# Modern AI Software Engineering: The Orchestrator's Playbook
> A practical, opinionated course for engineers who have stopped *writing* code
> alone and started *directing* the systems that write it with them.
---
## What this course is
Most "AI for developers" material teaches you to talk to a chat box. This course
teaches you to **run a workshop**: design specialized agents, hand them
contracts they can verify against, keep their memory coherent across weeks of
work, and assemble the whole thing into a pipeline you can trust enough to
merge from.
It is built around the practitioner's reality of late 2025 / early 2026:
- Agentic CLI tools (Claude Code, OpenCode) are the new IDE.
- Specs are the new source of truth — not the prose kind, the *executable*
kind.
- Reasoning budgets, model selection, and context engineering are
load-bearing skills, not optimizations.
- The hard part isn't getting the model to write code. The hard part is
knowing whether the code it wrote is the code you actually wanted.
## Who this is for
You'll get the most out of this course if you can already:
- Read and ship production code in at least one language.
- Comfortable with the shell, `git`, and CI fundamentals.
- Have used an AI coding assistant for more than a weekend and felt its
ceiling.
You do **not** need to be an ML engineer. We treat LLMs as **components**, not
as research objects.
## How the course is organized
Seven modules, designed to be read in order, each ending with a hands-on lab
and a set of pitfalls collected from real practice.
| # | Module | Labs | What you'll leave with |
|----|--------------------------------------------------------------------------|----------------------------------------------------------------------------|----------------------------------------------------------------------------------------|
| 1 | [Foundations & The "Thinking" Economy](modules/01-foundations.md) | [4 labs](labs/01-foundations-labs.md) | A mental model of LLMs as priced reasoning engines; rules for picking the right model. |
| 2 | [Advanced Spec-Driven Development](modules/02-spec-driven-development.md)| [5 labs](labs/02-spec-driven-development-labs.md) | The ability to write specs an agent can implement *and verify against*. |
| 3 | [Mastery of Agentic CLI Tools](modules/03-agentic-cli-tools.md) | [4 labs](labs/03-agentic-cli-tools-labs.md) | A configured, governed Claude Code / OpenCode setup with house rules. |
| 4 | [Extending AI Intelligence](modules/04-extending-intelligence.md) | [4 labs](labs/04-extending-intelligence-labs.md) | Custom skills, MCP servers, and hooks that turn the CLI into your platform. |
| 5 | [Memory Orchestration & Context Engineering](modules/05-memory-orchestration.md) | [4 labs](labs/05-memory-orchestration-labs.md) | Cross-session memory that survives compaction and onboards new agents in seconds. |
| 6 | [The Meta-Agent Factory & Verification Frontier](modules/06-meta-agent-factory.md) | [5 labs](labs/06-meta-agent-factory-labs.md) | A roster of specialized agents that review each other before code reaches you. |
| 7 | [The Human-in-the-Loop & Strategic ROI](modules/07-human-in-the-loop.md) | [4 labs](labs/07-human-in-the-loop-labs.md) | A defensible answer to "what is *your* job now?" and a number to put next to it. |
Each module also closes with one introductory lab inline; the
expanded lab files in [`labs/`](labs/README.md) take each concept
further with 3–5 hands-on exercises per module.
## How to use this material
- **Read sequentially the first time.** Concepts from later modules
(verification, memory) only make sense after the foundations.
- **Do the labs.** Every module ends with one. They are short, real, and
designed to be done inside an actual agentic CLI session.
- **Treat the examples as starting points, not artifacts.** Configurations
rot. The reasoning behind them is what you should internalize.
## A note on tooling
Examples lean on **Claude Code** as the reference agentic CLI because it is
currently the most feature-complete platform for the patterns we teach
(skills, MCP, hooks, sub-agents, plan mode). Where a pattern is portable, the
text calls it out and shows the equivalent in **OpenCode** or another
provider. The principles outlive any single tool.
---
*Last revised: 2026-05.*