Skip to content

Guides

Is Vibe Coding Risky? What to Know Before You Build

Is vibe coding risky? Learn where AI-written code goes wrong, which projects are safe to try, and a practical checklist for building apps without coding.

On this page
  1. What Popular Science warned about vibe coding
    1. What counts as vibe coding
    2. Why gadget tinkering is the example that raised alarms
  2. Why vibe coding is risky when you can't read the code
    1. Guessing instead of knowing
    2. Fixing one bug, creating another
    3. Security holes you won't notice
  3. Two kinds of vibe coding: new apps vs. modifying things you rely on
    1. Low-stakes projects that suit AI building
    2. High-stakes projects to leave to maintained software
  4. What to do instead: lean on maintained, community-backed tools
  5. A practical safety checklist for vibe-coded apps
    1. Before you build
    2. Before you publish
    3. After launch
  6. Build it with FromScratch
    1. A safe first project
    2. Using versions as your undo button
  7. FAQ
Person reviewing app code on a laptop, asking whether vibe coding is risky
Photo by Arnold Francisca on Unsplash

Is vibe coding risky? According to a recent Popular Science piece, it can be, especially when you point AI-written code at the devices you use every day. The warning is aimed at gadget tinkering, but it carries a useful lesson for anyone building apps without coding: the safety of vibe coding depends less on the AI and more on what you aim it at.

Popular Science describes vibe coding as the informal term for prompting an AI chatbot to create a program. You explain in natural language what you want, the AI writes it, and you use follow-up prompts to fix whatever goes wrong. The article notes that this has let non-programmers build their own sites and apps quickly, and that the practice has spread to hardware too, including people hacking a 1994 Jeep Cherokee or repurposing an Amazon Kindle.

What counts as vibe coding

In practice, it is any workflow where you describe the outcome and let AI produce the code, rather than writing or reading it yourself. A phone app that tracks your exercise, a browser game, a registration site for a sports club: all fit. The article is upbeat about this part. As long as your ambitions are modest, you get a usable, custom app within minutes. It also admits the results often need finessing and don't match what a team of professional developers would deliver.

Why gadget tinkering is the example that raised alarms

The article points to TVs as a common target, citing TechCrunch. People use AI-generated code to simplify app launchers, strip out sponsored content and ads, and turn off aggressive tracking features. The appeal is obvious. The problem is that AI code often arrives with bugs and unintended consequences, and if you don't understand what the code does, you can't fix it.

That is the core message worth keeping: code you can't read is code you can't repair.

Why vibe coding is risky when you can't read the code

Popular Science lists several ways things go wrong. Each one maps neatly onto the risks non-coders face when building their own software.

black flat screen computer monitor turned on near black wall
Photo by Zac Gudakov on Unsplash

Guessing instead of knowing

The article says the operating systems on TVs, tablets, laptops, speakers and fridges are complex, and that AI will often guess at the right approach if it doesn't know for sure. A guess that looks confident is the dangerous kind. The code runs, the screen changes, and nothing tells you the underlying assumption was wrong.

Fixing one bug, creating another

The source warns of creating new bugs while trying to fix existing ones. Anyone who has chatted their way through an app build has seen a version of this: you ask for a small change and something unrelated stops working. When you can't inspect the code, you find out only by testing.

Security holes you won't notice

The article also flags security vulnerabilities left open for hackers to target. This is the quietest of the vibe coding risks, because a security gap produces no error message. The app works perfectly until someone finds the gap. For an app that handles customer names, emails or payments, that matters far more than a visual glitch.

The article's worst cases for a TV are vivid: breaking key functionality (the Netflix app stops working for reasons you can't fathom), opening the set to malware, or a device stuck in an endless loop between shutting down and powering up. The common thread is a system you can't easily reset or replace.

Two kinds of vibe coding: new apps vs. modifying things you rely on

So is vibe coding safe? The article implies a useful answer: it depends on the target. It says a custom soccer score tracker running in your browser is not much of a problem, while bricking your main TV is. That distinction is the one to build your habits around.

New, separate app Modifying a device or system you depend on
Example Booking page, tracker, internal tool TV software, router, smart-home setup
Can you undo it? Usually, if you can restore earlier versions Often not
Worst case A broken or embarrassing app you can discard A bricked device, lost functionality
Touches other systems? Only what you connect Everything the device runs

Low-stakes projects that suit AI building

Good candidates are new, self-contained apps: a booking app, a simple dashboard, an internal tool for your team. If the first version is wrong, you throw it away or fix it in the preview. Nothing you rely on breaks while you experiment.

High-stakes projects to leave to maintained software

Anything that modifies a system you depend on belongs in the other bucket: firmware, router settings, the software running your TV, your smart-home hub. Before you start any project, ask three questions:

  1. Can I undo it? If the answer is "only by buying a new device," stop.
  2. What is the worst case? Compare a broken prototype with a broken TV.
  3. Does it touch other systems or personal data? The more it connects to, the more a hidden bug can spread.

