Rolling out an AI-assisted SDLC on a large legacy monolith

An enterprise event management platform (registration, attendee management, event websites, venue sourcing, analytics) built on a large, long-lived monolith, delivered by a team of about 30 engineers.

Bring AI coding agents into daily work on a large monolith, from story creation to QA, without lowering quality or losing engineer accountability for each change.

• Introduced Claude Code, Cursor, and agentic frameworks (BMAD-Method, Superpowers, Ponytail) across story creation, implementation, code review, and QA.
• Connected AI to full project context and JIRA through Atlassian MCP to create and update tickets.
• Grounded AI-assisted PR review in the full EPIC and child-story context.
• Added trace-mcp (a framework-aware code graph) and RTK (CLI output compression) to improve agent accuracy and cut token cost on the legacy codebase.
• Piloting the workflow with 4 engineers, and building an AI SDLC proof of concept with Claude Code headless mode and scripted runners.

• Engineers now refine AI-drafted stories instead of writing them from scratch.
• AI handles the first pass of PR review, so reviewers focus on design and risk.
• New developers contribute to the monolith within about one week.