A year ago, "vibe coding" was mostly a punchline. You'd see Twitter threads about someone building an app with ChatGPT, and the replies were... skeptical, to put it mildly.

But something shifted.

Solo founders with no engineering background are now shipping real products. 

I want to be clear, here I’m not talking about AI replacing developers, but rather, I want to talk about what happens when the cost and time of building drops significantly, and the real constraint becomes taste, speed, and distribution.

We looked at six products built primarily with AI-generated code. All are live. All are generating revenue. Here's what's actually working, and honestly, some of it surprised us.

The weekend that changed everything

Zohar Vanunu already had a full Figma design ready when he stumbled upon Lovable on Product Hunt. He wasn't looking to build the whole product himself. He just wanted to generate some real components for his devs.

But curiosity led to one prompt. Then something clicked.

"I already had the design," Zohar recalls. "I just wanted to generate real components for my devs. But once I saw Lovable working, I said to my wife: 'Don't talk to me this weekend. This is a moment in history.'"

He did 1,500 prompts in two days.

That free weekend changed everything. Zohar won the internal Lovable promptathon and received 12,000 credits. He didn't waste a single one.

That’s how he built MagiCan—an AI-assisted infinite canvas that lets creators visualize, map, and narrate ideas on a flexible workspace. 80–90% of it was built directly inside Lovable through prompt-driven development. Developers came in only at the end to clean up and productionize.

The founder has no engineering background. And yet he shipped a production-ready SaaS that saved an estimated $200K in development costs.

How an Indie hacker reached $75k MRR in no time

If you've spent any time in the indie hacker world, you probably know Pieter Levels. The Dutch serial entrepreneur and digital nomad built Nomad List and Remote OK while travelling the world: solo, profitable, and famously lean.

So when Pieter decided to build a browser-based multiplayer flight simulator, people paid attention.

Fly Pieter is a low-poly game where players fly planes, see each other in real time, and can engage in combat. It runs entirely in your browser. And roughly 80% of the codebase was AI-generated using Cursor.

The frontend is built with HTML, JavaScript, and Three.js. Multiplayer functionality: broadcasting player positions every 100ms was built with a Python WebSocket server reportedly written using Grok 3, while ChatGPT handled debugging. 

The numbers as of March 2025:

• $75K MRR
• ~$52K/month in ad revenue from 22 in-game banner ads
• 89K total players, 26K peak concurrent

That's $127K monthly from a browser game built mostly by AI.

But here's what makes Pieter's story instructive: he's not a gamer, and he's not a game developer. He's someone who understands distribution, monetization, and shipping fast. The AI handled the code. Pieter handled everything else.

The ceiling isn't the technology anymore. It's knowing what to build and how to get it in front of people.

The database-first revolution

Kyle Ledbetter and Andy Keil took a completely different approach, and it might be the most replicable playbook in this entire piece.

With Dreambase, they built three products in nine months. That's a pace that would take most teams years.

Their secret? They flipped the typical AI-assisted development process on its head.

Most builders start with UI mockups, generate some code with AI, and then discover their beautiful interface can't handle real data. Kyle and Andy did the opposite:

Their counterintuitive process:

• Start with database schema: map data relationships upfront, often using AI to help design them
• Create wireframes that reflect these data structures, not aspirational UI dreams
• Build prototypes connected to real data models from day one
• Let the interface emerge based on how data actually needs to be displayed

Why does this work so well with AI tools? Because AI understands data structures. Feed it proper schema context, and it becomes dramatically more helpful for both coding and design. No painful "prototype to production" rewrites when you discover your AI-generated UI can't handle real data.

Dreambase itself is an AI-native analytics product for Supabase users. It connects directly to a Supabase project, pulls schema, relationships, and security config, then generates reports and internal views through natural-language queries.

The team maintains a single markdown "AI Requirements" doc containing database schema, user flows, design rules, and constraints. This doc feeds Cursor, Claude Code, v0, Lovable, and Bolt simultaneously. They run parallel "bake-offs" across tools and combine the best outputs.

Where they're at:

• 100+ users
• Five-figure revenue
• Team size: 2
• Time to initial product: ~3 months

4 days to MVP, 4 weeks to $12K

Sebastian Volkis gave himself a week to launch.

He built an MVP in four days.

TrendFeed is a SaaS tool that identifies viral news from credible sources, evaluates their virality potential, and uses AI to generate short-form video content for Instagram and TikTok.

Here's the thing: there's no handwritten code in the stack. None.

It's built on Next.js and React, with ShadCN for the UI, Supabase for the database, and Vercel for deployment. The stack is evolving to include Clerk for auth and Whop for payments.

The traction:

• £9,222 (~$12K) in the first four weeks
• £5.5K on launch day alone
• 22 lifetime customers at £249 each (launch cohort)
• Launched via webinar-based controlled sales funnel

What I find smart about Sebastian's approach is the GTM strategy. He skipped traditional self-serve SaaS entirely, launching via webinar to a controlled audience. This de-risks the "is the product ready?" question entirely. You don't need perfect code when you control the sales environment.

The marketing initially leveraged Sebastian's existing social audiences, then shifted to brand-owned accounts to reduce key-person dependency and support a potential future exit.

Four days of building. Four weeks to five figures. That's the math.

