AI-DLC Workflows is an open-source methodology and rule set from AWS Labs that turns generic AI coding agents into disciplined software engineering collaborators by giving them adaptive, phase-based workflow steering rules. With nearly 2,000 GitHub stars and 329 forks in just six months, the project is rapidly becoming a reference implementation for how enterprise teams structure agent-driven development. The framework is intentionally agent-agnostic and IDE-agnostic, working with Claude Code, Amazon Q Developer, Kiro, Cursor, Cline, GitHub Copilot, OpenAI Codex, and custom agents. ## Why AI-DLC Matters Most AI coding agents today operate as one-shot generators: a developer prompts, the agent writes code, and quality control happens entirely in the developer's head. This scales poorly for production engineering work where requirements, design, testing, and operations all need explicit treatment. AI-DLC introduces a structured AI-Driven Development Life Cycle that segments work into three sequential phases - Inception, Construction, and Operations - and uses adaptive rules to decide which stages actually deserve effort for a given request. The result is a methodology that brings the rigor of a traditional SDLC to agentic coding without slowing down trivial changes. ## The Three Phases The Inception phase answers WHAT and WHY. The agent analyzes requirements, drafts user stories, sketches application design, and assesses risk and complexity. The Construction phase answers HOW. The agent performs detailed component design, generates code, configures builds, designs test strategies, and runs quality assurance. The Operations phase handles deployment automation, infrastructure setup, monitoring configuration, and production readiness validation. Each phase produces structured artifacts in an aidlc-docs/ directory that humans can review and version-control. ## Adaptive Intelligence and Risk-Based Execution A central design principle is that AI-DLC only executes stages that provide genuine value for the specific request. For a trivial bug fix, the agent may skip the full Inception ceremony. For a complex new feature touching multiple services, it executes the full pipeline with detailed risk analysis. The framework examines existing codebases and complexity signals to make this decision, allocating comprehensive treatment to high-risk changes while keeping simple changes efficient. ## Question-Driven Human Control Rather than relying on free-form chat, AI-DLC uses structured multiple-choice questions written to files. The agent proposes choices, the developer answers in the file, and execution proceeds based on those answers. Each phase requires explicit human review and approval of the execution plan before code is generated. This keeps humans firmly in the decision loop on critical choices while still letting the agent handle the heavy lifting. ## Extensibility and Multi-Agent Support AI-DLC ships as a set of platform-agnostic rule files in aidlc-rules/, including a core workflow file and detailed phase rules. Teams place these files in the appropriate directory for their agent of choice, for example .github/copilot-instructions.md for GitHub Copilot or the equivalent location for Claude Code, Cursor, or Cline. Organizations can layer custom rules on top for security policies, compliance requirements, or in-house architectural conventions, making it suitable for regulated enterprise environments. ## Limitations AI-DLC is a methodology layer, not a coding agent itself, so its quality depends entirely on the underlying agent and model. The structured question-answer flow adds friction for very small tasks where developers may prefer to skip ceremony. Documentation generated in aidlc-docs/ requires team discipline to keep up to date as code evolves. Adoption requires changing developer habits to start requests with the AI-DLC invocation, which can be a cultural shift for teams already comfortable with chat-style prompting.