technical-decision-record

You use Cursor. Or Copilot. Or both. You've gotten faster at writing code.

But you're still just using autocomplete with extra steps.

• Your .cursorrules file is ignored half the time
• Context resets every session
• You're prompting, not architecting
• AI helps you code, but doesn't know YOUR stack

You're using AI. You're not building with AI.

This course fixes that.

---

```

CURSOR/COPILOT CLAUDE CODE

───────────────────────────────────────────────────────────────────

Single-file autocomplete → Multi-file agents with memory

"Chat in sidebar" → Skills that encode YOUR expertise

.cursorrules (ignored) → CLAUDE.md (sacred architecture)

Generic AI assistant → Domain expert that knows your stack

Every session starts fresh → Compound knowledge across sessions

───────────────────────────────────────────────────────────────────

```

Duration: 3.5 hour livestream (built in public)

Outcome: Your first Claude Code skill shipped, production setup complete

---

| Segment | You'll Ship | Time |

|---------|-------------|------|

| [0. Cold Open](segments/00-cold-open.md) | See a working agent demo | 5 min |

| [1. Architecture](segments/01-architecture.md) | CLAUDE.md for your project | 20 min |

| [2. First Skill](segments/02-first-skill.md) | Complete skill with SKILL.md | 40 min |

| [3. Plugin Architecture](segments/03-plugin-architecture.md) | Agent + command | 35 min |

| [4. MCP Integration](segments/04-mcp-integration.md) | External API connection | 25 min |

| Break | Catch up | 10 min |

| [5. Telemetry](segments/05-langsmith-telemetry.md) | LangSmith tracing | 20 min |

| [6. Guardrails](segments/06-git-guardrails.md) | Safety constraints | 15 min |

| [7. Domain Expertise](segments/07-domain-expertise.md) | Your framework as a skill | 25 min |

| [8. Subagents](segments/08-subagent-patterns.md) | Specialist delegation | 15 min |

| [9. Production Setup](segments/09-production-checklist.md) | Complete, verified setup | 10 min |

| [10. Team Adoption](segments/10-team-adoption.md) | Rollout playbook | 30 min |

Every segment ends with something you built yourself.

---

This course is for you if:

• [ ] You use Cursor, Copilot, or similar AI coding tools
• [ ] You want to go beyond autocomplete to building agents
• [ ] You have domain expertise worth encoding (frameworks, patterns, processes)
• [ ] You want your team to adopt AI-assisted development systematically

This course is NOT for you if:

• You've never used AI coding tools (start with basics first)
• You just want faster autocomplete (Cursor is fine for that)
• You don't have a specific domain or codebase to apply this to

---

1. Live Build, Not Lecture

I build everything live during the stream. You follow along. No slides-only segments.

2. Real Patterns from Production

Every pattern comes from my actual Claude Code setup: 17 custom skills, multiple plugins, LangSmith traces showing real workflows.

3. Encode YOUR Expertise

By the end, you've encoded your own framework/process as a Claude Code skill. Not a toy example—your actual workflow.

4. Team-Ready

The final segment is a complete team adoption playbook. This isn't just personal productivity—it's how you roll out to your org.

---

This course uses exported LangSmith traces to show exactly how Claude Code works:

| Trace | What It Shows |

|-------|---------------|

| [Skill Activation](traces/) | How skills load and route |

| [Agent Workflow](traces/) | Multi-step agent execution |

| [MCP Integration](traces/) | External API calls |

| [Subagent Delegation](traces/) | Specialist handoff |

Not conceptual diagrams. Actual traces from production use.

---

##