Master the art of AI-assisted development. Learn to orchestrate intelligent agents, manage context like a pro, and ship 10x faster.
Vibe Coding is a new paradigm where you orchestrate AI agents instead of writing every line of code yourself. You become the architect, reviewer, and conductor—while AI handles the implementation.
"The goal is not to write code faster, but to solve problems at a higher level of abstraction."
— Boris Cherny, Creator of Claude Code
Master these core principles to unlock the full potential of AI-assisted development.
The CLI is your primary interface. It's faster, more powerful, and gives you direct access to the AI agent's full capabilities.
Context is your most precious resource. Learn the 20-30/10 rule to keep your AI agent performing at peak efficiency.
Run multiple AI agents simultaneously. While one builds, another tests, and a third documents.
Trust, but verify. Set up automated checks so the AI can validate its own work before you review.
Choose the right tools for your workflow. Here's what the pros use.

Anthropic's terminal-first AI coding agent. Full shell access, agentic loops, and deep context understanding.
VS Code fork with AI superpowers. Composer mode, Tab completion, and multi-file editing.
Open-source CLI tool for AI pair programming. Great for local models and custom setups.
Codeium's AI-native IDE. Cascade feature for multi-step reasoning and code generation.
The original AI coding assistant. Best for inline completions and quick suggestions.
High-performance editor with built-in AI. Rust-based for speed, with collaborative features.
A systematic approach to AI-assisted development that maximizes quality and speed.

Start with /plan to let the AI analyze the problem and propose a solution before writing any code.
claude
> /plan Implement user authentication with JWT tokensReview the AI's plan. Ask clarifying questions. Challenge assumptions. Iterate until the approach is solid.
What edge cases should we handle for token expiration?Switch to /act and let the AI implement the solution. It will write code, run tests, and iterate.
> /act
[Claude begins implementing...]The AI runs linters, type checkers, and tests automatically. Review the output and fix any issues.
npm test
✓ 12 tests passed
✗ 1 test failed: token expiry edge caseReview the changes, commit with a meaningful message, and update documentation as needed.
git add .
git commit -m "feat: add JWT authentication with refresh tokens"Techniques from Boris Cherny and the vibe coding community to maximize your effectiveness.

Summarize history to save tokens and maintain reasoning quality.
Keep source files under 20%, conversation under 30%, system overhead under 10%.
Use grep/search to find relevant code instead of loading entire files.
Include all build, test, and lint commands so the AI can run them.
Specify naming conventions, patterns, and architectural decisions.
If you correct the AI, add the lesson to CLAUDE.md so it persists.
ESLint, Ruff, or similar tools catch syntax errors immediately.
TypeScript or mypy catch logic errors before runtime.
Give the AI a target to hit with automated verification.
Ask it to review its own code for security, performance, and edge cases.
"Act as a Principal Engineer" triggers more rigorous analysis.
Ask for failure modes before implementing the solution.
Official guide to the CLI agent, slash commands, and configuration.
Mastering the AI-first code editor, Composer, and Tab features.
Standard for connecting AI models to your data and tools.
The original manifesto and evolving playbook for vibe coding.