What to do instead: lean on maintained, community-backed tools

Popular Science's practical advice is not "don't customize your gadgets" but "don't vibe-code a replacement for something mature software already does." It names three examples:

red padlock on black computer keyboard
Photo by FlyD on Unsplash
  • Projectivy Launcher for Google TV models: an ad-free launcher with lots of customization options.
  • OpenWrt for compatible routers: built and maintained by a large enthusiast community, it can block ads and trackers, control which devices access the network and when, monitor bandwidth, and integrate a VPN at the router level.
  • Home Assistant for smart homes: an alternative management system that gives you more control than the comparable apps from Amazon, Google and Apple.

The article is honest about the trade-off. You are still trusting third-party code you didn't write. The difference is that these projects are run by capable developers and backed by big user communities who watch for security and compatibility problems. Many eyes on the code is the safeguard a one-off AI-generated script lacks.

The takeaway for app builders: before you prompt an AI to build something, check whether a maintained product already solves the problem. Building your own makes sense when you need something specific to your business, such as your intake form, your booking rules or your client list, and not when you are recreating a well-tested tool.

A practical safety checklist for vibe-coded apps

If you are building a new app, here is how to vibe code safely. These steps are our own working practice, not claims from the article, and they follow its central logic: keep the blast radius small and keep a way back.

Before you build

  • Start with an isolated new project. Don't connect it to a live system or a device you rely on.
  • Ask for changes in small steps. One change per message lets you see what each one did, which counters the "fix one bug, create another" problem.
  • Keep sensitive data out at first. Use sample data until you've reviewed how the app stores and shows information.

If you are new to this, the beginner's guide to a safe start walks through a low-risk first project.

Before you publish

After launch

  • Keep making changes in small steps and re-test the parts that matter, such as login and forms.
  • Watch what real users do. Unexpected inputs are where hidden bugs surface.
  • If the app grows to handle payments or sensitive records, revisit the security checklist before adding more.

Common vibe coding mistakes to avoid: building on top of a live system, publishing without testing the unhappy paths (wrong password, empty form, duplicate booking), and loading real customer data before reviewing how it's handled.

Build it with FromScratch

FromScratch is an AI app builder that fits the "new, separate app" side of the line above. You describe the app in chat, and an AI agent builds a React front-end, a Python (FastAPI) back-end and a database. It appears in a live preview within minutes, so you can test it before anyone else sees it. Here is a safe first project: a booking app for a small studio, kept entirely separate from your devices and core systems.

A safe first project

  1. Open FromScratch and start a new project. You are creating a fresh app, not touching anything you already run.
  2. Paste this prompt into the chat:

Build a simple booking app for my small studio: clients pick an available time slot, enter their name and email, and receive a confirmation on screen. Add an admin page, protected by login, where I can see and cancel bookings. Start with a clean, simple design and show me a live preview.

  1. Review the live preview. Book a slot yourself, then try to break it: leave fields empty, book the same slot twice, open the admin page without logging in.
  2. Refine in small steps. Ask for one change at a time, for example "require login for the admin page" or "add a search bar to the bookings list."
  3. Attach your real content if you need it. You can upload PDFs, images or spreadsheets so the app uses your actual services and hours.
  4. Publish when you're satisfied. One click gives you a public URL, or you can use your own custom domain. A free plan is available to start.

For more on this kind of project, see how to build a booking app with AI.

Using versions as your undo button

The biggest difference from tinkering with a TV is reversibility. Every change in FromScratch is saved as a version, and you can restore an earlier one. If a request breaks something, you go back instead of debugging code you can't read. That doesn't replace testing or a security review, but it removes the "stuck in a loop with no way out" scenario for the app itself.

FAQ

Is vibe coding safe?

It is safer when you build a new, isolated app with a preview, restorable versions and login than when you alter devices you depend on. AI code can still contain bugs and security gaps, as Popular Science points out, so testing and a pre-launch security review remain necessary.

Can I vibe code my TV or router?

The source advises against it. Popular Science recommends maintained, community-backed alternatives such as Projectivy Launcher for Google TV, OpenWrt for compatible routers and Home Assistant for smart homes. A mistake there can break functionality or leave a device unusable.

Do I need to know how to code?

Not to start. You do need to test carefully, add login before real users arrive, and review security before launch. If you want a step-by-step path, the guide to building an app without coding is a good place to begin.

What's the difference between vibe coding and an AI app builder?

Vibe coding is the practice of describing what you want and letting AI write the code. An AI app builder wraps that in structure: a live preview, saved versions you can restore, and publishing. Those pieces make the process easier to test and reverse. See how no-code tools and AI app builders differ.

What should I check before publishing?

Check that login works and protects private pages, that forms handle empty or wrong input, and that no sensitive data is exposed. Confirm you can restore an earlier version, then work through the AI app security checklist before sharing the link.

Written by

FromScratch TeamEditorial team

The team building FromScratch, the AI app builder that turns a plain-language description into a real, published full-stack app.