C
Vibe Coding//Lesson 01

Vibe Coding

30 min·theory

Vibe Coding

🎯 After reading this lesson

After finishing this lesson, you will be able to confidently do the following 3 things.

  • The core patterns of vibe coding (having AI write code using natural language)
  • 3 mechanisms for saving tokens (the unit of text AI processes at once)
  • ✅ Common code traps that AI tends to create

Keep the learning objectives as a checklist and close the lesson once you can answer all of them.

✨ The People Who Created Vibe Coding — 4 Teams, 4 Panels

01
Andrej KarpathyAndrej Karpathy
AI Researcher · EducatorOpenAI → Tesla → OpenAI → Eureka Labs1986–Present

Coined the term 'Vibe Coding' — naming the era of delegating code to AI

  • 2015 OpenAI co-founder and research scientist
  • 2017 Tesla AI Director — responsible for the Autopilot vision system
  • 2023 Rejoined OpenAI, leading multimodal and educational content initiatives
  • 2025 Coined the term 'Vibe Coding' on X (Twitter) — the phrase spread as a new buzzword
Standard for deep learning educational content. Redefining education for the AI era with Eureka LabsVIBE CODING · Term Creation
02
GitHub Copilot TeamGitHub Copilot Team
AI Pair Programming PioneerMicrosoft + OpenAI2021~Present

Bringing AI pair programming to 1 million developers — the first mainstream code AI

  • 2021 GitHub Copilot technical preview released — powered by the Codex model
  • 2022 Official paid launch — $10/month subscription model
  • 2023 Copilot Chat and GPT-4 integration, expanded IDE integration
  • 2024 Copilot Workspace announced — autonomous handling of issues and PRs
Popularized the 'comment → code' paradigm. Pioneered the AI coding market itselfCOPILOT · AI Pair Programming
03
Anysphere TeamAnysphere Team
Cursor IDE FoundersAnysphere (Cursor)2022~Present

AI-native IDE — the team that rewrote what an editor means with Cursor

  • 2022 Anysphere founded by MIT alumni Sualeh Asif, Michael Truell, and others
  • 2023 Cursor 1.0 launched — VS Code fork with AI integration
  • 2024 Composer and Tab features drove adoption among 1M+ developers
  • 2025 $400M+ ARR achieved, effectively the category leader in AI IDEs
Established the standard for AI-native code editors. Ushering in the IDE = AI eraCURSOR · AI Native IDE
04
Anthropic Claude Code TeamAnthropic Claude Code Team
Agentic Coding PioneerAnthropic2024~Present

An autonomous agent that runs in the terminal — the standard for agentic coding

  • 2024 Claude 3.5 Sonnet's coding capabilities leap forward — new SWE-bench record
  • 2025 Claude Code CLI released — terminal-based autonomous coding agent
  • 2025 Claude Agent SDK released — establishing external tool and MCP standards
  • 2025 Claude 4 series automates multi-file editing, testing, and debugging
Defined the 'agentic coding' category itself. Leading the CLI and MCP standardsCLAUDE CODE · agentic coding
👥
In one line
Karpathy (coined the term) → Copilot (popularization) → Cursor (IDE integration) → Claude Code (agent). Four teams built the vibe coding era.

Why Vibe Coding Is a New Paradigm

In one line: From the era of writing code to the era of expressing intent. AI handles implementation; humans handle direction and verification.


Tool Mapping — The English in each cell is just the tool name; focus on the description

Use CaseStandard Tool
AI inside the editorCursor · Windsurf · VS Code + Copilot (AI is embedded directly in the code editor)
Terminal agent (a mode where AI executes commands on its own)Claude Code · Aider · OpenAI Codex (AI autonomously edits files from the command line)
Instant app generationv0 (Vercel) · Bolt · Lovable (one sentence → a working app)
AutocompleteGitHub Copilot · Codeium (AI suggests the next line as you type)
SPEC (specification — organizing requirements as a document) → codeSpec-Kit · BMAD (give it a design document and it auto-generates code)

5 Core Reasons

ReasonMeaning
Speed5 min vs. 50 min difference for the same feature
Context window (the character limit AI reads at once)Claude 200K·1M tokens = understanding the entire codebase
Agentic workflowAI autonomously edits files, tests, debugs, and commits
MCP (Model Context Protocol — the standard channel for AI to communicate with external tools and databases)AI accesses databases, APIs, and local files all at once
Verification responsibilityThe accuracy of AI-generated code is ultimately the human's responsibility

Key insight: Vibe coding = prompt (instructions sent to AI) engineering + code review skills. Humans focus on intent, verification, and architecture.

🤖 Try asking AI like this

Once you know the concepts in this lesson, you can give AI specific instructions. Instead of a vague 'fix this,' make requests using the right vocabulary — that is the starting point for saving tokens (the unit of text AI processes at once).

  • "How can applying vibe coding concepts to this task reduce token usage?"
  • "Give me 3 example prompts (instructions) for giving AI precise vibe coding-related directions"
  • "Tell me which stage of the agent (a mode where AI uses tools autonomously) workflow this task belongs to"

Why This Saves Tokens

Without knowing the concepts, you have to ask "What does that mean?" again after receiving an AI response. That follow-up question consumes tokens. Learning the concepts once means the conversation ends in one exchange.

Vibe Coding - Vibe Coding