Real Vibe Coding Apps (What Indies Built in 2026)

The 2026 record for vibe coding apps is no longer the demo video; it is the ledger. WizardGenie ships browser games. Sorceress Code ships internal tools. Cursor, Lovable, Bolt, and Replit ship the rest. The wins underneath every one of those platform names are real software products with real customers paying real money: a multiplayer flight simulator at $138K MRR, a healthcare staffing platform at $1M ARR, a fashion AI dashboard at $800K ARR, a Brazilian education product that grossed $3M in 48 hours. This is the honest tour of what indie developers actually built in 2025 and 2026, verified against the live record on May 13, 2026, with the pattern under the wins called out so the next builder has the playbook in hand.

What “real vibe coding apps” actually means in 2026

The phrase vibe coding apps covers a wider category in 2026 than it did when Andrej Karpathy coined the term in February 2025. The original meaning was the loose, agentic, low-discipline workflow where the developer described what they wanted, the model wrote the code, and the developer steered with vibes rather than line-by-line authorship. Eighteen months later, the working definition has broadened: vibe coding now describes any project whose author spent the bulk of their time describing intent and reviewing diffs rather than typing the implementation. The apps that came out of that workflow span browser games, mobile consumer apps, healthcare staffing platforms, fashion AI dashboards, and education products. The defining property is not zero human code — almost every shipped product had a review pass at some point — but a fundamentally agentic authoring loop.

The category also got more honest in 2026. The first wave of vibe coding apps headlines in early 2025 were dominated by demo videos and toy projects. The second wave is dominated by actual revenue numbers and actual customer counts. The six case studies in this tour are picked because every one of them has independently reported revenue or user counts, an actual deployed URL, and a credible build trail. The throwaway prototypes that flooded ProductHunt in April 2025 are not on this list; the products that shipped, kept shipping, and built durable revenue are. That filter is what separates “vibe coding apps” the meme from vibe coding apps the business category.

1. fly.pieter.com — the flight sim that proved the loop

Pieter Levels built fly.pieter.com, a free-to-play multiplayer 3D browser-based flight simulator, in roughly three hours on February 22, 2025, despite never having shipped a game before. The build used Cursor as the agentic IDE, Claude as the underlying model, and Three.js for the 3D rendering layer. The game launched February 23, hit $38,360 in monthly recurring revenue by day ten, $87,000 by day seventeen, and roughly $138,000 per month by November 2025 — about seventy percent of Levels’ total income at the peak.

The economics are worth tracing because they are the lookalike pattern most other shipped vibe coding apps follow. Monetization combines three streams: in-game advertising blimps that brands rent for roughly $5,000 per month each, premium plane skins that unlock at $29.99 a one-time purchase, and branded 3D objects that companies pay to embed in the world. None of those streams require a payment processor more complex than Stripe Checkout, and the multiplayer infrastructure runs on a single browser-served WebSocket loop with no dedicated server fleet. The player count peaked between twenty-two thousand and twenty-six thousand simultaneous flyers, with 320,000 total users and 74 million impressions logged in the first month. The lesson is not that anyone can ship a viral flight sim in three hours; the lesson is that an agentic IDE plus a real-time 3D library plus a follow-the-money monetization design can clear half a million dollars a year on a hand-coded weekend project, which was structurally impossible before the agent loop matured.

2. WW2 Dogfight Arena — zero human-written code, twenty euros to ship

Nicolas Zullo built WW2 Dogfight Arena, a multiplayer World War II dogfight simulator playable at fly.zullo.fun, with a hard constraint: no human-written code at all. The 20-hour build used Cursor as the IDE, Claude Sonnet 3.7 in thinking mode for code generation, and Grok 3 for scope-and-plan work. Total prompts consumed: roughly five hundred. Total spend on the agent: around €20. Total revenue: roughly €2,000 per month from in-game ad sponsorships within the first weeks of launch, plus 1.5 million views on X, 45,000 players in the first week, and enough server load that Zullo had to scale capacity twice in the first ten days.

The case study in the Vibe Coding Wiki entry on the project is unusually detailed about the prompt-to-code ratio. Most of the sessions were the developer asking the model to add a feature, watching the model write the implementation, playing the result, and prompting the model again on the failure modes the playthrough surfaced. The pattern that comes out of the trail is the one almost every shipped vibe coding app shares: the agent writes ninety percent of the code; the developer writes ninety percent of the iteration prompts; the playable artifact is the test suite. The build was entered in the Vibe Jam 2025 contest Pieter Levels organized in March 2025, which collected over 1,170 game submissions with about 500 finalists and a $17,500 prize pool sponsored by Bolt and others. WW2 Dogfight Arena finished as one of the standout entries.

