Built a full AI SaaS with Claude Code — here's what actually worked (and what didn't)

New here. Built an AI product photography platform from scratch using Claude Code over the past 2 years — React + Express + 15 AI models integrated (Gemini, Kling, Veo, Flux, etc.).

The biggest lesson: Claude Code is insanely good at building features fast, but the real bottleneck is always architecture decisions BEFORE you start coding. I wasted weeks letting Claude scaffold things that I later had to completely restructure because I hadn't thought through the data model.

What worked: giving Claude extremely specific context about my existing codebase (CLAUDE.md files, strict rules) and treating it like a senior dev who needs a clear brief — not a magic wand.

What didn't: asking Claude to "figure it out" on complex multi-service integrations. The more ambiguous the prompt, the more refactoring later.

Curious about the builders here — do you find Claude Code works better for greenfield projects (building from zero) or for adding features to an existing codebase?

12 comments