This is a valid approach, but not because of its "speed".
When you start a project, you have a great many things you don't know and need to learn. This is the case in 99% of projects out there. By definition, you cannot at the start of it optimize for things you don't know anything about, because... how, exactly?
If you work in a particular stack a lot, prepare a starter project template. Put some time into it, set it up so that it is optimized for letting you learn the things you don't know and helping you discover and make architectural decisions as they become necessary to make. And then just go wild.
This is the bottom-up programming approach advocated by pg in his book, IIRC. Have a setup where you can iterate quickly in the small, and make it easy to abstract and refactor as you go. You don't need a Lisp for that, just some tools with proper configs. It's also what R&D guys tend to do naturally with their Jupyter Notebooks - the difference, other than the kinds of problems faced, is that devs don't have the luxury of leaving the code in a form of REPL session transcript.
When you start a project, you have a great many things you don't know and need to learn. This is the case in 99% of projects out there. By definition, you cannot at the start of it optimize for things you don't know anything about, because... how, exactly?
If you work in a particular stack a lot, prepare a starter project template. Put some time into it, set it up so that it is optimized for letting you learn the things you don't know and helping you discover and make architectural decisions as they become necessary to make. And then just go wild.
This is the bottom-up programming approach advocated by pg in his book, IIRC. Have a setup where you can iterate quickly in the small, and make it easy to abstract and refactor as you go. You don't need a Lisp for that, just some tools with proper configs. It's also what R&D guys tend to do naturally with their Jupyter Notebooks - the difference, other than the kinds of problems faced, is that devs don't have the luxury of leaving the code in a form of REPL session transcript.