Now accepting guest editorial from SAP, Oracle, Workday and ServiceNow practitioners.Pitch a guide
The Sign-Off Table

Explainer

AI-DLC: the AI-Driven Development Lifecycle, in diagrams

What AWS’s AI-DLC is, how it works, where quality fits, and what it means for SAP, Oracle, Workday and ServiceNow release teams.

Where it comes from

AI-DLC was introduced by AWS in July 2025 on the AWS DevOps Blog, by AWS principal solutions architect Raja SP, and set out in an AWS method definition paper. AWS open-sourced its AI-DLC workflows in November 2025 and presented the method at re:Invent 2025. Its premise is that bolting AI onto existing methods (“AI-assisted” development) leaves most of the benefit on the table, and that handing AI the whole job unsupervised is not safe. AI-DLC sits between the two: AI drives, people decide.

1. The core loop: AI plans, people approve

AI-DLC reverses the usual conversation. For each activity, AI creates a plan, asks clarifying questions and implements only after a person validates it. The method compares these human checks to a loss function: they catch errors early, before they spread into later work.

  1. AI

    Creates a plan

  2. AI

    Asks clarifying questions

  3. Human

    Validates and corrects

  4. AI

    Implements the approved plan

Repeated for every lifecycle activity, in minutes or hours rather than weeks

The AI-DLC interaction pattern. Source: AWS DevOps Blog, July 2025.

2. Three phases, each with a ritual

Inception decides what to build and why, in a facilitated, whole-team session AWS calls Mob Elaboration. Construction decides how, in Mob Construction. Operations deploys and runs it, with AI watching telemetry and proposing actions for people to approve.

Phase 1

Inception

What are we building, and why?

Mob Elaboration

  • · User stories
  • · Non-functional requirements
  • · Risks
  • · Measurement criteria
  • · Units and suggested Bolts

Quality

QA is in the room. Risks and measurement criteria are defined before any code.

Phase 2

Construction

How do we build it?

Mob Construction

  • · Domain design
  • · Logical design
  • · Code and unit tests
  • · Deployment units

Quality

AI generates functional, security and performance tests; people review the scenarios; AI runs them and proposes fixes.

Phase 3

Operations

Is it running well?

AI-assisted operations

  • · Deployment
  • · Observability
  • · Runbook actions

Quality

AI reads telemetry and proposes runbook actions; people approve before anything runs.

Each phase hands richer context to the next. Artifacts are saved as shared context and stay traceable both ways.

AI-DLC’s three phases with their outputs and where quality work sits. Source: AWS AI-DLC method definition.

A note on versions: AWS’s open-source workflow repository has since grown to a more detailed model with additional phases, numbered stages and verification gates between phases. It changes often, so the diagrams here follow the original three-phase method that the newer versions build on.

3. Intents, Units and Bolts

Work starts as an Intent, a statement of purpose. Inception breaks it into Units, cohesive pieces of work comparable to epics. Units are built in Bolts, the method’s smallest iteration, measured in hours or days rather than the weeks of a sprint.

Intent

A high-level statement of purpose

e.g. “Move order-to-cash to S/4HANA without losing a day of shipping”

Unit

credit management

Unit

billing output

Cohesive, self-contained pieces of work, comparable to an epic

Bolt 1

hours–days

Bolt 2

hours–days

Bolt 3

hours–days

Bolt 4

hours–days

The smallest iteration. Bolts replace sprints and can run in parallel or in sequence.

From Intent to Units to Bolts. The ERP examples are illustrative.

4. Inside Construction

Each Unit moves from a technology-independent domain design to a logical design that adds non-functional requirements, then to code with unit tests, then to deployment units. AI generates functional, security and performance tests along the way; people review the scenarios; AI runs them, traces failures back to the code or configuration that caused them and proposes fixes for approval.

  1. Domain design

    Business logic, independent of technology

  2. Logical design

    Non-functional requirements, patterns and decision records

  3. Code + unit tests

    AI generates, runs and analyzes the tests

  4. Deployment units

    Tested for function, security, NFRs and operational risk

The Construction flow. Source: AWS AI-DLC method definition.

AI-DLC compared with a typical agile lifecycle

Typical agile SDLCAI-DLC
Who drivesPeople direct each task; AI assistsAI proposes plans and asks questions; people validate
Unit of iterationSprints of one to four weeksBolts of hours or days
When QA joinsOften after development, at system or integration testFrom Inception, in Mob Elaboration
DesignOutside the method (Scrum and Kanban leave it to teams)Built in, through domain and logical design
TestsWritten and maintained by peopleGenerated by AI, reviewed by people, run and analyzed by AI

5. An ERP lens

AWS wrote AI-DLC for building software. ERP programs configure and extend packaged platforms, organize work around business processes and live with vendor release cycles. Here is how the quality work of an SAP S/4HANA program can map onto the AI-DLC phases. This mapping is ours, not AWS’s.

Inception

  • Business process scope from the process library (O2C, P2P, R2R)
  • Simplification items and custom code impact from the Readiness Check
  • Risks mapped to processes, interfaces and roles
  • Measurement criteria written as test exit criteria

Construction

  • Configuration, extensions and custom code built as Units
  • Automated end-to-end process tests generated and reviewed per Bolt
  • Test data created or reserved by each test
  • Integration and role tests before a Unit is "done"

Operations

  • Risk-based regression on every transport and release
  • Cutover rehearsals and hypercare as operational runbooks
  • Production incidents fed back to the process library and tests

The Sign-Off Table’s adaptation for ERP programs. Not part of AWS’s published method.

The Sign-Off Table’s ERP adaptation of AI-DLC.

Read the full argument in AI-DLC for ERP teams, or see how it builds on a Testing Center of Excellence.

Put it to work

Free AI-SDLC templates

AI use policy, agent guardrails register, AI risk register mapped to OWASP and NIST, an Inception workshop pack, an Intent → Unit → Bolt planner, a review checklist for AI-generated tests and a metrics tracker.

Browse the AI-driven SDLC templates

Sources

AI-DLC is an AWS methodology. The Sign-Off Table is not affiliated with AWS.