Claude Code built 90% of my SaaS — here's what it can't do

I've been using Claude Code as my primary development tool for the past months on a React + Express SaaS (7 languages, Stripe billing, Supabase auth, multiple AI API integrations). Here's an honest breakdown of where it excels and where it still breaks down.

What Claude Code handles remarkably well:

Full-stack feature implementation from a single prompt (routing, API, frontend, DB migration)
- Refactoring across multiple files while keeping everything consistent
- Writing complex backend services (API integrations, webhook handling, credit systems)
- Debugging — it reads error traces and fixes issues faster than I ever could manually
- Translation management across 7 language files simultaneously

Where it consistently struggles:

Complex CSS/visual design — it writes functional Tailwind but has no real "eye" for spacing, hierarchy, or visual polish. You still need a human design sense.
- Large file context — once a component exceeds ~500 lines, it starts losing track of what's already in the file. You need to architect smaller modules proactively.
- Third-party API quirks — when an API behaves differently than its docs say (which happens constantly), Claude Code will keep trying the documented approach instead of adapting.
- State management across sessions — it doesn't remember what it decided 3 hours ago unless you maintain explicit documentation.

The meta-lesson: Claude Code is an extraordinary implementation tool, but it doesn't replace architectural thinking. You still need to know WHAT to build and WHY. The "how" is where it saves you weeks.

Curious about your experience — do you find Claude Code better at greenfield projects (building from scratch) or at adding features to an existing codebase?

2 comments