Software Development

Vibe Coding Is Great for MVPs, But Sloppy Vibe Coding Is Ruining Our Apps

Under The Headline
Under The Headline

8 min read

Vibe coding makes building an MVP faster than ever, and I genuinely love it for rough ideas. Yet as unreviewed AI apps flood the web, sloppy vibe coding is quietly dragging down software quality, security, and the everyday user experience we all rely on.

Picture this: it is a rainy Sunday afternoon, you have a half-baked idea for a minimalist bookmarking app, and instead of wrestling with setup files for three hours, you open an AI editor and type a casual paragraph about what you want. Four minutes later, it works. The buttons click, the dark mode looks slick, and you feel like a wizard.

I have had that exact Sunday afternoon. Honestly, it is thrilling. After years of spending late nights hunting down missing semicolons or fighting with CSS layouts, watching an idea materialize at the speed of thought feels like a genuine superpower.

Then, out of old habit, I actually opened the files the AI had generated. That was when the magic wore off.

Buried beneath that gorgeous interface were hardcoded API keys, zero checks on what users could type into forms, and database queries written in a way that would quietly melt down the moment fifty people used the app at the same time.

That little reality check summed up something I have been feeling all year. I do not hate vibe coding—not even close. For throwing together an MVP, it is fantastic. What bothers me, and what I think is visibly dragging down the quality of modern software, is sloppy vibe coding.


Why I Still Love Vibe Coding for MVPs

When Andrej Karpathy coined the phrase "vibe coding" back in early 2025, he was talking about a fun, low-stakes way of building—fully giving in to the vibes, letting the AI handle the code, and not sweating the details for quick weekend experiments.

Used that way, vibe coding is brilliant. If you are testing whether anyone actually cares about your startup idea, spending two months agonizing over clean architecture is just procrastination with extra steps. You need something real in front of humans by Friday.

From my own side projects, letting AI take the wheel early on has obvious perks:

  • Zero blank-page dread: You go from a shower thought to a clickable prototype before your coffee gets cold.

  • Anyone can build a rough draft: Designers, writers, and tinkerers no longer have to beg a developer friend just to test a concept.

  • Real feedback, fast: You find out what people hate on day two instead of month two.

As a digital sketchbook, vibe coding is unbeatable. The trouble starts when people mistake the rough sketch for a finished building—and start charging rent.


Where Things Go Sideways: The Rise of Sloppy Vibe Coding

Somewhere along the line, "forget that the code even exists" stopped being a playful tip for weekend hacks and turned into a badge of honor for shipping real products. Scroll through tech social media today and you will see endless posts bragging about launching full-scale apps in a weekend without reading a single line of code.

To me, that is the line between healthy prototyping and sloppy vibe coding. It happens when builders hit "Accept All" on every AI suggestion, skip testing entirely, and push unreviewed code straight to real users.

"AI pushed the cost of writing code down to almost zero, which tricked a lot of us into thinking the cost of maintaining software dropped to zero too."

In my opinion, we are already feeling the fallout every time we open a new web app. Software lately feels weirdly fragile. Buttons stop responding for no obvious reason, forms break the moment you enter an unexpected character, and basic features feel held together by digital duct tape.


Four Hidden Traps Behind the "Ship Fast" Hype

When you never look under the hood, the mess rarely shows up in a thirty-second launch video. It shows up three weeks later, usually on a Friday night. Based on my own experiments and conversations with other builders, sloppy vibe coding tends to spring four predictable traps:

  • Security becomes pure luck: AI models are trained to make your prompt work, not to protect your users. Unless you specifically know what to ask for, they will happily leave private keys exposed in the browser or leave your database wide open.

  • Instant spaghetti code: Because nobody is guiding the structure, every follow-up prompt just tapes a new patch over the old one. Before long, fixing a typo on the settings page somehow breaks user logins.

  • Phantom packages and "slopsquatting": Security researchers have pointed out that AI tools sometimes hallucinate software package names that do not actually exist. Attackers have started registering those fake names with malicious code—meaning anyone who blindly runs an AI-suggested install command could infect their own project.

  • The debugging doom loop: When a real bug hits a codebase you never read, you cannot reason your way out of it. Your only move is pasting error messages back into the chat box and hoping the AI guesses right, while watching it break three other things in the process.


Pretty Screens, Fragile Foundations

What makes sloppy vibe coding so tricky is how convincing it looks at first glance. Today's AI models are remarkably good at making things look professional. They know modern styling frameworks inside out, pick tasteful fonts, and nail smooth transitions.

Personally, I think that surface polish creates a dangerous illusion. A sleek interface tricks both the creator and early users into assuming the engineering underneath is just as solid. It is the software equivalent of slapping fresh paint over rusted pipes and calling it a remodel.

More than that, I worry it is hollowing out the craft of building things people can count on. Good software has always been about caring for the unglamorous moments—what happens when someone's Wi-Fi drops mid-checkout, when a user pastes an emoji into a phone number box, or when traffic suddenly spikes. Sloppy vibe coding skips all those questions because the happy path looks great in a screenshot.


When Nobody Knows How the App Actually Works

There is another side to this that rarely gets talked about: what happens when you actually have to live with what you built.

Even messy human-written code usually has a story behind it. Someone remembers why a weird workaround exists or where the payment logic lives. With sloppy vibe coding, nobody holds that map in their head—not even the person who typed the prompts.

Imagine getting woken up at 3:00 a.m. because customers cannot log in, opening your project, and staring at thousands of lines of code you have never read. You do not know which file handles sessions, how data is cached, or why three different date libraries are installed. At that point, you are not running an app; you are negotiating with a stranger's fever dream.

Even Andrej Karpathy himself has since pointed toward a more structured approach—calling it "agentic engineering"—where builders still rely heavily on AI agents, but step back into the role of planner, reviewer, and quality gatekeeper. That evolution says a lot to me: even the person who sparked the vibe coding wave knows vibes alone cannot hold up real software.


Keeping the Vibes Without Losing the Craft

I am not nostalgic for the days of writing repetitive boilerplate by hand. I still use AI coding tools almost daily, and I have no plans to stop. The answer is not ditching vibe coding; it is knowing when the experiment ends and real engineering begins.

If I had to boil down the rules I now stick to for my own projects, it comes down to four simple habits:

  1. Vibe the prototype, engineer the product: Use conversational prompting freely while exploring an idea. The second real user data or payments enter the picture, slow down and read the code.

  2. Never ship code you cannot explain: You do not have to write every line yourself anymore, but you should be able to open any file and explain what it does and how it might break.

  3. Treat AI like an eager junior teammate: Let it draft features, write boilerplate, and suggest tests—while you stay in charge of architecture, edge cases, and security.

  4. Double-check every dependency: Before installing any package an AI recommends, take ten seconds to verify that it is real, maintained, and safe.


A Quick Reality Check Before You Hit Deploy

At its core, an app is a quiet promise to the people using it. When someone signs up, saves their personal notes, or types in their credit card number, they are trusting that someone on the other side cared enough to build something sturdy.

Vibe coding handed us an amazing gift: the ability to turn rough ideas into working prototypes in an afternoon. Let us enjoy that freedom, play with weird ideas, and build faster than ever. Just do not let the rush of shipping quick turn into an excuse for shipping junk.

Enjoy reading on PBLISHD?

Bring your reading experience closer. Install PBLISHD on your device.