word of the Year nine months later
of tasks in Veracode’s AI-code study produced OWASP Top 10 flaws — the reason gates exist
what a failed build costs on opix — the meter only runs for results you keep
Asked before you asked.
What is vibe coding?+
Vibe coding is building software by describing it to an AI in natural language and letting the AI write the code — in the strict sense, without reviewing what it writes. Andrej Karpathy coined the term in February 2025, and Collins named it Word of the Year that November. In practice the word now covers the whole prompt-to-app workflow, reviewed or not.
Is vibe coding bad?+
It’s a trade, and by 2026 the evidence on both sides is solid. It’s genuinely fast and has opened software-building to people who don’t code. It also has documented costs when output ships unreviewed: Veracode found OWASP Top 10 flaws in 45% of AI-generated code tasks, and real incidents — exposed user data, a deleted production database — trace back to unchecked builds. Vibe coding isn’t bad; shipping unverified code to production is. That’s a platform problem, and it’s fixable.
Which vibe coding platform should I use?+
Judge on five things: a live preview you can click, code you can read and export, a database you own, a real deploy path, and what happens when a build fails. Lovable, Bolt.new, v0 and Replit are the established names, each with real strengths. opix’s case: every build passes a type-check, a runtime smoke test and a visual check; failed builds cost $0; and every app is standard Next.js on its own Postgres, exportable to GitHub. If you need native mobile apps, opix isn’t your tool — Replit and Bolt build those.
Can vibe-coded apps go to production?+
Yes — plenty are there already. The documented failures, like Lovable apps exposing data through misconfigured access rules or Replit’s agent deleting a live database, were failures of verification and guardrails, not proof it can’t be done. Production readiness means auth generated correctly, a database you own and can back up, versions you can roll back, and code a human can review. That’s what opix’s gates and export path are for — and for apps handling money or personal data, we’d still tell you to have a developer read the code before launch.
What’s the difference between vibe coding and AI-assisted programming?+
Review. Simon Willison’s 2025 line still holds: building with an LLM without reviewing the code it writes is vibe coding; reading, testing and understanding the output is AI-assisted programming. opix is built so you can move between the two — describe changes when the vibes are good, open the code view when it matters.
Do I need to know how to code to use a vibe coding platform?+
No — describing, refining and publishing on opix require no code. The difference between a toy and a platform is what happens when the AI gets something wrong: on opix you point at the element or paste a screenshot, failed attempts cost nothing, and if you eventually hire a developer they inherit a normal Next.js repository, not a proprietary export.
What does vibe coding on opix cost?+
Starting is free: welcome credits, no card. Paid plans start at $29/month for 120 credits — a little above Lovable and Bolt at $25 and Replit at $20 — and one-time credit packs never expire. The difference is how failure is billed: a live meter prices every build, hard spend caps bound the month, and failed builds always cost $0.
AI app builder
The product tour: the same gates, metering and ownership applied to full-stack app building.
Lovable alternative
Vibe coding head-to-head: how opix compares with the category’s biggest name, with sourced pricing.
Why failed AI builds should cost $0
The failure-billing argument in full — why the meter should only run for results you keep.
Give in to the vibes. Keep the code.
Free welcome credits, no card. Every build passes the gates — and failed builds cost $0.