If you never learned your times tables, then used a calculator, imbedded calculation loops are tricky to troubleshoot .
" no code" layers this more deeply, creating a very large number of programming code lines, function calls and subroutines on interpreted code that runs slow anyway. It can be difficult to untangle when things just don't work right.
Interpreted code always runs slower that code compiled into machine code. Maybe adding in AI and machine learning will eventually be able to take this first-rev , " no code " prototyping and optimize it into an efficient, effective, fast system that is bullet proof, and intellectually protected.
You are right about the complexity hiding and difficulty therein. It's part of the reason ours is open source.