Tags: vibe-coding, ai, llm, agentic, prompting, cursor, cline, copilot, beginner Last updated: 2026-06-27

Vibe Coding Cheatsheet

What Is Vibe Coding?

Vibe coding is a new way to build software where you describe what you want in plain English and an AI assistant writes the code for you. You “vibe” with the AI — you review, accept, reject, and guide its output in a conversation instead of typing every line yourself.

The term was coined by Andrej Karpathy (co-founder of OpenAI, former Tesla AI lead) in early 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.

In practice, it means you prompt an AI, watch it generate entire files, iterate through conversation, and only intervene when the output isn’t what you intended. It’s like pair-programming with a tireless junior developer who types at 10,000 words per minute.

Why Vibe Code?

BenefitWhat It Means
SpeedBuild a full prototype in hours, not weeks
Low barrierNo need to memorise syntax or APIs
ExplorationTry 5 approaches quickly, keep the best one
LearningWatch the AI solve problems and learn patterns
AccessibilityNon-engineers can build working software

Core Terminology

TermDefinition
LLM (Large Language Model)The AI brain (Claude, GPT-4, Gemini). Predicts the next token in a sequence.
PromptThe natural-language instruction you give the AI. Good prompts = good code.
Context WindowHow much conversation the AI can remember. Larger windows = more code per session.
Agent / AgenticThe AI reads files, runs commands, creates files, and loops autonomously.
BYOK (Bring Your Own Key)You provide your own API key. Pay per generation, no subscription lock-in.
MCP (Model Context Protocol)Open standard for AI agents to connect to external tools (databases, APIs, game engines).
ScaffoldingGenerating the boilerplate folder structure, config files, and package setup.
HallucinationThe AI invents functions or APIs that don’t exist. Review everything.
Vibe CheckRunning AI-generated code to see if it works. Pass/fail, usually a surprise.
TokenThe unit LLMs read. ~0.75 words per token. Billed per token (input + output).

Types of Vibe Coding

Prompt-Driven (Chat)

Agentic (In-IDE)

Multi-File / Project-Gen

Multi-Agent

The Vibe Coder’s Workflow

  1. DESCRIBE → “Build a to-do app with dark mode and local storage.”
  2. GENERATE → AI creates the project, files, and code.
  3. VIBE CHECK → Run it. Does it work?
  4. ITERATE → “Add drag-and-drop reordering.” AI updates code.
  5. REFINE → “Make it faster.” AI suggests optimisations.
  6. SHIP → Deploy. Move on to the next feature.

Choosing Your Stack

You AreTry
Total beginner, never codedChatGPT + Replit Agent
Learning to codeCursor or Cline in VS Code
Indie game devCline + Unity/Godot (BYOK)
Web dev, shipping fastCursor or Windsurf
Building a SaaSCline with MCP servers
Want everything in browserBolt.new or v0.dev
Enterprise / large teamsGitHub Copilot Agent Mode

Essential Skills to Learn

SkillWhy It Matters
Prompt writingVague instructions = vague code. Be specific about tech stack, patterns, and constraints.
Reading codeYou don’t need to write it, but understanding the output is essential for debugging.
Reviewing & rejectingAI makes mistakes. Learn to spot hallucinations, over-engineering, and security flaws.
Version control (Git)Commit after every working iteration so you can roll back. AI changes are fast and sometimes destructive.
DebuggingWhen the vibe check fails, you need to describe the error back to the AI clearly.
System designAI can build components; you need to know how they fit together.

Common Mistakes

MistakeFix
Prompt too vagueSpecify language, framework, file structure, and style preferences.
Not reading generated codeReview every file before accepting. AI invents functions that don’t exist.
No version controlCommit before every major AI change. AI can nuke your working code.
Ignoring context limitsLong sessions degrade quality. Start fresh sessions for new features.
Blindly trusting AITest everything. AI doesn’t understand your project the way you do.
Under-specifying securityTell AI explicitly about auth, validation, and sanitisation requirements.

History & Milestones

WhenWhat Happened
2020GitHub Copilot launches — first mainstream AI code assistant (autocomplete style).
Late 2022ChatGPT released. Developers realise they can prompt for entire functions.
2023GPT-4 improves code generation significantly. Cursor IDE launches with AI-first UX.
Early 2024“Agentic” coding emerges — AI runs terminal commands, reads files, writes multi-file changes.
Late 2024Cline launches as open-source BYOK VS Code agent. MCP protocol standardises tool connections.
Early 2025Andrej Karpathy coins “vibe coding” — describes building apps entirely through AI conversation.
Mid-2025Multi-agent frameworks and AI project generators become mainstream.
NowVibe coding is a recognised development paradigm. You’re holding the cheatsheet.

Philosophy

Vibe coding isn’t about replacing developers. It’s about shifting from “how do I write this?” to “what do I want to build?”