Skip to content

Latest commit

 

History

History
71 lines (53 loc) · 2.22 KB

File metadata and controls

71 lines (53 loc) · 2.22 KB

AI Native Development Workshop

Project Overview

This is a workshop project teaching AI-native development practices to engineers. The workshop covers multi-agent systems, context engineering, and building applications with Claude Code.

Workshop Focus Areas

  • AI Development Spectrum: Understanding vibe coding vs code assist vs AI-native development
  • Context Engineering: The core skill for effective AI-assisted development
  • Multi-Agent Workflows: Orchestrating parallel agents for complex tasks
  • MCP Servers: Extending AI capabilities with external integrations

Tech Stack for Demo

  • Frontend: Next.js
  • Backend: Convex
  • AI Tooling: Claude Code CLI
  • MCP: Context7 for documentation lookup

Key Concepts to Demonstrate

Plan Mode vs Agent Mode

  • Use Plan Mode for complex or risky changes - think first, execute later
  • Use Agent Mode for well-defined tasks with autonomous execution
  • When unsure, start with Plan Mode

Context Engineering Principles

  • Input quality determines output quality
  • Structure rules files well
  • Provide relevant examples
  • Reference existing patterns in the codebase
  • Manage token limits by prioritizing relevant information

Model Selection

  • Opus 4.5: Deep reasoning, complex architecture decisions
  • Fast models: Quick refactors, small fixes
  • Gemini 3.0: Visual/UI understanding
  • Match the model to the task at hand

Workshop Best Practices

  1. Break down tasks effectively
  2. Set clear boundaries for each agent
  3. Review at checkpoints, not just at the end
  4. Stay in control - automation serves you
  5. Use @mentions strategically to manage context

File Structure

.claude/
  commands/     # Custom slash commands (/review, /test, /deploy)
  mcp.json      # MCP server configuration
  rules.md      # Project-specific rules (alternative to CLAUDE.md)

MCP Servers Used

  • Context7: Real-time documentation lookup during coding
  • Additional servers can be configured in .claude/mcp.json

Demo Flow

  1. Initialize Next.js + Convex project
  2. Set up context rules
  3. Plan mode: outline app structure
  4. Agent mode: implement features
  5. Use Context7 MCP for docs lookup
  6. Code review workflow
  7. Iterate based on feedback