An end-to-end autonomous SDLC pipeline that takes any IT requirement — from a one-line enhancement request to a full PRD — and delivers production-deployed code. Replaces a 16-step manual process with a single human trigger.
As a Product Owner managing 3 scrum teams at Charter Communications, I lived inside this 16-step process daily — writing requirements, creating JIRA tickets, briefing engineers, reviewing PRs, coordinating deployments. Each step had handoff latency, context loss, and meeting overhead. I built this pipeline to collapse those 16 steps into 2 human touchpoints, using AI to handle everything in between. The goal wasn't to replace the team — it was to eliminate the coordination tax so the team could focus on decisions that actually require human judgment.
The pipeline is not limited to PRDs. Any structured requirement can be the entry point:
| Input Type | Example | Output |
|---|---|---|
| Product Requirements Doc (PRD) | New feature spec with user stories, personas, KPIs | Epic → multiple Stories → full feature branch |
| IT System Enhancement | "Add ACH AutoPay support to the checkout flow" | Single Story → targeted code change → deploy |
| Bug Report | "Order confirmation email not sending for smartwatch orders" | Bug ticket → fix → test → patch deploy |
| POC / Spike | "Evaluate feasibility of fraud score integration at order entry" | Spike ticket → prototype code → findings doc |
| Technical Debt | "Refactor eligibility logic into reusable service layer" | Tech debt story → refactor → regression tests |
| Compliance / Policy Change | "Flag NY State vulnerable customers in order flow" | Requirement → targeted implementation → audit trail |
The Claude AI decomposition layer adapts ticket scope, story count, and acceptance criteria depth based on the size and type of the input — a one-liner enhancement becomes a single Story; a full PRD becomes an Epic with 5–10 Stories.
Requirement Input (any format)
│
▼
Claude AI — interprets requirement type + scope
│ decomposes into appropriately-sized
│ JIRA tickets with EARS acceptance criteria
│
▼
Atlassian Rovo MCP — creates structured JIRA tickets
│ (Epic → Stories, or standalone Story/Bug/Spike)
│
▼
WebStorm IDE — JIRA integration pulls ticket automatically
│ Claude AI coding agent reads the spec
│ implements the code (spec-driven development)
│
▼
GitHub — code committed to feature branch
│
▼
GitHub Actions — CI/CD pipeline (test → build → push)
│
▼
Render — auto-deploy to production
Only two checkpoints in the entire pipeline:
- Providing the requirement (any format)
- Reviewing JIRA tickets before code generation begins
Everything else — decomposition, spec writing, ticket creation, coding, CI/CD, and deployment — runs autonomously.
The JIRA ticket IS the spec.
All tickets are written in EARS format (Easy Approach to Requirements Syntax) with precise, testable acceptance criteria. When WebStorm's JIRA integration surfaces the ticket to the Claude AI coding agent, Claude has everything it needs to implement — no Slack threads, no handoff calls, no ambiguity.
The same pipeline works across scales because the decomposition layer adjusts to the input:
- A two-sentence enhancement → one well-scoped Story
- A multi-page PRD → a full Epic hierarchy with prioritized Stories
| Layer | Tool |
|---|---|
| Requirement Decomposition | Claude AI |
| JIRA Ticket Creation | Atlassian Rovo MCP |
| IDE + Coding Agent | WebStorm + Claude AI |
| Version Control | GitHub |
| CI/CD | GitHub Actions |
| Deployment | Render |
- MCP Chatbot — a live project built end-to-end using this pipeline
- Weather Feature SCRUM Specs — example EARS-format ticket specs that feed the coding agent