Understand how Claude works, learn what agents are, and build your first real AI project. The foundation for everything.
What agents are, how they think, and why they're the future of AI applications.
Context windows, tokens, reasoning, and how to structure your prompts for Claude.
How to use Claude Code effectively to write code, debug, and build projects.
Build a real working agent that takes input, reasons, and produces output.
An agent is not just a chatbot. It's an AI system that:
Gets a task, breaks it down, figures out what to do—without you telling it each step.
Runs code, calls APIs, writes files, reads data—actually does things in the real world.
Checks if something worked. If not, tries again. Learns from mistakes in the same session.
Multi-step workflows, large projects, complex decision trees—all automatic.
Real Example: You tell an agent "research product-market fit for AI apps" and it writes Python scripts to scrape data, analyzes trends, writes reports, and emails you the results. All in one go.
Claude has a "working memory" called a context window. It can hold:
Pro tip: Put your most important instructions at the END. Claude pays more attention to recent context.
Everything Claude processes is measured in "tokens":
Pro tip: Be specific. Rambling costs more money AND gets worse results.
Claude has a "thinking" mode (slower, more accurate) and a "speed" mode (faster, good enough):
Claude Code is different from ChatGPT or Copilot. It's not just autocomplete—it's a full coding agent.
What it does: You describe what you want. Claude Code writes it, runs tests, fixes errors, and shows you the result. It can create full projects in minutes.
Native Mac/Windows. Keyboard: Cmd+K (Mac) or Ctrl+K (Windows). Type what you want, Claude Code builds it.
Command line: type claude "your task here". Perfect for automation scripts.
Build a Research Agent that takes a topic and writes a research report.
What it does: You give it a topic (e.g., "latest AI safety research"). The agent researches it, finds real sources, writes a structured report with citations.
What you'll learn:
Time to build: 2-3 hours
Steps:
Instructions you give Claude about how to behave. Like a personality blueprint.
The art of asking Claude the right question to get the best answer (covered in Phase 2).
Asking Claude to "think out loud" before answering. Makes it smarter.
Claude can call external services (like Google Search, Twitter API) to get real data.
Claude's "thinking" mode. It spends tokens thinking before answering. Slower but smarter.
How much information Claude can hold at once. More = better understanding, costs more.