Back to projects

A

D

L

C

the duck dose
with Claude
AI × Human Workflow

How humans and Claude ship software, together.

The AI Development Life Cycle (ADLC) is an evolution of the classic SDLC where AI acts as a first-class collaborator at every stage — discovery, design, build, review, test, and ship.

7
Stages
4
Human roles
1
AI partner

Who is in the loop

// roles_legend
Product ManagerOwns the why and the what.
Tech LeadOwns architecture and quality bar.
Software EngineeringBuilds, ships, and operates the code.
QAGuards quality and user trust.
Claude AIPair-partner across the entire lifecycle.
Workflow

The ADLC, top to bottom

Seven stages of building software, each split between what the team owns and what Claude assists with.

Discovery & Ideation

Human
Product ManagerTech Lead
  • PM gathers user pain points and defines the problem statement.
  • Tech Lead validates feasibility and surfaces tech constraints.
Claude AI
Claude AI
  • Summarizes user interviews and clusters feedback themes.
  • Generates competitor scans and opportunity matrices.
// stage_01 :: discovery

Requirements & Specs

Human
Product ManagerQA
  • PM writes user stories and acceptance criteria.
  • QA reviews testability and edge cases.
Claude AI
Claude AI
  • Drafts user stories from PM notes and expands edge cases.
  • Generates Gherkin scenarios and clarifying questions.
// stage_02 :: requirements

Architecture & Design

Human
Tech LeadSoftware Engineering
  • Tech Lead chooses patterns, data models, and integration boundaries.
  • SE reviews and refines the component design.
Claude AI
Claude AI
  • Drafts sequence diagrams and proposes data schemas.
  • Surfaces existing-codebase patterns to follow.
// stage_03 :: design

Implementation

Human
Software Engineering
  • SE writes and refines code; owns design judgment calls and merges.
Claude AI
Claude AI
  • Generates boilerplate and scaffolds components.
  • Explains unfamiliar code and proposes diffs as a pair-programmer.
// stage_04 :: build

Code Review

Human
Tech LeadSoftware Engineering
  • Engineers review architectural intent, naming, and business correctness.
Claude AI
Claude AI
  • Auto-summarizes PR diffs and flags risky changes.
  • Checks the PR against conventions defined in CLAUDE.md.
// stage_05 :: review

Testing & QA

Human
QASoftware Engineering
  • QA executes exploratory testing and owns release sign-off.
  • SE fixes defects and adds regression coverage.
Claude AI
Claude AI
  • Generates unit and integration test cases from acceptance criteria.
  • Drafts regression matrices and bug-reproduction steps.
// stage_06 :: qa

Deployment & Monitoring

Human
Tech LeadSoftware EngineeringQA
  • Tech Lead approves the release; SE runs the deploy pipeline.
  • QA monitors smoke tests and key signals in production.
Claude AI
Claude AI
  • Generates release notes and summarizes log spikes and alert anomalies.
  • Suggests rollback steps and feeds insights back into Discovery.
// stage_07 :: ship
Migration Path

From SDLC + Scrum to ADLC

How a Scrum team evolves into an AI-augmented delivery org. Five phases, each with its own outcome and concrete moves.

  1. Phase 01
    Baseline — SDLC with Scrum

    Outcome — Stable Scrum delivery cadence with no AI involvement.

    • Ceremonies: Sprint Planning, Daily Standup, Sprint Review, Retrospective, Backlog Refinement.
    • Roles: PM (as Product Owner), Tech Lead, Software Engineering, QA.
    • Artifacts: backlog, sprint board, burndown, velocity.
    • Pain points: manual story writing, slow code review, repetitive test design.

  2. Phase 02
    Awareness & Pilot

    Outcome — Team is trained on Claude AI and one ceremony is piloted.

    • Define usage guidelines and data / security boundaries.
    • Pick a low-risk pilot — e.g. AI-assisted backlog refinement.
    • Establish a shared prompt library the team can iterate on.
    • Measure pilot impact in the very next retrospective.

  3. Phase 03
    Ceremony Augmentation

    Outcome — Claude AI participates in core Scrum ceremonies.

    • Backlog refinement: Claude drafts stories and acceptance criteria from PM notes.
    • Sprint planning: Claude estimates complexity and surfaces dependencies.
    • Daily standup: Claude summarizes overnight PR and CI activity.
    • Sprint review: Claude drafts demo scripts and release highlights.

  4. Phase 04
    Lifecycle Integration (ADLC)

    This is ADLC

    Outcome — Claude AI is present across every SDLC stage end-to-end.

    • All 7 ADLC stages active — see the diagram above.
    • PR review uses a Claude-first triage pass before human review.
    • Test cases are auto-generated from acceptance criteria.
    • Release notes and runbooks are drafted by Claude; QA shifts toward exploratory testing.

  5. Phase 05
    Continuous Optimization

    Outcome — A closed feedback loop with continuously tuned AI usage.

    • Tracks DORA metrics alongside AI-effectiveness indicators (cycle time, review turnaround, defect-escape rate).
    • Retrospectives include a recurring "AI working agreements" review.
    • Prompt library and CLAUDE.md are continuously refined as the codebase evolves.
    • Insights flow back into Discovery, mirroring the ADLC feedback cycle.

© 2026 the duck dose · ADLC is an internal workflow concept.

built with Next.js · Tailwind · Framer Motion · and Claude