MCP vs Sub-Agents vs Rules vs Skills: The AI Features Everyone Confuses
Most developers have heard these buzzwords. Few actually know what they do.
If you've used Claude Code, Cursor, Codex, Windsurf, or other AI coding assistants, you've probably come across terms like MCP, Sub-Agents, Rules, and Skills.
They sound similar—but they're solving completely different problems.
Let's break them down.
:-)
🧠 Think of an AI Developer as a Human Engineer
Imagine you hire a software engineer.
What would make them productive?
📖 They follow your company rules
🛠 They have specialized skills
👥 They ask specialists for difficult tasks
🌐 They use external tools like GitHub, Slack, Jira, databases, browsers, etc.
That's exactly what these four concepts provide.
:-)
📜 1. Rules – Tell AI How to Behave
Rules are permanent instructions.
Examples:
Always write TypeScript
Never use any
Follow Clean Architecture
Use Tailwind CSS
Write unit tests
Don't modify generated files
Think of Rules as your team's coding standards.
Purpose: Control AI behavior.
:-)
🛠 2. Skills – Teach AI How to Do Something
Skills contain reusable knowledge.
Examples:
Create a React component
Review security issues
Build a REST API
Write Docker files
Generate documentation
Design database schemas
Instead of explaining every time, the AI loads the right skill when needed.
Purpose: Give AI expertise.
:-)