Go from prompting Claude to building with it
A hands-on, 14-week path through everything "using Claude" actually covers — prompt engineering you can rely on, the Anthropic Messages API and tool use, Claude Code for agentic terminal workflows, and the design patterns behind real AI agents. All 14 weeks are live, ending in a capstone where you ship a real Claude-powered agent.
Prerequisites & tooling
What you should already know
- No prior AI/ML or prompt-engineering experience needed — Weeks 1–5 start from zero
- Comfortable in a terminal and reading basic code helps from Week 6 onward (the API and Claude Code modules)
- Any programming language is fine for the code examples — they lean Python and TypeScript, but the ideas transfer
- Pairs well with the LLM & ML course if you want the theory underneath the model, but stands entirely on its own
Stack you'll work with
A free Claude.ai account gets you through Week 5. An Anthropic API key (a few dollars of usage covers the whole course) unlocks Weeks 6 onward.
The full 14-week roadmap
Five modules: prompting fundamentals, using Claude in daily work, the Anthropic API and tool use, Claude Code for agentic development, and agent design capped by a capstone. All 14 weeks are live below.
How Claude & LLMs Actually Work
Before you write a single prompt, get the mental model right.
- Tokens, context windows & next-token prediction, in plain terms
- The Claude model family — Opus, Sonnet & Haiku, and when each earns its cost
- claude.ai vs. the API vs. Claude Code — one model, three very different jobs
Prompt Engineering Fundamentals
The handful of techniques that separate a vague ask from a prompt that reliably works.
- Being clear and direct: context, constraints & explicit success criteria
- Few-shot examples — showing Claude the shape of a good answer
- XML tags for structuring long or multi-part prompts
Advanced Prompting & Structured Output
Get Claude reasoning out loud, staying in character, and answering in a shape your code can parse.
- Chain-of-thought reasoning, and when it actually helps
- System prompts & role prompting
- Forcing reliable JSON/structured output & reducing hallucination
Claude for Research, Writing & Long Documents
Using claude.ai as a real work surface, not just a chat box.
- Projects & persistent context for work that spans sessions
- Working with long documents, PDFs & large source files
- Drafting, editing & iterating without losing your own voice
Claude for Data & Repeatable Workflows
Where Claude genuinely speeds up your week — and where it doesn't.
- Analyzing data & generating reports from spreadsheets and CSVs
- Building reusable prompt templates for recurring tasks
- A practical efficiency checklist for everyday knowledge work
The Messages API — Requests, Responses & Streaming
Your first line of code talking to Claude instead of a chat window.
- Auth, the Messages API shape & system prompts in code
- Streaming responses token by token
- The Python & TypeScript SDKs, side by side
Tool Use: Giving Claude Real Actions to Take
The feature that turns a chatbot into something that can actually do things.
- Defining tools with JSON Schema
- The tool-use loop: request, execute, return, repeat
- Chaining multiple tools together in one turn
Vision, PDFs & Multimodal Inputs
Beyond text — reading images, documents, and reasoning through harder problems.
- Sending images & reading them accurately
- Native PDF support & long-document handling
- Extended thinking for harder, multi-step problems
Prompt Caching, Cost & Choosing a Model
Ship something that's fast and cheap enough to actually run in production.
- Token economics — what you're actually paying for
- Prompt caching for repeated context
- Picking Opus, Sonnet or Haiku per task, plus the batch API for bulk work
Claude Code Fundamentals
An agentic CLI that reads, edits and runs your code — with you in control.
- Install, your first session & the permission model
- CLAUDE.md — giving Claude Code persistent project context
- The core loop: read, edit, run, verify
Claude Code Power Features — Hooks, Subagents & MCP
The features that turn Claude Code from "helpful" into "part of your actual workflow."
- Hooks for automating behavior around tool calls
- Custom slash commands & subagents for repeatable workflows
- Connecting MCP servers to extend what Claude Code can reach
Real-World Claude Code Workflows
Put it all together on tasks that look like your actual job.
- A TDD loop: write the test, let Claude Code make it pass
- Large refactors & multi-file changes without losing control
- Code review, git/PR workflows & plan mode for risky changes
Designing AI Agents with Claude
The patterns behind "agent," beyond the buzzword.
- Agent design patterns: orchestrator-worker, ReAct & evaluator-optimizer
- When an agent is the right shape — and when a single prompt is
- MCP as a common tool ecosystem across agents
Capstone: Ship a Claude-Powered Agent, End to End
Apply every module in one project — prompting, the API, tool use, and agent design.
- An agent or app built on the Messages API with real tool use
- A working interface people can actually try
- Portfolio-ready writeup of the design decisions you made