Vibe coding: The pros and cons and how to avoid the risks

I used to hate vibe coding Every time I tried it, I spent more time fixing AI’s mistakes than just writing the code myself Then something changed We used vibe coding to build a new feature for Viteval (our OSS evals framework) The catch? We spent time upfront making sure there was great markdown context for the AI to reference Here’s what worked: - Added an Agent MD to the repo - Linked in key context (testing, coding rules, architecture) - Let AI take the first pass The result? 90% of the feature was vibe coded. A couple reviews, some testing, and it shipped Do others feel vibe coding is ready for production or still just for prototyping??

to all the vibe coders out there: i’ve been building with AI for months, and here’s my take. at first, vibe coding feels magical. you can ship features in hours, explore frameworks without reading docs, and get a prototype out overnight. but the dark side shows up fast. the codebase gets messy. ai stops helping because it doesn’t know where to put new features. you spend more time refactoring than building, and the hardest part is that, you never develop real engineering intuition. that’s why vibe coding alone doesn’t get you from “prototype” to “product.” AI can give you something that seems to work. But only engineers can build systems that scale, get users, and survive. every product that lasts has one thing in common: real engineering underneath. There’s no shortcut. so if you want to launch something real, yes, use AI. it’s amazing for speed, validation, and momentum. but you also need to grow as a real engineer. that’s why I built vibecheck (link in comment). it keeps the velocity of vibe coding, while training you to think like an engineer. because ai builds prototypes. engineers build products.

WHY VIBE CODING FAILS Vibe coding is fun. It feels like magic. But if you’ve tried building anything serious with it, you’ll know the cracks show up fast. Here are 5 of the biggest problems and how to fix them SECURITY LANDMINES 🛡️ AI code often hides vulnerabilities like SQL injections or unsafe libraries. Fix: Always run vibe-coded output through a security scanner and dependency checker. Review it like you would an intern’s work. SPAGHETTI SCALING 🍝 What looks fine in a demo quickly collapses when you try to scale. Fix: Set architecture rules early, document as you go, and don’t skip writing tests. Speed is useless if the foundation is shaky. BLIND TRUST 🙈 Too many people paste AI code straight into production. That’s how bugs and debt creep in. Fix: Don’t just run it. Ask the AI to explain its own code back to you. If you don’t understand it, don’t ship it. CONTEXT DRIFT 🔁 Long chats make AI forget details. Suddenly variables are renamed or features disappear. Fix: Break work into chunks and lock in working pieces as files. Keep your spec in one place so you can always reset. GREAT PROTOTYPES, WEAK PRODUCTION 🚧 AI will get you 80% there. The last 20% is edge cases, optimisation, and scaling that still needs a human. Fix: Treat vibe coding as scaffolding. Refactor and harden before you ship. THE BOTTOM LINE 📌 Vibe coding is a turbo boost, not autopilot. Use it for speed and creativity but back it up with discipline, testing, and real engineering. That’s how you turn vibes into working code. #vibecoding #aicode #vibecodingbugs #AIcodegeneration #AIprompt

💻 Vibe Coding: When the Code Writes You Back There’s a moment in every developer’s journey when the screen stops feeling like a tool… and starts feeling like a canvas. The syntax fades. The bugs become puzzles. The hours disappear. You’re not just writing code, you’re in the code. That’s vibe coding. It’s the late‑night flow where your playlist syncs with your keystrokes. It’s the quiet satisfaction of watching a feature come alive exactly as you imagined it. It’s the art of turning logic into something that feels right, not just runs right. For me, vibe coding isn’t about chasing perfection, it’s about chasing connection: Connection between idea and execution Connection between developer and problem Connection between technology and creativity Whether you’re building a hyper‑realistic AI art pipeline, scaling a backend for millions, or just tinkering with a side project — find your vibe. Protect it. And let it carry you to the next breakthrough. What’s your vibe coding ritual? #VibeCoding #DeveloperLife #FlowState #CodingCulture #TechCreativity

Specs, specs, specs... The vibe coding workflow seems to be well established now. You need to write decent specs about what the software should do. Then you should plan step-by-step how to get the implementation done. Make AI follow the plan. And then probably test that it actually works. That's pretty much how software has been done for the past forever(?) so it's a little amusing to see how the wheel gets reinvented now with "revolutionary AI-native approaches". I predict we'll soon add "spec" into the usual design/develop/test sprint rotation.

