The Way Houyhnhnms Compute (2015)
ngnghm.github.io
ngnghm.github.io
Seems like you should be able to say "new values of y are entered by users, x is the sum of all y, z = 2x, every night text me the value of z" and compile that to a cross-platform native UI + cross-cloud api gateway + serverless function + persistence layer in a few lines of code as an afternoon exercise in an elementary school class, and such a runtime could extend its use cases to a decent portion of the CRUD much of industry toils away at, without having to itself become essentially complex.
There's lots of assumptions in our systems that are counter to simplicity, or simply counterintuitive. People expect things to be declarative (a = 0; b = 2; a = b; What? But a can't be 2 and 0...) persistent and consistent by default (what do you mean I have to save to disk and re-load it every time, why do two people both think they have this same table reserved, wtf is a partition), replicated by default (why can't my database be like google sheets... wait, why can't it BE a google sheet?), safe by default (why is this negative when it only ever goes up... overflow? wtf, why is that even allowed!? Wait, there are special things called bignums... why isn't that just every number all the time?) etc.
It seems lots of production software for humans that operates on human timescales can take the performance hit that these more ergonomic set of assumptions would require and many would be better off for it.
They could be the initial set of assumptions in an academic setting so you don't have to trip over the machine all the time in the early stages. You could then pare back one at a time with exploration of the benefits of the trade-off and cases where they are necessities (e.g. higher txn rates than a synchronous multi-master datastore could handle) so that instruction in the realities of computing come in bite-size and are paired with ample motivation to get over each hurdle (instead of upfront pain followed by latent payoff).
Without scraping the sharp edges of implementation detail, we could focus on design of these things -- both in terms of considering performance or maintenance headaches and in terms of applicability to real-world usage -- so there'd be more commonality between the fiddler who means well "non-coder" type who has unmaintainable SFDC schema or pet spreadsheet hell messes and the annoyed reality won't succumb to his ideal "software engineer" type banging out strongly typed highly normalized models in their ivory tower with no sympathy for end user (and yet also nothing to which they can apply their proud knowledge of algorithms or data structures or machine sympathy).
I have often wondered how a hypothetical alien would perceive our computer systems... It's wonderful to see someone has thought about it too (and they made it fun to read about!)