Claude Code Course: agentic coding from first prompt to autonomous agent

Prompt engineering is the art of crafting one-off queries to a model (e.g. ChatGPT, Claude in chat). Agentic engineering is designing systems where an agent (Claude Code, Claude Agent SDK) executes multi-step tasks autonomously — using tools (bash, file editor, MCP), keeping context between steps, and self-correcting. It's the difference between a single prompt and a complex pipeline. The course covers both, but the center of gravity has shifted to agentic.

Yes. Plan Mode is the Claude Code feature that forces the agent to present a step-by-step plan before touching any files. Use it before large refactors, migrations, and brownfield investigations. Module 8 shows 5 situations where Plan Mode is essential (e.g. refactoring 14 files, legacy migration, production debugging) and 3 where it wastes time (hot fix, obvious 1-line change, prototype). After that module, Plan Mode becomes your default before any change larger than 50 lines.

Subagents are parallel agents running in separate context windows that report to an orchestrator (the main agent). In practice, when you ask for an audit of 200 files, the main agent distributes the work across 5–10 subagents, each handles its share, and the orchestrator aggregates the results. Speed-up: 3–5×. Module 6 shows patterns for when subagents help (research, audits, parallel test runs) and when they hurt (sequentially dependent tasks). Hands-on: subagent pipeline that audits a repo in 4 minutes.

Yes. Module 5 (30 pages) is dedicated to MCP servers, including writing your own from scratch in Python and Node.js. Architecture: stdio vs SSE, security model, error handling, deploy. Three ready-made servers as templates (GitHub bridge, Linear bridge, Notion bridge) to fork. After the module you can build your own MCP server for a team workflow in under an hour.

Yes. Computer Use became GA in September 2025 — Claude sees a screenshot and clicks UI coordinates. The course covers use cases (smoke tests after deploy, scraping apps without an API, automating repetitive workflows in tools with no API) as well as limits (slow, expensive, requires supervision). It also shows how to combine Computer Use with Plan Mode: the agent plans what to click, then clicks.

Yes — that is one of the two main use cases for this course (vibe coding). Claude Code writes the code for you; your role is to describe what you want, verify the output, and decide on changes. What you need: patience and logical thinking. What you don't need: a programming language, terminal experience, or a CS course. The first module and a dedicated FAQ section for vibe coders walk you step by step from installing the CLI to your first working application.

Vibe coding is a new way of working with AI where you don't write code yourself — instead you direct Claude (or another agent) through your product vision and it delivers the implementation. You assess, correct, and verify. The course covers both sides: vibe coders (how to direct an agent when you don't know the code yourself) and developers (how to use Claude Code to ship faster in production).

Yes. Claude Code is a CLI agent that works alongside GitHub Copilot, Cursor, and other IDEs. Copilot is autocomplete in the editor; Claude Code is an agent that plans, edits multiple files, runs bash commands, and uses MCP servers. These are different layers of the stack. The course shows how to integrate them and when to use each one.

The full PDF (220 pages) takes 8–10 hours to read. With hands-on work (5 portfolio projects + hook setup + first MCP server) that's a weekend or 7 evenings of about 1 hour each. Lifetime access means you set your own pace.

Yes. Module 6 covers the full Claude Agent SDK (agent loop architecture, tools, memory, subagents). Module 7 covers the Anthropic API (Messages, Batch, prompt caching, tool use, Files, Vision). The complete Anthropic ecosystem in one course.

Cursor is an AI-first IDE (editor with a model inside). Copilot is autocomplete in VS Code. Claude Code is a CLI agent with an agentic loop, hooks, slash commands, MCP servers, and custom skills. Cursor edits, Copilot suggests, Claude Code plans and executes multi-step tasks. The course covers all three and the workflow for each.

Yes. Module 5 (30 pages) is dedicated entirely to MCP. Architecture: stdio vs SSE, writing your own server in Python and Node, 3 ready-made servers as templates (GitHub, Linear, Notion). MCP is a game-changer for developers deploying Claude in production, so it gets thorough coverage.

Module 1 (25 pages) walks you through it: installation on Mac / Linux / WSL, API key setup, first hello-world project, .claude/ folder structure, IDE integration. The only assumption is basic git and bash (cd, ls, cat). If you don't have those, spend one evening on bash + git, then start your first Claude Code session.

No. Claude Code can be run with an Anthropic API key (pay-per-use, no subscription) or with a Claude Pro / Max subscription. The course covers the costs of both options, when each makes sense, and how to choose a model (Haiku 4.5 / Sonnet 4.6 / Opus 4.7) based on your budget and task.

No. A hook is a JSON config in settings.json, optionally with a shell command. Module 4 (25 pages) explains all 5 hook types (PreToolUse, PostToolUse, UserPromptSubmit, Stop, SessionStart), provides 10 ready-made hooks to download, and walks through the hookify pattern with examples. After the module you can write your own hook in 10 minutes.

Yes. Module 6 (25 pages) is dedicated to the Claude Agent SDK. Agent loop architecture, tools, system prompt, file-based memory, multi-step tasks, error handling, parallel subagents. Hands-on: build an agent that autonomously triages GitHub issues. After the module you have a template for every future agent project.

Yes, in two sections: Module 7 (prompt caching 50% off + Batch API 50% off) and Module 9 (cost monitoring, alerts, token budgeting). In practice, students reduce their Anthropic bill by 50–80% within a month. The course pays for itself in the first week of production use.

Yes, but it requires specific patterns: CLAUDE.md per module, an MCP server with code search, slash commands per workflow, and file-based memory. Module 8 covers production workflow patterns for brownfield / legacy / monorepo projects. It shows how Claude Code scales from a solo dev to a 20-person team.

Yes, all updates are free forever. New modules are added when a significant feature ships (e.g. new Computer Use, Memory Tool 2.0, Files API v2). Currently 10 modules; planned growth to 14–16 over the year.

PDF (220 pages) + Discord. No video in v1. Deliberate choice: developers read faster than they watch, and a PDF is easier to search, copy code from, and consume at your own pace. Video may appear in v2 as a bonus.

Yes — you have 14 days to withdraw under consumer protection law. The right to withdraw expires once you access the PDF with your explicit consent. The price is a one-time payment with no subscription.