Built in 24h, sold for $15k

Josh Pigford is a well-known builder in the Twitter/X community, the kind of person who ships constantly and thinks in public.

One day, Josh spotted an idea on Greg Isenberg's ideabrowser.com. Instead of bookmarking it, he decided to build it.

That's how Namesnag was born.

Namesnag is an AI-powered tool that scans expired .com domains to help users find available names for new projects. Josh built it end-to-end with Claude Code in approximately 12 hours. The backend runs on Ruby on Rails and PostgreSQL—a traditional, proven stack, just accelerated by AI.

What happened next:

• $500 MRR within 24 hours of launch
• Sold for $15,000 a few weeks later

This is the micro-exit playbook in action. See an idea, build fast, validate demand, flip for a quick multiple.

The math is obvious when build time collapses to hours. If it doesn't work, you've lost a day. If it does, you've got an asset worth five figures. That's a risk/reward ratio any builder can appreciate.

25 years of fashion expertise, 9 months to €700K ARR

Henrik and Peter first worked together eight years ago at a Swedish retail technology company, building digital in-store experiences for brands like Axel Arigato and Bestseller.

Between them, they have 25 years of combined fashion and retail experience and 15 years in commercializing technology solutions.

When they built Lumoo, they weren't experimenting. They knew exactly what problem they were solving.

Lumoo is an AI-powered content platform for fashion, retail, and beauty brands. It generates virtual try-ons, on-brand model images, dynamic videos, and complete lookbooks at scale—the kind of content production that traditionally requires expensive photoshoots, models, and post-production teams.

The entire product was built using Lovable, integrating approximately 30 synchronized AI models. It connects directly with ERP and e-commerce systems.

The numbers:

• €700K ARR in ~9 months (aiming for €1M ARR by end of 2025)
• 15+ brand customers, including Gant, AWNR Group, Zoovillage, Brothers
• Commercial traction began in February 2025

What makes Lumoo different from the other products in this piece is that Henrik and Peter didn't just use AI to build the product; they use AI at runtime to deliver value. The product is AI.

This dual use (AI-built, AI-powered) is where the leverage really compounds. They used AI tools to build faster, and they built something that uses AI to make their customers faster.

Deep domain expertise plus AI-accelerated development. That's a powerful combination.

What patterns are we noticing?

Six products. Six different builders. Six different approaches.

But some patterns keep showing up:

Experience matters more than ever. Every builder in this piece knew what to build before they started. Pieter understands distribution. Kyle and Andy understand data architecture. Henrik and Peter understand fashion retail. Sebastian understood his audience. Josh understood the micro-exit playbook. Zohar had his design ready.

AI builds the product, but it doesn't tell you what's worth building.

Small teams, outsized output. Every product here was built by one or two people. This means that AI not only speeds up development but also changes the staffing equation entirely.

Speed compounds. Namesnag went from idea to live product in 12 hours. TrendFeed's MVP took four days. MagiCan emerged from a single weekend. When build time collapses, you can afford to be wrong more often, which means you find what's right faster.

Distribution still wins. Fly Pieter's success isn't about Three.js. It's about 89K players. TrendFeed launched via webinar, not Product Hunt. The winning products have practical GTM strategies, not just GitHub repos.

The risks are real

I don't want to oversell this, don’t get me wrong. The approach also presents real challenges that should be considered. 

Debugging gets strange. When 80% of your code is AI-generated, tracing bugs is harder. Most of these builders used ChatGPT and similar tools for debugging; essentially, they’re using AI to fix AI.

Technical debt accrues differently. AI-generated code can be verbose, inconsistent, or subtly wrong in ways that compound. Zohar's approach, bringing in developers at the end to "clean up and productionize", acknowledges this.

Scaling limits are unknown. A 26K concurrent player game and a €700K ARR platform are real traction. But we don't yet know how these codebases hold up under 10x growth or multi-year maintenance.

You still need to know what you're building. Every success story here features someone with deep domain knowledge or proven product instincts. AI accelerates execution, but it doesn't replace judgment.

What to focus on if you’re building with AI?

A practical checklist, if you’re considering building your product with AI:

Start with what you know. The builders who succeeded here brought domain expertise or proven product sense. AI amplifies these, but it doesn't create them.

Timebox aggressively. Josh built Namesnag in 12 hours. Sebastian gave himself a week and shipped in four days. Set hard limits. If you can't get to something shippable fast, this approach might not be right for that particular problem.

Plan for cleanup. Budget time (or hire help) to review and stabilize AI-generated code before scaling.

Where does this leave us?

We're in an interesting moment.

The tools to build software are more accessible than ever. The cost of a first version is approaching zero. Builders with domain expertise but no engineering background are shipping production products.

And yet, most AI-built products still don't ship, don't monetize, or don't survive contact with real users.

The products in this post are the exceptions, but they're becoming less exceptional every month.

The pattern is clear: experienced builders who know what to build, using AI to collapse the timeline, then focusing their energy on distribution and monetization.

This isn't the end state. The codebases are young, the scale is modest, and the long-term maintenance story is unwritten.

But the gap between "AI prototype" and "production product"? It's closing. Fast.

Building with AI or thinking about it? At Solveo, we work with founders and product teams navigating this shift. Whether you're exploring what's possible, scaling what's working, or figuring out where to start, let's talk.

‍