AWS ships a rulebook to keep AI coding agents from going rogue

What it does

AI-DLC is a set of markdown rule files you download and drop into your project’s configuration directory. The rules define a three-phase adaptive workflow—inception, construction, and operations—that your AI coding agent is supposed to follow. It supports Kiro, Amazon Q, Cursor, Cline, Claude Code, and GitHub Copilot, each with their own directory convention (.kiro/steering/, .cursor/rules/, .clinerules/, etc.).

The interesting bit

The whole thing is essentially a governance layer expressed as structured text. Rather than trusting an AI to improvise a development process, you hand it a literal rulebook—core-workflow.md plus conditionally-loaded detail files for specific phases. It’s the bureaucratic paperwork approach to prompt engineering, and AWS seems to think this is more reliable than hoping your agent stays on track.

Key highlights

• Ships as a release zip with two folders: aws-aidlc-rules (core workflow) and aws-aidlc-rule-details (phase-specific deep-dives)
• Supports six major AI coding platforms with platform-specific setup paths
• Designed to keep the human “in control of the process” per AWS’s own framing
• Includes an “experimental” AI-assisted setup option, which is either convenient or ironic depending on your cynicism level
• Methodology is backed by an AWS blog post and a separate “Method Definition Paper” hosted on what the README admits is a temporary Amplify URL

Caveats

• The README contains a TODO about replacing the Amplify URL with something permanent, so the methodology paper link may rot
• Setup is manual copy-paste into dot-directories; no package manager or automated distribution
• Windows users get special warnings about File Explorer’s “Extract All…” creating wrapper folders, suggesting this has already confused real users

Verdict

Worth a look if you’re already using AI coding agents and want to impose some process discipline on them. Skip it if you were hoping for a tool or framework—this is documentation-as-infrastructure, and you’ll need to buy into the methodology for it to matter.