Claude Academy: AI-Native SDLC Playbook Course Guide

TL;DRAnthropic's Claude Academy offers a free 14-lesson course that takes AI-assisted coding from 'individuals using Claude Code' to 'an organization-wide development lifecycle.' Four core concepts — intent.md, CLAUDE.md, Skills, and Hooks — wire together into a complete AI-native SDLC.

#claude-code #sdlc #claude-academy #agent-skills #hooks #ci-cd #governance #context-engineering

Table of Contents

Course Structure at a Glance
Four Core Concepts
Who Should Take This Course
Related Articles on This Site
Overall Takeaway
References

Series: AI-Native SDLC Playbook (0 / 15)

AI-Native SDLC Playbook L1: When Code Generation Is No Longer the Bottleneck

Most teams use Claude Code as a personal productivity boost — each engineer chats with Claude in their terminal, and output quality depends on individual prompt skills. That improves personal throughput, but the organization's development process stays unchanged.

Anthropic's The AI-Native SDLC Playbook on Claude Academy addresses exactly this gap. Across 14 lessons and 6 stages — from requirements capture to post-deployment monitoring — it maps out how to embed AI agents into every phase of the SDLC. This article is a reading guide to help you build a mental model before taking the course and identify which chapters matter most for your role.

Course Structure at a Glance

The 14 lessons follow the traditional SDLC stages but redefine who does what at each step:

Stage	Lesson	Core Question
Intro	1. Introduction	Where do bottlenecks form when AI accelerates code generation?
Plan	2. Capture as intent.md	How do vague conversations become machine-actionable documents?
Design	3. Requirements and design	Why can spec and design collapse into a single session?
Build	4. Plan mode as default	Why should Claude Code sessions start in plan mode, not coding?
	5. The CLAUDE.md	How do you encode org conventions, architecture, and common mistakes into repo-level context?
	6. Skills as institutional knowledge	How do you package review standards and deploy procedures as version-controlled skills?
	7. Parallel sessions and subagents	How do you coordinate multiple Claude sessions working simultaneously?
Test	8. Give Claude a feedback loop	How does Claude self-verify before submitting work?
	9. Continuous evals in CI	How do you regression-test agent output inside a CI pipeline?
Deploy	10. AI in the PR review loop	How do you layer PR review — what goes to the agent, what stays with humans?
	11. Hooks as approval gates	How do you use deterministic gates to prevent irreversible agent actions?
	12. CI/CD integration	How do you wire all of the above into an automated deploy pipeline?
Maintain	13. Closing the loop on metrics	How does post-deployment monitoring feed back into new intent.md files?
Closing	14. Closing thoughts	Resource roundup and next steps

Four Core Concepts

The entire course rests on four artifacts. Understanding them is understanding the course's design philosophy.

intent.md — Requirements in Machine-Readable Form

In traditional development, requirements are scattered across Jira tickets, Slack threads, and meeting notes. intent.md consolidates them into a single version-controlled file structured as "problem + expected outcome" — readable by humans, directly actionable by Claude.

According to the course, intent.md is not a spec. It's a statement of "what we want to solve." The spec is generated by Claude during the Design stage based on the intent, then reviewed by humans before entering Build. This separation matters — it lets non-technical stakeholders drive requirements without needing to write technical specifications, while engineering still gets structured input.

CLAUDE.md — Repo-Level Agent Context

CLAUDE.md lives at the repo root and tells Claude "how this project works." It covers:

Code style and lint rules
Architecture decisions (why this framework, which patterns are intentional)
Common mistakes ("don't use X because Y")
Commit format and branching strategy

The course's emphasis isn't on how to write a CLAUDE.md — that's already in the Claude Code docs — but on why it's organizational infrastructure. When every repo has a well-maintained CLAUDE.md, onboarding time drops and agent-generated code style stops varying from person to person.

Skills — Organizational Standards as Reusable Skills

If CLAUDE.md is "static context," then a Skill is "a triggerable standard operating procedure." A skill is a folder containing a SKILL.md that describes trigger conditions and execution steps, plus any templates or reference files needed.

The course positions skills as institutional knowledge — the tacit knowledge that only veterans carry: "which checks to run before opening a PR," "the full steps to deploy to staging," "the breaking-change history of this API." Packaging those as skills lets the agent execute them consistently, independent of any individual engineer's memory.

Hooks — Deterministic Approval Gates

Hooks are the governance backbone of the entire course. Unlike CLAUDE.md and Skills (which are advisory — Claude should follow them), Hooks are deterministic — they fire shell commands before and after agent actions, blocking if the check fails.

Typical hook scenarios:

PreToolUse: intercept before Claude runs git push to verify the target branch
PostToolUse: run lint after a file write, requiring fixes if it fails
Stop: print a diff summary before Claude finishes, ensuring nothing was missed

As the course puts it: "Humans remain accountable for every decision that requires judgment." Hooks are the mechanism — automate the checks that don't require judgment, and surface the ones that do.

Who Should Take This Course

This course is not a Claude Code beginner tutorial. It assumes you can already use Claude Code for personal development; the question is "how do you scale it to a team."

The roles that benefit most directly:

Role	Key Lessons
Engineering Manager / Tech Lead	L1–L3 (strategy), L10–L11 (governance)
Platform Engineer	L5–L6 (CLAUDE.md / Skills), L8–L9 (CI evals), L11–L12 (Hooks / CI/CD)
Security / Compliance	L11 (Hooks as approval gates), L13 (metrics feedback)
Individual Developer (looking to level up)	L4 (plan mode), L7 (parallel sessions), L8 (feedback loop)

If your team is still at the "everyone uses Claude Code on their own" stage, start with L5 (CLAUDE.md) and L11 (Hooks) — these two have the highest ROI. A well-maintained CLAUDE.md plus basic hooks can unify output quality without waiting for everyone to finish all 14 lessons.

Many concepts from this course have deeper standalone write-ups on quidproquo:

SDLC overview: Integrating AI Agents into the Development Workflow: A Five-Stage SDLC Approach — a broader industry perspective on Agentic AI across SDLC stages
Context Engineering: Harness Engineering: An Engineering Methodology for Reliable AI Agent Delivery — deep dive into CLAUDE.md, hooks, and skills in practice
Claude Code intro: Claude Code Startup Guide — if you haven't started using Claude Code yet, read this first

Overall Takeaway

The core argument of The AI-Native SDLC Playbook is simple: when code generation is no longer the bottleneck, the bottleneck shifts to everything around code generation — requirements capture, design review, testing strategy, deployment governance, post-launch feedback. The course doesn't introduce new tools; it provides a framework for wiring existing tools (Claude Code, CLAUDE.md, Skills, Hooks) into a complete development lifecycle.

The 14 lessons follow a loop: intent.md (requirements) → design → build → test → deploy → monitor → new intent.md. The final lesson (Closing the loop on metrics) explains how monitoring systems automatically produce new requirements documents, forming a continuous improvement cycle.

Free, 14 lessons, each works standalone. If you only have 30 minutes, watch L2 (intent.md) and L11 (Hooks) — these two define the line between "AI-native" and "AI-assisted."