3. Stoppr — the $12K MRR mobile app from a niche clone

David Adius (also reported as David Attias), a French former quantitative trader, built Stoppr, a science-based mobile app that helps Gen Z women quit sugar, in roughly two and a half weeks using Cursor on the $200-per-month plan. The app launched in under thirty days from first prompt to App Store live, hit $5,000 in first-month revenue, and crossed $12,000 monthly recurring revenue by month five. By that point Stoppr had logged 60,000 downloads and roughly 900 paying subscribers across weekly, monthly, and yearly tiers, with a 35% profit margin after operating costs.

The Stoppr playbook differs from the flight-sim wins in two important ways and is worth calling out because it generalizes more cleanly to first-time builders. First, Adius did not invent the product idea. He cloned the screen flow of Quitter, a similar app already grossing roughly $200,000 per month, and rotated the niche from a male audience to Gen Z women aged 13 to 25. The line he gave the press: copy every screen, copy every word, change the niche, switch the colors. Second, the build used Cursor for code, Figma for screen design, Mixpanel for analytics, and TikTok ads for distribution — a full off-the-shelf indie stack rather than custom infrastructure. The lesson the Stoppr case settles is that the “vibe coding apps” category does not require either a viral hook or a novel idea; cloning a known-working product into an underserved niche, with the agent loop carrying the implementation, can produce real five-figure-MRR revenue in months.

4. Lumoo — the Lovable-built fashion AI platform at $800K ARR

Lumoo is a Swedish AI-native content creation platform for fashion and retail brands, built by founders Henrik Skagerlind Fasth and Peter Thörngren on Lovable. The platform integrates virtual try-on, B2B sell-in tools, localization, and automated brand modeling, runs roughly thirty AI models in sync to generate brand-aligned content at scale, and integrates with ERP and e-commerce systems on the customer side. Within nine months of first prompt, Lumoo had crossed €700,000 in annual recurring revenue (roughly $800,000 USD), signed fifteen-plus customer brands including Gant, AWNR Group, Zoovillage, and Brothers, and was tracking toward €1M ARR by end of 2025. The founders publicly attribute the entire build to Lovable, with one quote sticking: this would have taken a minimum of two years to develop in another way.

The capital story underneath the Lumoo product is itself an argument for the vibe coding apps category. Lovable CEO Anton Osika led an angel investment round of roughly €550,000 (SEK 6 million) into Lumoo in late 2025, alongside other prominent fashion-and-tech investors. That mattered because it signaled that the platform vendor is willing to put its own capital into the businesses built on top of it — a vote of confidence Cursor, Bolt, and Replit have not yet replicated at the same scale. The broader platform context: Lovable itself crossed $400 million in annual recurring revenue in February 2026, added $100 million in revenue in a single month with 146 employees, signed roughly half of the Fortune 500 as customers, and reached an $8 million users-generating-100,000-projects-per-day footprint on the way there.

5. ShiftNex — the $1M ARR healthcare staffing platform

ShiftNex is a healthcare workforce staffing platform built on Lovable by founder Allan that uses AI to match nurses to patients on a per-shift basis. The platform reached $1 million in annual recurring revenue (about $83,000 monthly recurring revenue) in five months from first prompt to revenue milestone, with 5,000-plus active healthcare users on the platform. It is the canonical example of the SaaS-shaped vibe coding apps category in 2026: a real B2B problem, a recurring revenue model, and a product that from the outside is indistinguishable from a traditionally coded healthcare staffing system — no “built with AI” watermarks, no missing features, no hand-wave on compliance.

The why-it-worked analysis the case attracted is the more useful half of the story. Healthcare staffing is mission-critical, recurring, and operationally expensive; software that improves matching and scheduling has direct monetary value to hospitals. ShiftNex did not have to build category awareness for what it did because the category already existed; what the agent loop unlocked was the ability to ship a credible product into that category fast enough to land paying customers in five months instead of the eighteen-to-twenty-four month average a traditional development cycle would have required. The pattern repeats across the rest of the SaaS-shaped wins on this list: the durable vibe coding apps are not novel categories; they are existing categories that suddenly got cheap to ship into.

