1. What is Vibe Coding?
Vibe Coding (or Vibecoding) is an AI-powered software development method in which programmers describe requirements in natural language and let the Large Language Model (LLM) automatically generate source code.
This term was coined by Andrej Karpathy — AI scientist, co-founder of OpenAI and former AI Director at Tesla — in February 2025:
"There's a new kind of coding I call 'vibe coding', where you fully give in to the vibes, embrace exponentials, and forget that the code even exists. It's possible because the LLMs are getting too good."
— Andrej Karpathy, February 2, 2025
The term quickly became so popular that:
- Merriam-Webster listed it as "slang & trending" in March 2025
- Collins English Dictionary choose it Word of the Year 2025
- Y Combinator reported that 25% of startups in the Winter 2025 batch have a codebase that is 95% generated by AI
2. How Vibe Coding works
Instead of the traditional process think → write code → test → debug, Vibe Coding according to the process:
┌─────────────────────────────────────────────────┐
│ VIBE CODING WORKFLOW │
├─────────────────────────────────────────────────┤
│ │
│ 1. 💭 Describe (Mô tả yêu cầu bằng tiếng Việt/│
│ English cho AI) │
│ ↓ │
│ 2. 🤖 Generate (AI sinh code tự động) │
│ ↓ │
│ 3. 👀 Review/Accept (Xem kết quả, accept/reject)│
│ ↓ │
│ 4. 🔄 Iterate (Nếu chưa đúng, mô tả lại) │
│ ↓ │
│ 5. ✅ Ship (Deploy sản phẩm) │
│ │
└─────────────────────────────────────────────────┘
Karpathy describes his experience:
- Talk to AI via voice (SuperWhisper)
- Always press "Accept All" without reading the diff
- Copy-paste error messages without any further comments
- Code evolves beyond the developer's ability to understand
3. Vibe Coding vs Traditional Coding vs AI-Assisted Coding
| Criteria | Traditional Coding | AI-Assisted Coding | Vibe Coding |
|---|---|---|---|
| Who writes the code? | Developer 100% | Developer + AI support | AI produces 95%+, developer guides |
| Understand the code? | Complete understanding | Understand and review AI code | May not be able to read/understand the code |
| Skills needed | Programming languages | Programming + Prompting | Prompting + Domain knowledge |
| Speed | Slow | Faster | Very fast (for prototype) |
| Code quality | Depends on the developer | High (reviewed) | Not guaranteed |
| Suitable for | Production systems | Every project | Prototype, side projects |
Simon Willison (famous developer) clearly distinguishes:
"If an LLM writes every line of your code, but you've reviewed, tested, and understood it all, that's not vibe coding in my book — that's using an LLM as a typing assistant."
4. Popular Vibe Coding Tools (2026)
4.1. GitHub Copilot (VS Code)
Leading AI coding tool from GitHub/Microsoft, deeply integrated into VS Code:
- Inline Suggestions: suggests code while typing
- Agent Mode: AI automatically implements features across multiple files
- Plan Agent: plan before coding
- Cloud Agent: run agent on cloud, create PR automatically
- Copilot CLI: agent runs from terminal
- Custom Agents & MCPs: expanding AI capabilities
- Supports many models: GPT-5.4, Claude, Gemini 3.1 Pro
4.2. Cursor
Specialized IDE for AI coding, fork from VS Code. Cursor Composer is an outstanding feature for multi-file editing.
4.3. Replit Agent
Cloud IDE platform with integrated AI agent, allowing build and deploy right in the browser.
4.4. Lovable, Bolt, v0
"Prompt-to-app" platforms allow the creation of web applications using only text descriptions.
5. When should Vibe Coding be done?
✅ Suitable:
- Prototype & MVP: validate ideas quickly
- "Software for One": small personal tool (Kevin Roose, NYT)
- Weekend side projects: as Karpathy describes
- Learning: explore new languages/frameworks
- Boilerplate code: CRUD, forms, config files
- Rapid iteration: try many quick approaches
❌ Need to be careful:
- Production systems there are many users
- The system handles sensitive data (healthcare, finance)
- Safety-critical code (medical devices, automotive)
- Complex multi-file architectures not yet documented
- Open source maintainability (paper "Vibe Coding Kills Open Source", Jan 2026)
6. Limitations and risks have been noted
6.1. Security
VeraCode (Oct 2025): LLMs are better at generating functional code but security does not improve accordingly. CodeRabbit (December 2025): AI co-authored code available 2.74x more security vulnerabilities than human-written code.
6.2. Technical Debt
GitClear (2025): Code duplication increases 4 times, code churn nearly doubles, refactoring decreases from 25% to less than 10%.
6.3. Productivity Paradox
METR (Jul 2025): Experienced open-source developers in practice 19% slower When using AI tools, self-assessment is 24% faster.
6.4. Actual Incident
- Loveable (May 2025): 170/1645 apps have vulnerabilities that expose personal information
- Replit (Jul 2025): AI agent deletes production database despite being told not to change anything
- Orchids (Dec 2025): security vulnerability on vibe coding platform
7. Why study this Series?
This series will teach you how Vibe Coding Responsibly — leverage the power of AI while ensuring:
- ✅ Understand code you ship (no blind accept)
- ✅ Security-first mindset when using AI
- ✅ Maintainable code for long-term
- ✅ Professional workflow suitable for production team
- ✅ Optimize productivity real (not an illusion)
Andrew Ng Offer a balanced perspective:
"Vibe coding is a bad name for a very real and exhausting job."
We will learn how to make it one professional skills, not just "vibes".
8. Prerequisites
- Know the basics of programming (any language)
- Have a GitHub account
- Install VS Code
- Basic English reading comprehension (the most effective prompts are usually in English)
9. Summary
| Concept | Key Takeaway |
|---|---|
| Vibe Coding | Programming by describing requirements to AI, may not review code |
| Namer | Andrej Karpathy (Feb 2025) |
| Top tools | GitHub Copilot, Cursor, Replit Agent |
| Suitable | Prototype, side projects, learning |
| Risk | Security, technical debt, maintainability |
| Series goals | Vibe Coding responsibly — fast + safe + professional |
In the next article, we will install and configure GitHub Copilot in VS Code — the main tool we will use throughout the series.