Introduction
If you've ever tried to learn AI, you've probably felt overwhelmed by terms like "large model," "Codex," "Skill," and "Prompt" floating around online. This guide breaks down these key concepts using real-life examples, so you can understand them clearly and apply them right away.
1. Large Language Model (LLM)
A large language model is like the brain of an AI system. Different models have different capabilities — some are better at reasoning, others at coding, and some at creative writing.
Example
- Model A: When told "the table is dirty," it only writes a note saying "the table is dirty" instead of cleaning it.
- Model B: When told "the table is dirty," it immediately grabs a cloth and cleans the table.
This shows that the model you choose directly determines how "smart" your AI will be. Common LLMs: GPT-4o, Claude 3, Gemini, Doubao, DeepSeek.
2. Prompt
A prompt is the command you give to an AI. The more specific your prompt is, the more accurate the AI's response will be.
Bad Prompt Example
Give me something to drink.This vague prompt might result in the AI giving you a watering can instead of a beverage.
Good Prompt Example
I am a human. It's very hot today, and I want to drink something cold, sweet, tasty, and cheap.This clear prompt will likely result in the AI giving you a suitable cold drink.
3. Agent
An Agent is an AI employee that can understand tasks, use tools, and work independently to complete them.
Example
When you ask an Agent to find a book: it first checks the fridge (no books there), then searches its internal knowledge base (no books there), and finally checks the external library and finds the book you want. This shows that an Agent can take initiative to solve problems without constant guidance.
4. Codex
Codex is a programming-savvy AI employee that specializes in writing and understanding code.
Example
If you ask an Agent to find a book, it will search for books. If you ask Codex to find a book, it might write a Python script to search for books automatically:
def find_book(book_title):
library = ["The Great Gatsby", "1984", "To Kill a Mockingbird"]
for book in library:
if book_title.lower() in book.lower():
return f"Found: {book}"
return "Book not found"
print(find_book("1984"))5. Skill
A Skill is a standard operating procedure (SOP) for an Agent. It tells the Agent how to perform a specific task consistently.
Example
If an Agent keeps finding the wrong books, you can give it a Skill document that says: "Book Search Skill: 1. Only search the bookshelf in the living room. 2. Look for books with red covers. 3. Check the first page to confirm it's the right book." Now, every time the Agent searches for a book, it will follow these steps and find the right one.
6. Workflow
A Workflow is a sequence of steps that an Agent follows to complete a task. A workflow for finding and returning a book might be: Find the book → Clean the book → Return the book to the shelf.
7. API
An API is a plug that connects an AI to other tools or software. If you want your AI to send emails, you can use an email API to connect your AI to an email service. This way, the AI can send emails without having its own built-in email functionality.
8. Knowledge Base
A Knowledge Base is a database that limits the information an AI can use. If you want your AI to only answer questions about your company's products, you can set up a knowledge base that only contains information about your products. This way, the AI won't use information from the internet or other sources.
Summary
- LLM: The brain of the AI
- Prompt: The command you give to the AI
- Agent: An AI employee that can work independently
- Codex: An AI employee that specializes in coding
- Skill: A standard operating procedure for an Agent
- Workflow: A sequence of steps for completing a task
- API: A plug that connects an AI to other tools
- Knowledge Base: A database that limits the AI's information sources
常见问题
How is this different from the July 2 "Don't Be Intimidated by AI Jargon" article?
The July 2 article covers 12 advanced concepts (Token, RAG, MCP, Context Engineering, Harness Engineering, Workspace Agent, etc.) with code examples — it's designed for intermediate readers who want to understand the full AI stack. This article covers 8 foundational concepts (LLM, Prompt, Agent, Codex, Skill, Workflow, API, Knowledge Base) using everyday analogies — it's designed for absolute beginners who just want to understand what these terms mean. Read this one first if you're brand new. Read the July 2 article when you're ready to go deeper into AI architecture and implementation.
Why do these 8 concepts matter for someone who just wants to use AI tools?
Because every AI tool you'll encounter — whether it's ChatGPT, WorkBuddy, Codex, or Doubao — uses these concepts. LLM determines how smart your tool is. Prompt determines whether you get useful output or garbage. Agent is what separates modern AI tools from simple chatbots. Skill/Workflow is how you automate your work. Knowledge Base is how you make AI answer from your company's documents instead of making things up. Understanding these 8 concepts isn't academic — it directly impacts how effectively you use whatever AI tool you've chosen.
Do I need to remember all the technical details in this guide?
No. Focus on the analogies, not the technical details. LLM = brain. Prompt = command. Agent = employee. Skill = SOP. Workflow = step sequence. API = plug. Knowledge Base = filter. If you remember these one-to-one mappings, you'll have the mental model you need to navigate any AI tool. The technical details in this guide are there so you can recognize the terms when you see them, not because you need to memorize them.
What should I learn after mastering these 8 concepts?
Move on to Practicing with real tasks. Each tool does these 8 things differently. Don't try to memorize — just be aware of these concepts and experience them firsthand.