Danny McMillan and Shubhash Sharma are back with another Claude Sessions episode covering both the back end and front end of building your Amazon business infrastructure with AI.
Shubhash walks through exactly how to register for Amazon's Seller Partner API — your free, direct access to your own sales, inventory, pricing, and order data — no third-party subscriptions required. Danny then breaks down the 165-feature design system he built to eliminate AI slop from websites, landing pages, and app interfaces.
Part 1: Amazon SP-API Setup (Shubhash)
What SP-API is — Amazon giving you a key to your own data warehouse: live inventory, real-time orders, pricing, catalog data, and sales reports
5-step registration process — Register as developer, create an app, select permissions, self-authorize, and connect to Claude Code to build dashboards
Common rejection reasons — Usually a missed checkbox or vague answer about data usage. Keep answers focused on personal brand development and safe data storage
Advertising API is separate — Different credentials, different registration, different refresh token. You cannot reuse SP-API tokens for ads
What you can build once connected — Custom dashboards, forecasting engines, inventory alerts, automated reporting — all built by Claude Code without knowing Python
Danny's guardrails — Hire a $50 Upwork specialist to help with paperwork submission, keep them on retainer for when APIs go down (especially Q4, Black Friday, Prime Day)
Part 2: The 165-Feature Design System (Danny)
The AI slop problem — Default fonts (Roboto, Arial), purple-blue gradients, three-column card layouts, floating animated orbs, oversized border radius — all telltale signs of generic AI output
15 anti-patterns cataloged — The system actively fights against common AI design defaults
Four-phase pipeline — Decide, Design, Build, Refine — with 15 databases and components extracted from 11 repos
Gap analysis scoring — Rates output out of 60 points. Seller Sessions Live went from 33 to 50; Databrill went from 48 to 55
Psychology of design baked in — Hick's Law (limit choices to 5-7), Miller's Law (chunk information in groups), Jacob's Law — all running automatically in the background
"Pretty doesn't convert" is a cop-out — Apple, Ralph Lauren, Sony all prove that quality design builds trust. The real issue was budget — now AI removes that barrier
Design is about subtraction — Cut 69% of animations in one project. Overcooking destroys user experience
25 quality gate techniques — Color tokens, typography rules, contrast ratios, accessibility (100+ rules), spacing, and composition patterns
Claude Loom workflow — Record feedback via Cmd+Shift+L, Claude extracts screenshots and browser URLs, and the system pushes back if changes violate the design system
Key Takeaways:
SP-API is free and gives you direct access to your Amazon data — do it tonight
The Advertising API requires a completely separate registration process
Have a backup developer on standby for API downtime, especially during peak sales periods
AI-generated interfaces all look the same because they default to the same fonts, colors, and layouts
A design system isn't about making things pretty — it's about trust, conversion, and consistent user experience across all devices
Before your customer reads a single word of copy, your page load time and visual quality have already made an impression
Coming Next:
Shubhash experiments with running AI models locally on an old MacBook using Ollama — cutting token costs to zero.
Podden och tillhörande omslagsbild på den här sidan tillhör
Danny McMillan. Innehållet i podden är skapat av Danny McMillan och inte av,
eller tillsammans med, Poddtoppen.