← Work

Multi-agent AI workflows in production, not a demo

Dealerware

Designed a multi-agent SDLC (Claude/Codex/Gemini, Challenger/Risk roles) that cut time to market from months to weeks.

Problem

Standard AI-assisted coding (single model, single pass) produced inconsistent quality and didn’t scale trust across a team — no structured way to check business intent against technical risk before code shipped.

Action

Designed and ran a multi-agent development workflow across Claude, Codex, and Gemini with explicit roles: a context-extraction agent, parallel “Challenger” (business-intent) and “Risk” (technical-risk) agents, and a red-green-refactor cycle gated by human approval and acceptance-criteria evaluations.

Result

Time to market dropped from months to weeks on affected work, with fewer defects escaping to QA and more consistent documentation as a byproduct of the structured process. This work changed how the broader team operated around AI usage (details beyond this are NDA-scoped).