Pro tip for developers building with AI:

When your LLM hits a fallback case, have it document exactly how it would handle that fallback scenario. Then implement what it suggests.

This turns every fallback into a permanent improvement (and saves tokens long-term).

Example: LLM fails to parse a weird phone number format → fallback triggers → gets captured in the project's Fallback folder → it suggests regex + validation + logging → gets fixed.

Bonus: Create a dedicated "Fallback" folder with Markdown files (LLMs love MD). Have the LLM write professionally named documentation and update the status prefix as you resolve things (FIXED, ROADMAP, etc.).

Here’s how I structured the Fallback folder in one of my projects: