Margenie
Agent-native brand ops
- Next.js
- Design system
- AI
Overview
Margenie is the ops brain I wished I had running Takeout Order — agent-native, approval-first, and design-system driven. It's a Next.js platform for Shopify brands juggling profit metrics, Meta ad diagnostics, creative refresh, and shipping across scattered tools.
Problem
AI assistants suggest changes but don't integrate with real workflows. Brands need recommendations they can inspect, changes they approve before they go live, and mechanics explained in plain language instead of black-box optimization.
Solution
Four pillars define the product: an Ops design system (OpsCard, checklists, mechanic pills), Supercomputer agent layer (playbooks, ranked fixes, streaming tool execution), Ad Ops with Meta mechanics (learning phase, overlap, breakdown insights), generative creative for ad refresh (fal.ai — Nano Banana 2 for image, Seedance 2 for video), and deep integrations with Shopify, Meta, and EasyPost.
- Human-in-the-loop: propose → explain mechanic → approve → execute
- Generative creative pipeline (fal.ai) for image and video refresh — same approval gate before publish
- declare_work_plan / checklist UX for visible agent progress
- Meta delivery diagnostics module with learning-stage sync
- Shared style guide and components across Ad Ops and chat surfaces
Stack
Next.js, React, TypeScript, Prisma, Meta Graph API, Shopify API — deployed on Vercel. I own architecture, design, and implementation; AI tools accelerate coding but every production decision and craft choice is mine.
Outcome
Production SaaS at margenie.co. Closest analog to Claude-powered workflows + design system code + MCP-style tools — steerable automation where humans stay accountable.