# Agentic Engineering Course

**From Foundations to Production — build, deploy, and maintain production-grade AI agent systems.**

## Quick Start

```bash
# Read this first:
less GETTING-STARTED.md

# Then start with Module 1:
less M1-FOUNDATIONS.md

# Need API keys?
less API-KEYS.md

# Something broken?
less TROUBLESHOOTING.md
```

## Course Map

| Module | Title | Est. Hours | Prerequisites |
|--------|-------|------------|---------------|
| M1 | Foundations | 4-6 | None |
| M2 | Architecture | 6-8 | M1 |
| M3 | Safety & Security | 5-7 | M1 |
| M4 | Multi-Agent Orchestration | 7-9 | M1, M2 |
| M5 | Production Patterns | 5-7 | M1, M2 |
| M6 | Economics & Evaluation | 4-6 | M1 |
| M7 | Advanced Topics | 5-7 | M1-M6 |
| M8 | Capstone | 8-12 | M1-M7 |

## File Structure

```
course/
├── 00-CURRICULUM.md         # Full curriculum with lesson plans
├── M1-FOUNDATIONS.md        # What are agents, harness vs model, decision frameworks
├── M2-ARCHITECTURE.md       # Tools, loops, context, memory, codebase patterns
├── M3-SAFETY.md             # 5-level security ladder, hooks, verifier pattern
├── M4-ORCHESTRATION.md      # Multi-agent patterns, delegation, P2P comms
├── M5-PRODUCTION.md         # CI/CD, shadow deploys, monitoring, rollback
├── M6-ECONOMICS.md          # Model pricing, cascade routing, evals
├── M7-ADVANCED.md           # Autoresearch, meta-agents, Beyond MCP
├── M8-CAPSTONE.md           # 3 project options with phases and rubrics
├── NON-TECHNICAL.md         # 7 decision frameworks for managers/PMs
├── ASSESSMENTS.md           # 56 quiz questions + answer keys + rubric
├── COMPETITIVE-ANALYSIS.md  # vs Anthropic, TAC, DL.AI, ClaudeFAST
├── DEBATE.md               # Armin vs George — the two poles of agent debate
├── FEYNMAN.md               # 15 core concepts explained in plain language
├── FIELD-MANUAL.md         # Consolidated technical quick reference (11 sections)
├── JEFF-INTEGRATION.md     # Jeff Emanuel's 14-tool flywheel mapped to course
├── REFERENCE-STACK.md       # Production 5-tool architecture
├── TOOL-REFERENCE.md        # 5-tool comparison reference
└── labs/
    ├── L1-first-agent/      # Your first single-tool agent
    ├── L2-multi-tool/       # Add file + search tools
    ├── L2-context/          # Sliding window + summarization
    ├── L3-whitelist-hook/   # L4 security whitelist implementation
    ├── L3-verifier/         # Read-only verification agent
    ├── L4-agent-chain/      # YAML-defined plan→build→review pipeline
    ├── L4-multi-team/       # Multi-team system setup
    ├── L5-observability/    # SQLite-based tool call tracing
    ├── L5-cicd/             # Golden dataset + regression gate
    ├── L6-eval-harness/     # pass@k evaluation system
    ├── L6-cost-optimization/# Cascade routing optimizer
    ├── L7-autoresearch/     # Self-improving experiment loop
    └── L7-meta-agent/       # Agent that builds agents
```

## How Each Module Is Structured

```
Lesson X.Y: Title
    Type: Concept | Technical | Architecture | Lab | Assessment
    Est. Time: XX min
    
    Body:
    - Core teaching points
    - Examples from tac/ repository
    - Code snippets where applicable
    
    Key sources in your repo: [file paths]
```

## Resource Index

Every lesson references specific files from the `tac/` directory:

| Module | Primary Sources |
|--------|----------------|
| M1 | single-file-agents/sfa_poc.py, mythos-learnings.md |
| M2 | single-file-agents/codebase-architectures/, 20 sfa_*.py files |
| M3 | bash-damage-from-within/, the-verifier-agent/, damage-control.ts |
| M4 | lead-agents/, ui-agents/, ceo-agents/, pi-vs-claude-code/ |
| M5 | paperclip/doc/DEPLOYMENT-MODES.md, hooks-multi-agent-observability/ |
| M6 | aiproxy/benchmarks, benchy/, agentic-coding-tool-eval/ |
| M7 | mythos-learnings.md, brand-monitor/autoresearch, beyond-mcp/ |
| M8 | brand-monitor/, the-verifier-agent/, ceo-agents/ |
| Reference | REFERENCE-STACK.md, TOOL-REFERENCE.md, COMPETITIVE-ANALYSIS.md |

## What This Course Covers That No Other Does

- **Original security research**: mythos-learnings (reward hacking, grinding, confabulation cascades from Anthropic's 244-page paper)
- **Hands-on security ladder**: bash-damage-from-within L1→L5 with runnable demos and the L3 marquee break
- **Production verifier pattern**: read-only agent that independently verifies builder's work
- **Autoresearch loop**: brand-monitor's self-improving agent with real latency data (52→46ms)
- **Complete multi-team architecture**: depth-3 delegation (orchestrator → leads → workers) from your own ceo-agents/lead-agents/ui-agents
- **Model economics**: LLM pricing 100x range, cascade routing, the 3x rule, pass@k math