How I'd build Tradewind
I'd reach for Next.js with TypeScript on the frontend, Next.js API routes for the generation service, Postgres for multi-tenant data, Stripe for billing, and the Claude API for copy generation. The Shopify, Amazon, and Etsy integrations would use their official SDKs to read catalogs and stage listings. I'd estimate 110-130 hours to ship something sellers can use and pay for.
Day-by-day plan
- Days 1-2: Provision Postgres schema for users, teams, API keys, and listing records. Set up NextAuth for OAuth with email fallback. Multi-tenant row-level security policies.
- Days 3-4: Wire Stripe billing with three tiers (free trial, starter $29/month, pro $79/month). Metered billing for listings over the plan limit. Webhook handlers for subscription status changes.
- Days 5-6: Integrate Shopify API. Build the onboarding flow to connect a Shopify store, read product catalog, and display a preview of 50 products in the app.
- Days 7-8: Add Amazon Seller Central and Etsy API integrations using the same preview pattern. Handle auth token refresh and rate limiting across all three.
- Days 9-10: Build the generation interface. Bulk-select products, choose tone (professional, casual, marketing-focused), category override. Real-time progress bar and downloadable CSV of results.
- Days 11-12: Integrate the Claude API with a system prompt tuned for SEO keywords and platform-specific formatting. Add a draft/publish workflow and conflict detection to prevent overwriting recent manual edits.
- Days 13-14: End-to-end testing, error recovery for failed generations, demo video, and Shopify App Store listing draft.
What's hard about this build
The core risk is API surface area. Shopify, Amazon, and Etsy all have different rate limits, authentication flows, and data models. Amazon's SP-API is notoriously fragile during quota windows. You also have to detect when a user has manually edited a listing after you generated one, otherwise you'll clobber their work. The copy quality is only as good as the prompt, and sellers have wildly different catalog quality - thin products with no images, misspelled descriptions, incomplete specs. Claude will hallucinate keywords that don't exist in the platform's search index. You need continuous feedback loops and maybe a human-in-the-loop for the first batch. Finally, platform algorithm changes are silent. Amazon could deprecate the field your whole value prop rides on, and you'd only find out when sellers complain.
What's fast because of AI
Claude compresses three weeks of work into three days. I use it to generate the full Stripe webhook test suite, catching edge cases like proration logic that would take days to manually enumerate. It scaffolds the NextAuth flow and Postgres schemas with the right indexes pre-baked. For the Claude API integration, I iterate the system prompt in conversation, testing edge cases like "product name is just emojis" or "category is blank", and Claude flags the gaps immediately. UI boilerplate for forms, tables, and progress states comes from a few lines of guidance. And copywriting the product UI itself - microcopy for empty states, error messages, onboarding hints - Claude writes first drafts that I refine in 15 minutes instead of an hour each.
How I'd hand it off
I'd record a Loom walkthrough of the full onboarding flow from store connection through a batch generation and CSV export. Leave a runbook in Linear covering deployment commands, credential rotation, and how to manually trigger a generation if the async queue breaks. I'd do 30 days of pager duty included in the estimate, monitoring error rates and Stripe webhook failures. All Stripe, Shopify, Amazon, and Etsy API credentials go into a 1Password vault with shared access. The Linear board has open issues flagged as blockers for your next hire or post-launch priority.