> it does not seem a lot more different that starting with a prototype
People underestimate this point. Spreadsheets are one of the best possible approaches for prototyping data-driven applications, thanks to its dual data/code nature and the ease to build and extend data types (just add more columns to a table); it's like building the application inside a debugger.
Non-developers don't have anything else that resembles a debugger, anywhere in the common approaches of software for end-users in industry.
Combine this with spreadsheet applications working as integrated environments -a single tool that handles all the computing needs of a project, without having to build a toolchain-, and it's no wonder than it's the preferred method for non-programmers to build custom automation workflows when their needs aren't supported by any specific software.
> Spreadsheets are one of the best possible approaches for prototyping data-driven applications,
Bingo. Back when I worked in big corp Excel was the prototyping tool for the masses. If someone in a remote office built something in Excel that solved a problem that was useful to other offices my team would come in and use that work to build a real system. It was great because a lot of the requirements discovery work happened organically before we would even hear about the process/Excel.
Admittedly, we were a small team and could only take on so much work. This led to Excel being used beyond its capabilities which leads to a host of other problems.
People underestimate this point. Spreadsheets are one of the best possible approaches for prototyping data-driven applications, thanks to its dual data/code nature and the ease to build and extend data types (just add more columns to a table); it's like building the application inside a debugger.
Non-developers don't have anything else that resembles a debugger, anywhere in the common approaches of software for end-users in industry.
Combine this with spreadsheet applications working as integrated environments -a single tool that handles all the computing needs of a project, without having to build a toolchain-, and it's no wonder than it's the preferred method for non-programmers to build custom automation workflows when their needs aren't supported by any specific software.