Updated November 2026

AI-Driven Development Life Cycle (AI-DLC): Methodology and Practice

Class Duration

14 hours of live training delivered over 2-3 days to accommodate your scheduling needs.

Student Prerequisites

Experience on a software delivery team as a developer, tech lead, architect, product owner, QA engineer, or engineering manager
Familiarity with Scrum or Kanban
Basic knowledge of Git and pull requests; developers drive the coding agents, and other roles need little coding
Some experience with an AI coding assistant is helpful

Target Audience

Delivery teams and the people around them: developers, tech leads, architects, product owners, QA engineers, and engineering managers. The course is for teams that already use AI coding agents and need a shared method with clear roles and human approval gates. Teams can pair it with Spec-Driven Agentic Coding with Kiro or Spec-Driven Development with AI Coding Agents.

Description

AI-DLC is a delivery method from AWS that puts AI at the center of the work. AI drafts plans, questions, requirements, designs, code, and tests, and the team approves each step. Short bolts replace sprints, units of work replace epics, and the team reviews AI proposals together in mob sessions. This course teaches the method and runs it with AI-DLC Workflows, the open-source implementation in awslabs/aidlc-workflows. Students work through each phase in Kiro, Claude Code, GitHub Copilot, Cursor, or Codex CLI, and plan a rollout for their own organization.

Learning Outcomes

Explain how AI-DLC differs from AI-assisted and AI-autonomous development.
Describe the AI-DLC phases and the artifacts each produces.
Organize work as intents, units of work, and bolts.
Facilitate Mob Elaboration and Mob Construction sessions.
Run approval gates and request changes precisely.
Choose a workflow profile, depth, and test strategy for a change.
Install and run AI-DLC Workflows with the aidlc CLI.
Run AI-DLC in Kiro, Claude Code, Copilot, Cursor, or Codex CLI.
Adopt AI-DLC on existing codebases with reverse engineering.
Turn gate corrections into durable rules with the learning loop.
Map AI-DLC onto Scrum or SAFe practices.
Define adoption metrics and plan an organization-wide rollout.

Training Materials

Comprehensive courseware is distributed online at the start of class. All students receive a downloadable MP4 recording of the training.

Software Requirements

Git, the aidlc CLI from AI-DLC Workflows, at least one supported coding agent with an active plan or model access (Kiro, Claude Code, GitHub Copilot, Cursor, or Codex CLI), and a GitHub account with a repository for the hands-on portions.

Training Topics

Why the Lifecycle Changes When AI Drafts the Work

AI bolted onto existing rituals
AI-assisted, AI-autonomous, and AI-driven development
AI plans, asks, then implements
Bolts instead of sprints
Units of work instead of epics
Persistent context in the repository

AI-DLC Principles and Core Concepts

AI proposes, humans approve
Approval gates at every stage
Intents, spaces, units, and bolts
Determinism in tools, judgment in humans
From rules-based workflows to the current engine
Native harness integrations
The Classic profile

Ideation and Inception: Mob Elaboration

Intent capture and framing
Feasibility, constraints, and scope
Team formation and rough mockups
Requirements analysis and user stories
Domain and contract design
Units generation and delivery planning
Starting from a PRD or vision
The unit dependency graph
The bolt plan
Running Mob Elaboration

Construction and Operation: Mob Construction

Functional design per unit
NFR requirements and design
Infrastructure design
Code generation after plan approval
Unit checkpoints
The walking skeleton
Build, test, and CI pipeline stages
Running Mob Construction
The Operation phase
Deployment stages

Artifacts, State, and the Audit Trail

The intent record under aidlc/spaces/
What each stage writes
Requirements, stories, and bolt plans
aidlc-state.md and resumable sessions
Parking and resuming work
Reviewer agents and findings
The audit/ trail
Sensors for traceability and coverage
Lint and type-check sensors

Adaptive Depth and Workflow Profiles

Workflow profiles
feature, mvp, and poc profiles
bugfix, refactor, and security-patch profiles
Minimal, Standard, and Comprehensive depth
Independent test strategy levels
The adaptive composer
Changing scope at a gate
Matching ceremony to risk

Roles and Responsibilities

Decisions each role owns
Domain agents and their human counterparts
Answer modes
Approve or Request Changes
When to take over
Integration leads and unit teams

Installing and Running AI-DLC Workflows

Installing the aidlc CLI
aidlc config --harness
aidlc doctor and aidlc update
Starting work with /aidlc
Checking progress with /aidlc --status
What the team commits
aidlc.settings.json
Optional MCP servers
Amazon Bedrock as a model provider
Weak models and skipped reviews

AI-DLC Across Coding Agents

Kiro IDE and Kiro CLI
Claude Code skills, agents, and hooks
GitHub Copilot CLI and VS Code
Cursor IDE and CLI
Codex CLI and $aidlc
opencode support
How strictly each harness enforces gates

Greenfield and Brownfield Adoption

Greenfield MVP and feature runs
The technical environment document
Reverse engineering existing code
Practices discovery
Monorepos and multi-repo intents
Spaces for several teams
Starting small with Express runs

Team Knowledge, Rules, and the Learning Loop

Team knowledge folders
DocumentKB for existing documents
Org, team, and project rules
Phase rules
The learning loop
Plugins and custom stages
Custom sensors

AI-DLC in the Organization

AI-DLC with Scrum and SAFe
Bolts, backlogs, and ceremonies
AI-DLC and spec-driven development
Gate wait time and rework
Request Changes rate and lead time
Audit trail and commit provenance
Guard policy and unattended runs
A phased rollout roadmap
Pilot teams and champions