When I first heard about vibe coding, I thought it sounded like a shortcut that would break everything later. But then I tried it. I described my idea in plain English… and AI spun up the first version in hours. Not weeks. Not months. Hours. That’s the real power: speed. You can test an idea with customers right away, show a proof of concept, and learn what works. But here’s the catch: Getting to MVP doesn’t mean you’re ready for production. You still need: ✅ Testing ✅ Debugging ✅ Security checks Vibe coding helps you start the race. But fundamentals win the finish line. What’s one challenge you’ve faced with vibe coding? #VibeCoding

Here's what I see consistently across vibe coding communities → Someone builds an amazing app with AI in 30 minutes → They try to deploy it → 6 hours later they're still debugging environment variables or trying to fix one thing or the other → They give up and move on to the next project A whooping 80-90% of AI-generated code never sees production. Not because the code is bad. Because deployment is broken. CodeReady AI fixes this. And i cant wait to share it with the world. With CodeReady AI, you can now turn AI prototypes into production apps in minutes. Not days, not weeks, not months, not having to seek external help. Analyze your code, fix it (auth, db, API endpoints, security hardening) and deploy automatically. What do you struggle with the most when vibe coding? 👇

🚀 Let’s Talk About Vibe Coding! Many developers today are harnessing the power of vibe coding to make their workflow faster, smarter, and easier. 💡 What is vibe coding? It’s simply using AI to write, debug, and even optimize your code. Some years back, tech forums were our lifeline. Anytime a bug popped up, we’d rush to StackOverflow or developer communities. Debugging could take hours (or even days!) because solutions came in bits and pieces. Now, with AI-assisted coding, answers come in seconds. Just copy your error log from the console, paste it into an AI assistant and boom! You’ll likely get a fix (well… not always that easy, but close enough 😂). But it doesn’t stop at debugging anymore. Developers are now building complete projects end-to-end with AI support. At the end of the day, nobody really cares what’s running under the hood what matters is that your app works when you call on it. 👉 That’s the essence of vibe coding: working with AI-assisted bots to aid your coding journey. In my next post, I’ll share: • The best AI models I’ve used for vibe coding • The guardrails you must set if you want a reliable and functional program 🔥 Stay tuned. Vibe coding is here to stay. I am OliseyenumOmodi

Vibe check: passed ✔️ Production readiness: we need to talk 🤔   AI opens up incredible new ways of working, but turning those good vibes into working systems still demands architectural vision and seasoned expertise. Our honest thoughts on what vibe coding means for the future of software development, over on the blog: https://lnkd.in/evmRx9Me #vibecoding #code #AI #goodvibesonly

When I first started using code-generation models, the early build sessions felt magical. I’d be like a magician casting spells, creating the thing just because I have the power to speak it into existence. Then midway through, the vibe shifts. That initial sugar rush of everything feeling just so possible gets replaced by the inevitable swarm of runtime bugs — bugs that I didn’t even write! The whole session turns into a bit of a whack-a-mole game, and suddenly it’s not vibe coding… it’s vibe debugging. 🕹️ This has happened enough times that I’ve realized an important lesson: unless I fully comprehend the logic the model gives me, generative code = tech debt (pretty much a linear relationship). In thinking about how to solve this problem, I turned to my in-house Senior Software Engineer (salary: $20/month). My intuition was telling me I needed a more modular approach where I can split the application into its main parts and build each part independently, but I wasn’t quite sure how to do this. It pointed me towards this development concept of fixed-interface design. By interface I don’t mean user interface — I mean how the parts interface with each other. To get control back, I adopted a fixed-contract development approach: → Define each module’s contract — inputs, outputs, and guarantees. → Formalize the seams so modules can be built and tested independently. → Generate code only after the contracts that are relevant to that module’s operation are nailed down. In the app I’m building now, I’ve split it into four modules with six explicit contracts, where each module has its own mini-spec and each contract has its associated 'Contract Kit'. This small change transformed the process: codegen stays contained, debugging is localized, and I stay in control of the architecture. Overall it’s allowed me to keep my codegen model on a shorter leash as I go about building something that does what I need. It’s humbling that by far the slowest part of the development loop is me; but I’m learning a ton. 😄 #VibeDebugging #AI #AlgoTrading