6
Stages
Define the lifecycle from problem intent to production operation and continuous improvement.
Framework · Agentic AI · DataKnobs
Agentic Development Lifecycle
Build software where AI agents execute and humans steer. ADLC gives teams a practical operating model for agentic software delivery: role ownership, gates, task delegation, and calibrated autonomy.
6Stages
5Stage Gates
5Autonomy Levels
RICE-ADelegation Test
Complete ADLC guide by Prashant Dhingra
All modules
Six pages, one operating framework
Each module expands one part of the ADLC so teams can safely introduce AI agents into software delivery without losing human accountability.
Slide 01
Overview
ADLC Agenda & Overview
The complete map of the framework: six stages, five gates, five autonomy levels, and the end-to-end ADLC operating model.
Read module →Slide 02
Co-development
Agent & Human Co-Development
Agents own reversible, verifiable execution. Humans own irreversible judgement, risk acceptance, architecture, and stage approval.
Read module →Slide 03
Lifecycle
ADLC Six Stages
Intent & Framing, Specification, Planning, Execution, Verification, and Operation each with explicit agent and human roles.
Read module →Slide 04
Delegation
RICE-A Test
A task passes to the agent only if it is reversible, inspectable, constrained, low error cost, and autonomy-appropriate.
Read module →Slide 05
Autonomy
5 Autonomy Levels
From Suggest to Autonomous, ADLC defines how much execution can be delegated and when human review must remain in control.
Read module →Slide 06
Governance
5 Stage Gates
Intent Lock, Spec Approval, Plan Authorization, Verification Sign Off, and Release Authorization anchor human accountability.
Read module →Framework pillars
The ADLC is built around four control points
The framework is designed to increase speed without removing human judgement from decisions where accountability, risk, and architecture matter.
5
Gates
Force evidence before progress. Gates prevent agent errors from cascading downstream.
5
Levels
Calibrate how much autonomy the agent earns for each task type and risk profile.
A
RICE-A
Decide what the agent can safely own using reversibility, inspectability, constraints, error cost, and autonomy.
From intent to operation
A practical lifecycle for agentic delivery
ADLC is not just an AI coding workflow. It is a controlled lifecycle where requirements, execution, verification, and production readiness are separated into clear stages.
A
Intent & FramingDefine the goal, scope, and success criteria.
B
SpecificationConvert intent into testable acceptance criteria.
C
PlanningCreate a task graph, dependencies, guardrails, and autonomy levels.
D
ExecutionLet the agent build, test, self-correct, and produce deliverables.
E
VerificationReview evidence, test results, risk, and conformance.
F
OperationDeploy, monitor, learn, and feed production signals back into the lifecycle.
Governance model
Humans steer at the moments that matter most
ADLC moves human attention from low-value micro-review to high-value gatekeeping, risk acceptance, and release authorization.
Agents execute
Agents draft, generate, test, refactor, document, monitor, and propose actions within explicit task boundaries.
Humans steer
Humans define intent, approve architecture, accept risk, assign autonomy, review gates, and own outcomes.
Gates control progress
No stage advances until its output is proven, reviewed, and signed off by the right human gatekeeper.
Core thesis
Autonomy is earned, not assumed. Agents gain scope only when evidence, reversibility, and controls justify it.
The goal is not universal autonomy. The goal is the right level of autonomy for the right task at the right gate.
Agent Execute. Human Steer.
ADLC allows teams to capture agent speed while keeping accountability, judgement, governance, and risk decisions with humans.
Continue learning
Start with the ADLC overview
Read the agenda first, then continue through co-development, stages, RICE-A, autonomy levels, and stage gates.