«Vibe coding is not magic. It’s a crazy keyboard in the hands of someone who knows what to do with it»
The term that divided everyone
When Andrej Karpathy — one of the most influential people in AI, co-founder of OpenAI, former Director of AI at Tesla — first used the phrase “vibe coding”, he probably didn’t imagine it would become simultaneously a meme, a marketing term, and a reason for flame wars in every dev chat on the planet.
The idea is simple: you describe to AI what you want to get — and it writes the code for you. You don’t read every line, you “catch the vibe”, check the result and move on. Sounds like a dream. And partially it is.
But I’ve spent enough time with this approach to understand: between the promotional videos “I created an app in 10 minutes” and real production code — there’s a gap the size of the Grand Canyon.
My journey through all the tools
I’ve tried almost everything that’s on the market now. Cursor, OpenAI’s Codex, Antigravity and a dozen other smaller tools. The only thing I didn’t touch — Chinese models, on principle.
And after all the experiments, the winner for me is unambiguous: Claude Code from Anthropic based on the Opus 4.6 model. This is not advertising — this is the result of months of daily work. Opus understands project context, maintains architectural vision, generates code that you’re not ashamed to read. Other models are also good for individual tasks, but when it comes to serious development — Claude Code is unmatched so far.
“A derogatory term” or new reality?
I recently watched Peter Steinberger’s interview with Lex Fridman, and one moment hit very hard. Peter — the author of OpenClaw, today one of the craziest things in the world of vibe coding and AI agents — called the term “vibe coding” a derogatory word. And proposed an alternative: agentic engineering.
His explanation was genius in its simplicity: during the day he does agentic engineering. But after the third night he can switch to vibe coding — and usually regrets it the next morning.
Separately, he noted that the results of people who can’t program at all, he calls not pull requests, but prompt requests.
And I absolutely agree with this. There’s a fundamental difference between when an experienced engineer uses AI as a powerful tool to accelerate work, and when a person with no technical background tries to “vibe code” a production app. The first is an evolution of the profession. The second is a ticking bomb.
Crazy keyboard: my metaphor
I describe vibe coding like this for myself: it’s a crazy high-speed keyboard in your hands. It types tens of times faster than you. It can write in one night what used to take a week.
But if you don’t understand what exactly it’s typing — one day that keyboard will explode. And it’s good if this happens during development, not in production, when a client loses data or money.
Vibe coding without expertise is like giving a Ferrari to someone without a driver’s license. The first 200 meters will be thrilling. And then — a pole.
That’s why I’m still confident: you need to be a specialist and understand technologies to vibe code well. AI speeds up work — but it doesn’t replace understanding of architecture, security, performance, scaling. It doesn’t replace the experience that tells you “stop, something’s wrong here” before the code breaks.
Case #1: WebSocket chat in two nights
Theory is good, but let me show you practice. Kaplia Chat — a full-featured live chat on WebSocket that I wrote literally in two nights and already integrated into my client portal clients.kaplia.pro.
What’s under the hood: Node.js server, WebSocket connections, admin panel on React + Vite, Telegram integration, REST API for external services, webhook system for automation via n8n. Self-hosted, full data control.
What this chat can do:
- Spy Typing — admin sees what the client is typing before sending the message
- Real-time online statuses — who’s on the site right now, what page they’re viewing
- Activity Log — full history of actions: connections, disconnections, tab switching, site navigation
- Spam protection — rate limiting, message length limits
- Message pagination — lazy loading for admin panel and widget
- 10 notification sounds — from Chime to Coin, with preview listening
- Multi-language — Ukrainian, English, Russian
- Mobile responsive — burger menu, swipe gestures for sidebar
Two nights. Not two months. And this isn’t an MVP with two buttons — it’s a production-ready product with documentation, API and full deployment under Nginx with SSL.
Could I have written this without AI? Yes, I could. In two to three weeks. Vibe coding with Claude Code compressed that time by tens of times. But — and this is key — at every step I understood what was happening. I set the architecture, I reviewed the code, I made corrections. AI was a fast executor. I was the architect.
Case #2: AI Talks — when agents talk to each other
Another example — more for fun, but no less instructive. AI Talks (GitHub) — an app created in one day where AI agents autonomously communicate with each other.
You create agents: set a name, personality, choose a voice via ElevenLabs or OpenAI TTS API. Set a topic — and they start talking. With realistic voices, animated avatars, emotions on their “faces”. You can intervene during the conversation, interrupt them — and watch how they react. Sometimes this leads to such funny situations that it’s hard to believe it’s being generated in real time.
Under the hood: PHP backend as an API proxy, Alpine.js + Tailwind on the frontend, Web Audio API for syncing lip animation with audio, support for three LLM providers (OpenAI, Anthropic Claude, Google Gemini) — each agent can use a different model.
One day. A full-featured app with open source code, live demo and documentation. This is what vibe coding can give when the person behind the keyboard understands what they’re doing.
Agentic engineering: what it looks like in practice
Back to Peter Steinberger’s term — agentic engineering. I like it much more than “vibe coding”, and here’s why.
“Vibe coding” suggests you just “catch the vibe” and everything somehow comes together on its own. Reality is different. My workflow with Claude Code looks like this:
- I define the architecture — project structure, stack, code organization approach
- I formulate tasks — clearly, with context, with constraints
- AI generates code — quickly, a lot, often well
- I review every result — check logic, security, architecture compliance
- I correct the course — make changes, refine prompts, sometimes rewrite manually
This is not “vibe”. This is engineering work with an AI agent as a tool. You’re the technical director. AI is a fast but not always correct executor that requires your control and expertise.
Why without expertise — it’s dangerous
I see more and more people who “vibe coded” themselves an app, pushed it to production and sincerely believe everything works. Until the first serious incident.
AI doesn’t think about security the way an experienced developer does. It doesn’t always account for edge cases. It can generate SQL injection if you don’t know what to look for. It can create an architecture that “works” on 100 users and crashes at 10,000.
Vibe coding without understanding is technical debt you don’t even see. And when it comes — and it will come — you won’t be able to fix it because you don’t understand what broke.
New reality: what changed forever
Despite all the warnings, I’m not a skeptic. On the contrary. Vibe coding — or, if you will, agentic engineering — is a new reality that has already changed my work forever. I can’t imagine going back to the times when every line of code had to be written by hand.
Here’s what specifically changed:
Prototyping speed — idea from concept to working prototype in hours, not weeks. Kaplia Chat in two nights, AI Talks in one day — this is not an exception, this is the new norm.
Lowered barrier to experimentation — before, “what if we try it this way?” had no time. Now — try it, see, either keep it or delete it. The cost of an experiment dropped to zero.
Focus on architecture — paradoxically, AI freed up time for exactly what matters most: designing architecture, making strategic decisions, reviewing and optimizing.
Learning effect — working with AI, I constantly see alternative approaches to solving problems. It’s like having a colleague nearby who knows all programming languages at once.
Summary
Vibe coding is not a trend that will pass. It’s a fundamental change in how software is created. But it’s a tool, not a replacement for brains.
If you’re an experienced developer — vibe coding will make you many times more productive. You’ll create in nights what used to take weeks. You’ll be able to experiment, try, build pet projects and turn them into real products.
If you’re a beginner — learn the basics first. Understand architecture, security, algorithms. Then take AI as an accelerator. Not the other way around.
Because a crazy keyboard in the hands of a master — that’s a superpower. And in the hands of someone who doesn’t understand what they’re typing — it’s a bomb that will sooner or later explode.
