Kevin Stratvert - Claude Code Tutorial for Beginners

Short Summary

Main recommendation: Use Claude Code, Anthropic's AI coding assistant, to build and change apps through plain-language prompts, following a plan-first workflow: start in plan mode, let Claude ask clarifying questions, review the plan, then implement with frequent Git commits. The video demonstrates this by building a chore-management app from scratch, adding a fix to an existing app, and then making a real-time architecture change.

Key Points

• Start prompts with a bulleted list of high-priority requirements, then tell Claude to ask clarifying questions about requirements, engineering principles, and hard constraints; the presenter says this produces a richer planning discussion than a vague "make an app" prompt.
• Plan mode is the key safeguard: Claude will not write code until you confirm, so you can catch design choices you disagree with (for example, the plan defaulted to JavaScript, which the presenter changed to TypeScript).
• Commit early and often, so any regression from later prompts can be reverted to the last known good state.
• A CLAUDE.md file onboards Claude onto an existing codebase by describing the tech stack and structure, what each component does, and how Claude should work (such as running tests after every change); the presenter says the contents are injected into every session.
• For larger changes, the presenter uses Claude Code's recommended explore, plan, code, and commit approach, and asks Claude to explore relevant files and present three options before committing to one.

Details Worth Noting

• Concrete instructions:
  • Install Claude Code on Mac by copying the bash script from the homepage and running it in a terminal. On Windows, the provided script is PowerShell, which the presenter had trouble with, so use npm instead: install Node.js from nodejs.org, confirm with node -v, then run the npm install command copied from the video description's resources section, and run claude to set it up (choose a text style, then sign in with your Claude account).
  • In VS Code, open a terminal with Control+Shift+backtick, run git init, open a Command Prompt from the terminal dropdown, run claude, and press 1 to trust the directory.
  • To enter plan mode, hold Shift and press Tab until "plan mode" appears in the bottom left. Use Alt or Option+Enter to add new lines in the prompt.
  • Use the arrow keys to move between clarifying questions and press 1 to submit answers.
  • When reviewing a plan, select option three to request changes, type the change, and press Enter.
  • Choose "auto-accept edits" to speed up implementation, or manually approve each change if you are new to the tool.
  • Toggle "YOLO mode" by running Claude with the --dangerously-skip-permissions switch, then use Shift+Tab to switch to plan mode within it; in YOLO mode Claude runs commands without asking. The presenter warns of risk and suggests it only when you trust the tool.
  • Create CLAUDE.md with the prompt from the video's resources section (the /init command did not produce the three layers the presenter wants); view it formatted with Command+Shift+V (Control+Shift+V on Windows).
  • For bug fixes, use a short prompt that names it as a bug fix, describes the failure, and asks Claude to create a Git branch first.
  • Use /clear to clear conversation history and free context before working on existing code.
• The presenter notes that when Claude suggested a web server for real-time sync, the team member change propagated across two browser windows, and the presenter recommends reviewing generated code rather than trusting it blindly.
• The presenter treats Claude like a senior engineer on the team: direct, specific prompts, with code reviewed the same way as anyone else's.

Bottom Line

Claude Code works best as a planning partner rather than a one-shot code generator: clarify requirements in plan mode, document your standards in CLAUDE.md, commit frequently, and review everything it writes.