What if we aren’t really code constrained? The big AI victory case is coding, and yes we can crank out far more code far quicker now.
But for that to be a clear win, it requires that before AI there was loads of great software we wanted to write, but couldn’t afford to.
I am increasingly doubtful this was ever the case.
Software has no production cost and very small distribution cost. Once a team, out of the millions of programmers worldwide, comes up with a good new solution, everyone in the world can have access to it the same day. Good ideas spread immediately.
It’s often observed in the industry that pretty much everything in software was invented by the 1970s and since then we get maybe one big new thing per decade.
It’s also the case that most great software is worked out in brains and whiteboards during ideation, before any code is written. This means we have always been able to iterate on and discard ideas, explore the solution space, at a far grater rate than we can crank out code.
So, after 50 years of millions of programmers mostly failing to come up with important new software on even a yearly cadence - perhaps a coding speed boost isn’t really that relevant an upgrade to the system.
The limitation to AI coding productivity boosts might simply be that there aren’t that many great new software systems needing to be written.