I Watched a Founder With Zero Coding Experience Build an MVP in Four Hours
Let me tell you exactly what happened because I am still thinking about it three weeks later.
We had a client -- runs a chain of med spas in South Florida -- who wanted a patient intake system. Not a form. A system. Conditional logic based on treatment type, integration with their booking software, automatic pre-appointment instructions sent via SMS, and a dashboard for the front desk to see the day's schedule with prep notes for each patient.
Our original quote was $12,000 and six weeks. Instead, we sat her down with Claude and Lovable on a Tuesday afternoon as an experiment. By 6 PM, she had a working application. Not a wireframe. Not a mockup. A deployed, functional application that her staff started testing the next morning.
She described what she wanted in plain English. She said things like "when someone books a chemical peel, ask them about their retinol use in the last week" and "the front desk needs to see which patients have not completed their intake by 8 AM." The AI wrote the code. She tested it. She described changes. The AI made them. Repeat.
That is vibe coding. And it is the most consequential shift in software development since the internet moved from the server room to the browser.
She described what she wanted in plain English. The AI wrote the code. She tested it. She described changes. The AI made them. By 6 PM, she had a working, deployed application.
What Vibe Coding Actually Is
The term comes from Andrej Karpathy, who tweeted about "fully giving in to the vibes" when coding with AI -- not reading the code, not trying to understand every line, just describing what you want and trusting the AI to figure out the implementation. He meant it partially as a joke. The industry took it completely seriously because it described something real that people were already doing.
Vibe coding is writing software by describing behavior in natural language. You do not write code. You do not read code. You describe what you want the software to do, the AI generates the implementation, and you evaluate the result by using it. If it does not work right, you describe the problem and the AI fixes it.
It sounds absurd until you watch it happen. Then it sounds inevitable.
How It Works in Practice
There is no single "vibe coding tool." It is a practice that spans multiple tools and models. Browser-based tools like Lovable and Replit work well for prototyping and simple projects. But the real winner for production work is CLI agents like Claude Code that work in your actual codebase.
We found this out the hard way -- browser-based tools stalled on real client projects, and that painful experience pushed us toward Claude Code (we wrote about that journey in detail in our post on Replit, Lovable, and the tools that let anyone build). Claude Code runs in your terminal, in your actual project directory. The vibe coding workflow is the same: you describe what you want in natural language. But instead of working in a sandboxed browser environment, Claude Code reads your real files, runs your real tests, and makes changes with full context of your entire codebase. You still think about your business problem while the AI thinks about the implementation. But now the AI can actually finish the job.
Why This Time Is Different
We have heard "anyone can build software" before. Visual Basic. Dreamweaver. WordPress. Squarespace. No-code tools. Every five years, someone claims the end of professional programming is near.
Those predictions were always wrong because the tools lowered the floor without raising the ceiling. You could build simple things without code, but anything complex still required a developer. The gap between "what a non-developer can build" and "what a business actually needs" remained enormous.
Vibe coding is different because AI does not have a complexity ceiling in the same way. It can write authentication systems, design database schemas, implement payment processing, and handle error states. It can do the boring, intricate plumbing that always forced non- developers to hire developers. The gap is not gone, but it has shrunk from a canyon to a ditch.
Our med spa client did not build a landing page. She built a system with conditional logic, third-party API integration, role-based access control, and automated notifications. That was impossible for a non-developer twelve months ago. Flat-out impossible.
The Implications Are Massive
For Dev Agencies
The $5,000 to $25,000 "build me a basic web app" market is compressing rapidly. If a founder can build a working MVP in an afternoon, the value proposition of a six-week development engagement for a straightforward CRUD application is gone. Agencies that survive will move upmarket -- complex integrations, performance optimization, security hardening, scaling architecture. The commodity work is being automated.
We have already repositioned around this. Half our engagements now start with a client who has already built something with AI and needs us to make it production-ready. We audit, we refactor, we add the things the AI missed -- proper error handling, security, test coverage, monitoring. The starting point of the engagement moved from "blank page" to "working prototype." That is a better place for everyone.
For SaaS Companies
This should terrify the mid-market SaaS vendors charging $200/month for rigid, one-size-fits- all solutions -- and we have been sounding this alarm for a while now. When a business owner can describe exactly the tool they need and have it built in a day, why would they pay for software that does 80% of what they want with a UI designed for someone else's workflow?
The SaaS companies that will hold their ground are the ones with deep, domain-specific data advantages -- think healthcare compliance, financial regulation, industry-specific integrations. If your moat is "we built a nice interface for basic business logic," that moat just drained.
For Technical Hiring
The entry-level developer job market is going to transform. When an AI can write the code that a junior developer would write, the value of a junior developer shifts from "can write code" to "can evaluate AI-generated code, understand system architecture, and make judgment calls about trade-offs." The bar does not lower -- it changes shape.
Senior developers become more valuable, not less. Someone needs to review what the AI builds. Someone needs to design the system architecture that the AI implements. Someone needs to decide when the vibe-coded prototype needs to be rewritten properly and when it is good enough.
The Honest Limitations
Our med spa client's application worked. It also had no test coverage, inconsistent error handling, a database schema that would not survive a migration, and an authentication implementation that made our security engineer visibly uncomfortable. She could not have identified any of these issues because she did not know to look for them.
Vibe coding produces working software the same way that a home cook produces food -- it is functional and often good enough, but it is not restaurant quality. For internal tools, prototypes, and MVPs, that is perfectly fine. For customer-facing production systems handling sensitive data at scale, you still need professional engineering.
The danger is not that vibe coding does not work. The danger is that it works well enough that people do not realize what they are missing until something breaks. And by "breaks" we mean a security vulnerability, a data loss event, or a scaling failure at the worst possible moment.
Where This Is Heading
We are maybe 18 months into this shift and the tools are improving faster than anyone predicted. Claude's coding ability today is dramatically better than it was a year ago. And the real evolution is happening in how the tools work, not just how smart the models are.
The trajectory we see is clear: browser-based tools like Lovable and Replit will dominate prototyping and simple applications. CLI-based agents like Claude Code will dominate production work. The two layers will feed each other -- a founder prototypes in Lovable over a weekend, then a developer (or an AI-savvy consultant) takes that into Claude Code to harden it into production-quality software.
We know this because we live it. Every week, we take something that started as a vibe-coded prototype and run it through Claude Code to add the security, error handling, testing, and architectural rigor that the browser tools miss. The prototype gives us the "what." Claude Code gives us the "how it actually works in production."
Within two years, we expect vibe coding to be the default way that small businesses create internal tools. Within five years, it will be how most software prototypes start, even at large companies. The role of the professional developer will not disappear but will shift permanently toward architecture, review, optimization, and the hard problems that natural language descriptions cannot fully capture.
The barrier to building software has not just been lowered. It has been fundamentally redefined. Code literacy is no longer a prerequisite for creating functional software. That is a sentence that would have been laughable two years ago. Today it is an observable fact.
The vibes are real. And they are reshaping the entire landscape of who builds software, how, and for how much.
Code literacy is no longer a prerequisite for creating functional software. That is a sentence that would have been laughable two years ago. Today it is an observable fact.