Computers, Servers & the Internet

Before touching AI, you need one mental model: every piece of software is just computers talking to other computers over wires.

What is a Server?

A server is just a computer that is always on and waits for requests. When you open Instagram, your phone sends a request to Meta's servers. The server finds your photos and sends them back. That's it.

How the Internet Works (Simply)

What is an API?

An API (Application Programming Interface) is a defined way for programs to communicate. Instead of a human visiting a website, your code sends a structured request and gets data back.

# Calling the Claude API from Python — this is how you "talk to AI" from code
import anthropic

client = anthropic.Anthropic(api_key="your-key-here")

message = client.messages.create(
    model="claude-sonnet-4-5",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Summarize this email for me"}]
)

print(message.content[0].text)
# → "The email is about a meeting scheduled for Friday at 3pm..."

Key Terms Cheatsheet

What to Learn and Why

You don't need to master every language. For AI and automation, a focused stack beats broad shallow knowledge every time.

The Honest Priority Order

Python — The Essentials for AI Work

# Variables & data types
name = "Claude"
temperature = 0.7
messages = []          # list
config = {"model": "sonnet"}  # dictionary — used EVERYWHERE in APIs

# Functions — reusable blocks of logic
def ask_claude(question):
    response = client.messages.create(
        model="claude-sonnet-4-5",
        messages=[{"role": "user", "content": question}]
    )
    return response.content[0].text

# Calling your function
answer = ask_claude("What is machine learning?")
print(answer)

AI & ML — The Big Picture

Most people confuse these terms. Here's a clear, permanent definition that won't mislead you.

The Hierarchy

How an LLM Actually Works (Simplified)

Key ML Concepts You'll Encounter

Claude Models, API & Prompting

Anthropic's Claude is the most capable and safest frontier AI. There are three model tiers — choosing the right one saves cost and latency.

Claude Model Families

Prompt Engineering — Getting the Best Results

The difference between a mediocre AI output and an exceptional one is almost always the prompt. Here's what actually works:

# BAD prompt — vague, no context
"Summarize this"

# GOOD prompt — role, task, format, constraints
system_prompt = """You are an expert business analyst.
Summarize the following email in exactly 3 bullet points.
Focus on: action items, deadlines, and decisions made.
Be concise — each bullet should be under 20 words."""

user_message = "[paste email here]"

response = client.messages.create(
    model="claude-sonnet-4-5",
    system=system_prompt,
    messages=[{"role": "user", "content": user_message}],
    max_tokens=500
)

Prompt Engineering Principles

Tool Use — Giving Claude Hands

Claude can call external tools (functions) when it needs to fetch data, run code, or interact with systems. This is the foundation of agents.

# Define a tool Claude can use
tools = [{
    "name": "get_weather",
    "description": "Gets current weather for a city",
    "input_schema": {
        "type": "object",
        "properties": {
            "city": {"type": "string", "description": "City name"}
        },
        "required": ["city"]
    }
}]

# Claude decides WHEN to use the tool based on the conversation
response = client.messages.create(
    model="claude-sonnet-4-5",
    tools=tools,
    messages=[{"role": "user", "content": "What's the weather in Jaipur?"}]
)
# Claude responds with: tool_use block → {name: "get_weather", input: {city: "Jaipur"}}

Model Context Protocol

MCP is the USB standard for AI. It lets any AI model connect to any data source or tool through a single, consistent interface.

Why MCP Exists

Before MCP, every AI integration was custom code. You'd write a different connector for Gmail, Slack, your database, GitHub — all differently. MCP standardizes this: write one server, any MCP-compatible AI can use it.

MCP Architecture

What MCP Servers Can Expose

Popular MCP Servers Today

# Simple MCP server in Python (FastMCP library)
from mcp.server.fastmcp import FastMCP

mcp = FastMCP("My Company Tools")

@mcp.tool()
def get_customer_count(date: str) -> str:
    """Get number of new customers on a date"""
    # Connect to your database here
    result = db.query(f"SELECT COUNT(*) FROM customers WHERE date='{date}'")
    return f"New customers on {date}: {result}"

# Now Claude can answer: "How many customers joined last Monday?"

Building AI Agents

An agent is an AI that doesn't just answer questions — it plans, takes actions, observes results, and keeps going until a goal is achieved.

The Agent Loop

Multi-Agent Systems

Complex tasks benefit from multiple specialized agents working together — like a team rather than one person doing everything.

Agent Memory Types

Automation Workflows with Claude

Automation means getting computers to do repetitive work so humans can focus on thinking. Claude is the brain; workflow tools are the skeleton.

The Automation Stack

Real Automation Examples You Can Build

Example 1 — AI Email Triage

Example 2 — Competitor Intelligence Bot

Example 3 — Customer Support Automation

Example 4 — Code Review Agent

n8n — The Recommended Starting Tool

n8n is open-source, self-hostable, and has a visual workflow builder. It connects 400+ apps and has a built-in Claude/OpenAI node. You can start free at n8n.io.

Cloud — Deploy Your Systems

Cloud = someone else's computers you rent by the hour. AWS, Google Cloud, and Azure are the three giants. You only pay for what you use.

Why Cloud Matters for AI Systems

Cloud Services You'll Actually Use

Docker — The Packaging Standard

Docker packages your application with all its dependencies into a portable container. Works on your laptop, works in the cloud, works everywhere exactly the same.

# Dockerfile — how to package your Claude agent
FROM python:3.11-slim

WORKDIR /app
COPY requirements.txt .
RUN pip install -r requirements.txt

COPY . .

CMD ["python", "agent.py"]

---
# Deploy to Google Cloud Run (simplest production option)
gcloud run deploy my-claude-agent \
  --source . \
  --region us-central1 \
  --set-env-vars ANTHROPIC_API_KEY="your-key"

Your 90-Day Plan

Concrete, project-based. Every month ends with something real you've built and can show to anyone.

Month 1 — Foundations

Month 2 — AI Systems

Month 3 — Production