6. Q Group / Qconcursos — $3M in 48 hours from an existing audience

Q Group, the Brazilian EdTech operator behind Qconcursos — one of Brazil’s largest exam-prep platforms with 500,000-plus active paying students — built and launched a new premium product on Lovable in roughly fourteen days, with two developers, and grossed approximately $3 million in revenue within forty-eight hours of launch. The technical stack was the standard Lovable shape: React with Vite and Tailwind on the frontend, Supabase with PostgreSQL on the backend, and full GitHub sync so the company kept code ownership rather than vendor-locking into the platform.

The Q Group case is the one to study if the question is “how do real vibe coding apps make this much money this fast.” The product itself was a relatively straightforward upsell offer to an existing audience; what mattered was that the audience already trusted the brand and was sitting on a five-hundred-thousand-paying-student distribution channel that fired the moment the new product hit the page. The build cost was effectively zero against the revenue. The lesson generalizes to every other entry on this list: the product side of vibe coding apps is dramatically cheaper than it used to be; the distribution side is unchanged. The wins on this list either built distribution first (Levels, Adius via TikTok ads) or already had distribution (Lumoo’s fashion network, ShiftNex’s healthcare relationships, Q Group’s student base). The vibe coding apps that ship to no audience tend to die quietly even when the agent loop succeeds.

The pattern across real vibe coding apps that actually shipped

Six wins is enough to triangulate the pattern. The platform tracked the domain. Cursor with Claude and Three.js powered the browser-game wins; Lovable powered the SaaS and B2B wins; nobody on this list shipped a serious game on Lovable or a serious B2B SaaS on Cursor, because the agent loop on each platform is shaped around the asset and architectural primitives of its native domain. The build cost collapsed; the distribution cost did not. Every winner either had an existing audience (Levels, Q Group), bought distribution cheaply (Adius via TikTok), or sold into a market the founder already understood (Lumoo, ShiftNex). The review pass stayed in the loop. Even Zullo’s zero-human-code build had a human running every playthrough and prompting the model on the failure modes. The case studies that ship are the ones where the founder treated the agent as a fast typist with infinite patience and no judgment, not as an autonomous engineer.

The failure-mode list rounds out the picture. Vibe coding apps that try to invent a new category usually fail because the agent can ship the product but cannot ship the category. Vibe coding apps that skip the review pass tend to ship a working v1 and then collapse on the third or fourth feature when accumulated AI-written architectural debt eats the codebase — a Carnegie Mellon study cited across vibe-coding coverage in 2026 reports a 41% complexity increase on AI-assisted code and velocity gains that disappear within roughly two months without active code review discipline. Vibe coding apps that pair frontier-priced models on both ends of a Planner+Executor pipeline burn cash with no per-token differentiation. The pattern is the same across the wins and the losses: the agent loop is a force multiplier on whatever workflow discipline the founder already had; it does not substitute for the discipline.

How to ship your own real vibe coding app inside Sorceress

The Sorceress stack ships the same shape that produced the case studies above, with two structural differences. First, both WizardGenie and Sorceress Code ship an eight-model picker (Claude Opus 4.7, Claude Sonnet 4.6, GPT-5.5, Gemini 3.1 Pro, DeepSeek V4 Pro, Kimi K2.5, Grok 4.2, MiniMax M2.7), verified against the CODING_MODELS array in src/app/_home-v2/_data/tools.ts on May 13, 2026; the developer changes models per turn rather than committing to one for the project. Second, the asset pipeline lives in the same window: AI Image Gen for character art and props, Quick Sprites for animated sprite sheets, 3D Studio for image-to-3D and auto-rigging, Music Gen for instrumental and vocal tracks, plus SFX and Speech Gen for the audio surface. None of those tools force a context switch out of the agent loop, which is the exact gap that Cursor and Lovable leave open on the asset side.

For a browser-game-shaped vibe coding app along the lines of fly.pieter.com or WW2 Dogfight Arena, WizardGenie is the closer fit because it ships Phaser 4 and Three.js fluency in the agent and a live playable preview pane next to the chat. The starter prompt pattern that worked for Levels and Zullo translates directly: describe the world, describe the controls, describe the win condition, then iterate by playthrough. For a SaaS-shaped vibe coding app along the lines of ShiftNex or Lumoo, Sorceress Code is the closer fit; the parallel-agent flow plus the bring-your-own-key path runs the dual-agent Planner+Executor pattern at roughly one-fifth the cost of single-frontier work, which is the cost discipline the heaviest indie builders run by 2026. The longer treatment lives in the best vibe coding tools for games piece, the inside vibe coding platforms tour, and the model picker guide; the what is vibe coding explainer covers the category-level framing if the reader landed here cold.