AI Training for Developers

A hands-on training course on building software with AI. Four 90-minute sessions progressing from greenfield development to automated software factories.

---

Sessions

#	Session	Level	Focus
1	Greenfield Development	Beginner	Build from scratch — tools, prompting, specs, TDD
2	Brownfield Development	Intermediate	Navigate existing codebases — document, spec changes, code review
3	AI Agents	Advanced	Build AI agents — tools, memory, RAG, skills, subagents
4	Software Factories	Advanced	Automated delivery — role-based policies, spec loops, E2E testing

---

Session Summaries

Session 1: Greenfield Development

Build a working application from scratch using AI. Choose from three projects:

• Todo CLI — Command-line todo manager with categories and priorities
• Weather Dashboard API — REST API for weather data with caching
• Bookmark Manager API — REST API for saving, tagging, and searching bookmarks

Learn: AI tools landscape, rules/configuration, effective prompting, spec-driven development, TDD with AI.

Session 2: Brownfield Development

Work with Suroi (~114k lines of TypeScript), a real open-source battle royale game. Learn to:

• Document unfamiliar codebases with AI (3-tiered documentation)
• Spec changes using SDD (Spec-Driven Development)
• Code review changes against specifications

Learn: ModernPath 4-step process, AI-friendly architecture, documentation plans.

Session 3: AI Agents

Design, build, and rehearse practical AI agents with tools, memory, and retrieval:

• Exercise 1: Invent a CLI tool using AGENTS.md workflow
• Exercise 2: Build RAG-backed memory (embed, store, retrieve)
• Exercise 3: Combine tools + memory + guardrails into a full agent

Includes pre-built agents: lunch-selection-agent, weather-forecast-agent, prospecting-agent, tes-agent, holiday-planner.

Session 4: Software Factories

Move from individual agent usage to repeatable software factory workflows:

• Rehearsal 1: Minimal factory — single task execution
• Rehearsal 2: Role-based policies + session resume
• Rehearsal 3: Spec-driven delivery loops (implement → check → review)
• Rehearsal 4: Agent Factory — automated agent creation with E2E testing

Learn: Factory anatomy, role policies (analyzer/fixer/planner), spec-driven loops, quality gates.

---

Setup

Node.js (Sessions 1–2)

npm install

Python (Sessions 3–4)

# Install Python dependencies
pip install -r requirements.txt

# Set API key for Gemini
export GEMINI_API_KEY=<your_key>

# For Session 4, install Claude SDK
pip install claude-agent-sdk

CLI Tools (Session 4)

At least one CLI tool with active subscription:

claude --version   # Claude Code CLI
codex --version    # OpenAI Codex CLI
opencode --version # OpenCode CLI

---

Project Structure

README.md                        # This file
AGENTS.md                        # Agent configuration, workflows, documentation templates

session-1-greenfield/            # Session 1: Greenfield projects
├── a-todo-cli/                  # Todo CLI project
├── b-weather-api/               # Weather Dashboard API project
└── c-bookmark-api/              # Bookmark Manager API project

session-2-brownfield/            # Session 2: Brownfield work on Suroi
├── prompts/                     # Step-by-step prompt files
└── docs/                        # Documentation outputs

session-3-ai-agents/             # Session 3: AI Agents
├── agents/                      # Pre-built agent implementations
│   ├── lunch-selection-agent/   # Restaurant discovery + recommendations
│   ├── weather-forecast-agent/  # Weather forecasts + alerts
│   ├── prospecting-agent/       # B2B prospect search
│   ├── tes-agent/               # Finnish collective agreements
│   └── holiday-planner/         # Family holiday planning
├── exercises/                   # Hands-on exercises
├── gemini_agent.py              # Base agent template
└── index_site.py                # RAG indexing script

session-4-agentic/               # Session 4: Software Factories
├── claude-factory-rehearsal/    # Factory rehearsal scripts
│   ├── 01_minimal_factory.py    # Basic factory pattern
│   ├── 02_factory_catalog.py    # Role-based policies
│   ├── 03_resumable_factory.py  # Session resume capability
│   ├── 04_spec_loop_factory.py  # Spec-driven delivery loop
│   ├── 05_agent_factory.py      # Complete agent creation factory
│   ├── backend_runner.py        # Multi-backend adapter
│   └── spec.*.json              # Spec examples
└── AGENTS.md                    # Factory framework documentation

.agents/skills/                  # CLI tool documentation (SKILL.md per tool)
tools/                           # CLI tool implementations
slides/                          # Presentation slides

---

Key Concepts

Concept	Session	Description
Context Sandwich	1	Prompt structure: context → task → constraints
Spec-Driven Development	1, 2	Write specs before code, use AI to implement
3-Tiered Documentation	2	High-level → Subsystem → Module docs
Agent Loop	3	Plan → Execute → Observe → Repeat
RAG	3	Retrieval-Augmented Generation for memory
Software Factory	4	Policy wrapper: goal + tools + quality gates
Spec Loop	4	Implement → Check → Review until approved

---

Quick Links

• Session 1 README
• Session 2 README
• Session 3 README
• Session 4 README
• AGENTS.md — Workflow templates and conventions