Here’s one simple test: have AI “decompile” an existing codebase into a prompt, and then have it “compile” the prompt back into code. Is the app identical?