In seriousness, the upside with code assistants is increasingly clear. And may be the bet that these would continue to improve quadratically is worth taking if you're a VC. In fact, PG argues that buying into hype cycles is the default: https://paulgraham.com/herd.html
So far, the utility on existing codebases is less than zero.
> continue to improve quadratically
Using what metric?
I'm curious what definition of "Utility" you are going by here - at the most basic level of "smarter, more context-aware autocomplete", does it have zero or negative value to a developer? Would you disagree that even an experienced developer could save some amount of time, at the current state of the technology, even if only at the stages where you're just writing code you already have a pretty clear concept of in your head?
Unfortunately that's not how things work under capitalism. To the extent that coding assistants make a difference in productivity (in my experience quite limited gains), that will just mean that you get more work done for the same time and salary.