Short answer: yes. Vibe coding works. A year after Andrej Karpathy named it, the receipts are everywhere, paying users, real revenue, apps in production. The longer answer is more useful: it works brilliantly for some things, badly for others, and the gap between the two is what this post is about.
Below: what "works" actually means, eight real apps built with AI tools (with what they do and what they were built with), where vibe coding still falls down, and how to tell if your idea is a good fit.
What "works" actually means
Skeptics and believers usually argue past each other because they mean different things by "work". Three useful definitions:
- Ships at all. The app gets built and deployed, by someone who otherwise couldn't have built it. This is the easiest bar, and vibe coding clears it constantly.
- Holds up in production. Real users hit it, it doesn't fall over, data doesn't leak, the bills don't explode. Harder bar. Clears it often, with the right platform and a bit of care.
- Makes money or saves time. Paying customers, internal time saved, measurable outcomes. The hardest bar, and the one where the case studies below actually land.
By all three definitions, vibe coding is working in 2026. Not for every kind of software, but for a wide and growing slice.
8 real apps built with vibe coding
These are publicly documented examples from 2025 and 2026, founders posting on X, LinkedIn, and the tools' own showcases. They're not the only ones, just a useful cross-section.
1. Cluely (Roy Lee)
Roy Lee's "cheat on everything" AI assistant went viral in 2025 and reportedly crossed $7M ARR within months of launch. Lee has openly credited AI-assisted development tools for the speed of iteration. Whether you love or hate the product, the build velocity is the point: a small team shipping a real desktop app with AI in the loop.
2. Base44 (acquired by Wix)
Maor Shlomo built Base44, an AI app builder, largely solo. Wix acquired it in mid-2025 for a reported $80M. The whole story is a poster child for what AI-leveraged development looks like at the small-team end of the market.
3. PromptPal, internal tools, and "weekend SaaS"
On Lovable's own showcase and across Product Hunt, dozens of small SaaS products, prompt managers, niche CRMs, course platforms, indie analytics dashboards, were built with Lovable, Bolt.new, v0, or Replit Agent by solo founders. Not unicorns, but real products with real MRR.
4. Internal dashboards at startups
The least glamorous, most common win. Ops teams build their own admin panels, lead scorers, refund tools, and reporting dashboards in a weekend instead of waiting two quarters for engineering. Public case studies from Vercel (v0) and Lovable both highlight this category.
5. Landing pages and marketing sites
Almost every YC batch since W25 has founders shipping their marketing site with v0, Lovable, or Bolt before they hire a designer. It's no longer remarkable, it's the default. The pages convert because they're shipped in a day, not a month.
6. Replit Agent apps
Replit's own data (shared publicly through 2025) showed millions of apps generated by Replit Agent, with a measurable share running production workloads. Internal tools, Telegram bots, scrapers, mini CRMs, the long tail of "software a small business actually needs".
7. Indie iOS and web games
Game jams in 2025 and 2026 (including the Vibe Jam) produced hundreds of playable browser games built mostly through AI prompting. Most are toys. Some have real audiences and Patreon revenue.
8. Side projects that became companies
The most common path: someone vibes-codes a tool for themselves, posts it, gets users, charges money. Examples from the Lovable community alone include niche schedulers, fitness trackers, language tutors, and AI-powered content tools, all shipped without traditional engineering hires.
The common thread isn't the tool. It's the shape of the problem: a clear user, a narrow scope, and a willingness to ship version 1 in a week instead of version 3 in six months.
Where vibe coding still doesn't work
To be honest about it, here's what isn't on the list above, and why:
- Heavy backend systems. Distributed systems, low-latency infrastructure, payments engines. AI can scaffold, but the design decisions need real expertise. Vibe coding here is a recipe for incidents.
- Anything regulated. Healthcare records, financial brokerage, anything covered by HIPAA, PCI, or SOC 2 at scale. The platform safety nets aren't enough; you need audited code paths.
- Large codebases. Once a codebase is 100k+ lines, vibe coding gets messier. AI-assisted coding still helps, but pure "describe and ship" breaks down.
- Anything where being wrong is dangerous. Medical advice, legal decisions, safety-critical control loops. Don't.
Notice the pattern: vibe coding works exactly where most software actually lives, small apps, internal tools, MVPs, marketing sites, niche SaaS. It struggles at the extremes most founders never touch anyway.
Why it works (now)
Three things changed in 2025 that made the difference:
- Models got good at full-stack output. Claude Sonnet 4, GPT-5 class models, and their successors stopped hallucinating imports and started shipping working components.
- Platforms got opinionated. Tools like Lovable bake in auth, database, RLS, and deploy. You can't easily mis-configure your way into a leak.
- Hosting got cheap and instant. Edge runtimes, managed databases, free tiers that scale. The "but how do I host this?" tax is basically gone.
How to tell if your idea is a good fit
Quick gut-check before you spend a weekend on it:
- Can a real user explain what the app does in one sentence?
- Is the data model simple enough to draw on a napkin?
- Is the worst case if the app breaks an annoyed user, not a lawsuit?
- Would v1 fit in fewer than 10 screens?
Four yes-es and you're in the sweet spot. If you're getting mostly nos, vibe coding can still help you prototype, but plan for a real engineering phase after.
The honest verdict
Vibe coding works, for the kind of software most people actually need. It doesn't replace senior engineers building distributed systems, and it never claimed to. What it does is collapse the time from "I have an idea" to "real users are paying for it" from months to days. That's not a future promise. It's already happening, in public, right now.
If you're on the fence, the cheapest thing you can do is try it. Here's the 7-step guide to going from idea to live app, no code required.
FAQ
Are the apps people build with vibe coding actually good?
The best ones are indistinguishable from traditionally built apps. The worst ones are buggy demos that never made it to users. The variance is wide; the median is rising fast as platforms and models improve.
Can vibe-coded apps handle real traffic?
Yes, if you build on a platform that handles hosting, scaling, and database for you. Apps built on Lovable, Vercel, and Replit have served millions of requests without their builders writing a line of infra code.
What's the biggest reason vibe-coded apps fail?
Almost never the code. Usually the same reasons any startup fails: no real user need, poor distribution, founder loses interest. The most common technical mistakes are here.
Is vibe coding the same as no-code?
No. Here's the full comparison, but the short version: no-code locks you into a visual builder; vibe coding generates real code you can own and extend.
What if I'm worried about the risks?
Fair, and worth reading. This post on whether vibe coding is bad goes through the honest tradeoffs.
