Glonce

awslabs publishes AI-DLC, structured workflows for AI coding agents

toolsThe Glonce editorsHacker News🔥57Glonce rating57Impact45Novelty45Relevance95Surprise25Players60Credibility80
awslabs publishes AI-DLC, structured workflows for AI coding agents

AI-DLC (AI-Driven Development Life Cycle) is the subject of the awslabs/aidlc-workflows GitHub repository. Its README describes it as something that turns AI coding assistants into structured, verifiable software-delivery workflows. One harness-neutral core runs natively in seven named harnesses: Claude Code, Kiro CLI, Kiro IDE, Codex CLI, Cursor, opencode and GitHub Copilot.

The README's pitch is that ad-hoc AI coding loses context as projects grow. AI-DLC instead keeps requirements, decisions, implementation, tests and operational work connected through one audited lifecycle. The listed building blocks: 5 phases and 33 stages from initialization through operation; 14 agents (11 domain experts, 2 reviewers and an adaptive composer); 11 workflow profiles covering features, bug fixes, infrastructure, security, proofs of concept, enterprise delivery and other common work; human approval gates and source-bound review evidence; a 119-event audit trail plus persistent state, team knowledge and learned rules; and the same deterministic engine across every supported harness. Readers are pointed to a Workflow Profiles page to compare Classic, Express and the focused workflows, and to an AI-DLC Workflows 2.0 Specification for the architecture and methodology.

Installation is a one-line script. On macOS, Linux or WSL, a curl command fetches install.sh from the latest GitHub release and pipes it to sh; on Windows PowerShell, irm fetches install.ps1 and pipes it to iex. The installer sets up the native aidlc command and every harness runtime. Bun and Node.js are not needed for a native install, which runs its hooks and commands through aidlc. On Windows it installs for the current account and registers the bin directory in User PATH; running it from an elevated "Run as administrator" window triggers a warning and a prompt, because installing as administrator is less safe. The -NoModifyPath flag skips PATH changes. For people who prefer to manage project files themselves or cannot install a native executable, a manual-copy path exists: install Bun, download aidlc-copy-runtime-X.Y.Z.tar.gz from the release, and copy the complete runtime// directory into the project. That is the only route that needs Bun.

Setup per project: from the project root, run aidlc config --harness claude (or kiro, kiro-ide, codex, cursor, opencode, copilot), then aidlc doctor. Without --harness, aidlc config starts an interactive setup when a terminal is available. Then open the harness in the configured project and describe the work, for example: /aidlc Build a REST API for inventory management. Codex CLI uses $aidlc instead of /aidlc. In Kiro IDE, first choose aidlc in the chat panel's agent picker. AI-DLC selects a workflow from the request, asks for missing decisions, and stops at approval gates before moving forward.

Model setup belongs to the harness, and the shipped project configuration keeps the provider and model the user already selected. aidlc config providers can apply Amazon Bedrock settings on supported project surfaces or record manual setup for other harnesses. Kiro CLI and Kiro IDE need no provider answer because model access comes with Kiro. For Kiro's v3 engine on Kiro CLI 2.24.1 or later, the README says to use kiro-ide, which pins it; kiro is Kiro CLI (V2). The methodology itself is provider-independent. AI-DLC works best with capable reasoning models, and the current recommended model is Claude Opus 4.8.

The repository layout: core/ holds the hand-authored, harness-neutral methodology and engine, including core/tools/ with 93 aidlc-.ts engine and authoring tools; harness// holds thin harness-specific manifests and integrations; plugins// holds optional AIDLC plugins; scripts/ covers packaging, binary, installer and release tooling; tests/ holds smoke, unit, integration and end-to-end tests; docs/ holds user, harness-engineering and developer documentation. dist/ and dist-release/ are generated, ignored local outputs. Contributors are told to edit core/ or harness//, never generated dist output, and to build with bun install --frozen-lockfile and bun scripts/package.ts. The README ends with an Important note: generative AI can make mistakes, so review generated output and costs before acting on them.

Key facts

  • AI-DLC is a harness-neutral core that runs natively in Claude Code, Kiro CLI, Kiro IDE, Codex CLI, Cursor, opencode and GitHub Copilot.
  • The lifecycle spans 5 phases and 33 stages, with 14 agents (11 domain experts, 2 reviewers, an adaptive composer) and 11 workflow profiles.
  • It stops at human approval gates and keeps a 119-event audit trail plus persistent state, team knowledge and learned rules.
  • Install is a one-line curl or PowerShell script; a native install needs no Bun or Node.js, and projects are enabled with aidlc config --harness and aidlc doctor.
  • The README recommends Claude Opus 4.8 and says AI-DLC works best with capable reasoning models.

Why it matters

AI coding assistants tend to lose context as a project grows, and the README names that as the problem AI-DLC targets. Its answer is a single lifecycle that ties requirements, decisions, implementation, tests and operational work together, with approval gates and an audit trail. It is also harness-neutral: the same engine is meant to behave the same way across seven different coding tools, rather than being tied to one vendor's assistant.

Who it affects

Developers and teams already working in Claude Code, Kiro CLI, Kiro IDE, Codex CLI, Cursor, opencode or GitHub Copilot, who want a structured process on top of their assistant. The profiles named in the README (features, bug fixes, infrastructure, security, proofs of concept, enterprise delivery) suggest it is aimed at both small tasks and larger delivery work. Contributors can add runtimes through a Porting to a New Harness guide.

How to use it

Install with the curl script (macOS, Linux, WSL) or the irm script (Windows PowerShell) from the latest release. In the project root, run aidlc config --harness <claude|kiro|kiro-ide|codex|cursor|opencode|copilot>, then aidlc doctor. Open the harness and describe the work, for example /aidlc Build a REST API for inventory management; Codex CLI uses $aidlc, and Kiro IDE needs aidlc chosen in the agent picker first. Without a native install, the manual-copy path needs Bun and the aidlc-copy-runtime tarball from the release. The README gives no pricing or cost figures.

How solid is it

The material is the project's own README on GitHub, so every claim is the maintainers' description of their product. The README does not give a release date, a version number for AI-DLC itself, or maintainer names, and it offers no benchmarks or measured results showing that AI-DLC improves code quality or productivity. It does describe a substantial test setup (smoke, unit, integration and end-to-end tests) and a determinism check in the build.

Risks and caveats

The README itself warns that generative AI can make mistakes and tells users to review generated output and costs before acting on them. It says AI-DLC works best with capable reasoning models, so results on weaker models are a question for users to test. The install is a script piped straight into a shell, and the README notes that installing as administrator on Windows is less safe. Kiro IDE users opening an untrusted folder are warned that trusting it lets the folder's .kiro hooks run commands on their machine. Kiro users should also note the split between kiro (Kiro CLI V2) and kiro-ide, which pins the v3 engine.

“Ad-hoc AI coding loses context as projects grow.”

— AI-DLC README