Build Your AI Development Environment
A ten-lesson curriculum that takes you from a fresh install to an orchestrated ecosystem of specialist agents — instructions, skills, agents, and hooks, each doing the job it's built for.
Tool-portable by design. Examples use the GitHub Copilot CLI, but the concepts apply to Claude Code, Cursor, or any agentic tool.
Now playing · 01 — Foundation Setup
Ten lessons,
one arc
Conductor thinking — you direct the specialists who do. The palette advances through the spectrum as the curriculum does: coral at install, violet at the business ecosystem.
Foundation Setup
Install the Copilot CLI, create your global instructions (your personal orchestrator agent), set up your dotfiles repo so your config follows you to every machine.
- Install and authenticate the Copilot CLI
- Write your global instructions — the orchestrator that speaks for you
- Version your dotfiles so config follows you everywhere
The Context Hierarchy
How instruction layers stack and merge — what loads automatically vs. what you trigger, and the four-layer safety net (AI → editor → hooks → CI).
- Trace how global, project, and directory instructions merge
- Separate what loads automatically from what you trigger
- Map the four-layer safety net: AI → editor → hooks → CI
Automation isn't one thing — it's four layers, each with a different trigger. From always-on passive context down to mechanical shell scripts that fire at git events with no AI involved.
Four Automation Layers
The most important mental model in the course: instructions, skills, agents, hooks — why each belongs in its own layer, and why agents don't fire automatically (that's a feature).
- Place each capability in the right automation layer
- Understand why agents don't fire automatically — and why that's a feature
- See the trigger difference across all four layers (see diagram above)
The Agent Ecosystem
The 8-category specialist taxonomy and the full pipeline with feedback loops — orchestrator, planners, learners, doers, feedback, guardians, tool operators, presenters.
- Learn the 8-category specialist taxonomy
- Trace the full pipeline, feedback loops included
- Read before you write — where learners sit in the flow
Intent flows left to right through eight specialist categories. It's not one-way: Feedback and Guardians validate in parallel and loop issues back to the Doers before anything ships. Example agents shown as sublabels.
Content Types
The decision tree — agent vs. workflow vs. skill vs. instruction vs. prompt — and building your first workflow agent.
- Run the decision tree for any capability you need
- Tell agent, workflow, skill, instruction, and prompt apart
- Build your first workflow agent end to end
Five questions, five endpoints. Ask them in order — the first "yes" tells you what to build.
Building Real Agents
A complete custom agent from description field through hard rules and output format. Validators before doers.
- Write a precise description field — the agent's trigger surface
- Encode hard rules and a strict output format
- Build validators before doers
Workflow Agents & GitFlow
Git hooks for the mechanical layer, CI/CD enforcement, and a one-command pre-PR pipeline. Hooks enforce. Agents judge. You decide.
- Wire git hooks for the mechanical layer
- Enforce standards in CI/CD
- Run one command for the whole pre-PR pipeline
Each station on the path to a merged PR is handled by a specific layer — a hook, CI, or a workflow agent. Hooks enforce. Agents judge. You decide.
-
Write code
$ code .You -
Commit
$ git commitpre-commit + commit-msg hooks -
Push
$ git pushCI runs -
Pre-PR gate
$ pre-prworkflow agent -
PR opened → merged
$ gh pr mergeYou decide
Skills & MCP
Build a skill, set up MCP servers, understand the security model, and know when to use each tool type.
- Build a skill from scratch
- Set up MCP servers and understand the security model
- Know when to reach for each tool type
AI-First Project Setup
Kick off a project using AI as a design partner — docs, specs, design system, and a parallel build plan before one line of app code.
- Use AI as a design partner for docs and specs
- Establish a design system up front
- Draft a parallel build plan before writing app code
The Business Ecosystem
The same taxonomy applied to the product side — a second orchestrator for business work, and the bridge agent that connects product to engineering.
- Apply the specialist taxonomy to product work
- Stand up a second orchestrator for the business side
- Build the bridge agent that connects product to engineering
Two mirrored pipelines, one on each side of the business, joined by a single translating bridge agent in the center.
Instructions like code · specialists over generalists
Steal this setup
Start at Lesson 01 and build the whole environment — or clone the materials and skip straight to the parts you need.