I fundamentally disagree with this take. With regard to higher-level languages, code is a way to create and compose abstractions around data flow, plus a means of communicating this to other developers (including yourself in the future). Every project brings something new to the table. Machines "solving" this is tantamount to machines developing full-on human intelligence and judgement. The best they can provide is a static facsimile without much hope for future innovation or progress: a snapshot of programming best-ish practices circa 2024 or so.
And remember: "just needs to work" includes technical debt, which LLMs are especially great at accruing. When your codebase is composed of 10,000-line, randomly-generated PRs that nobody reads or understands, it's going to be full of untraceable heisenbugs and egregious security holes. What's going to fix them? More 10,000-line PRs? Is it turtles all the way down?
Don't mistake my rage at the egregious state of the industry with sadness. In a year, we've thrown much of everything we've learned about solid engineering out the window.
And remember: "just needs to work" includes technical debt, which LLMs are especially great at accruing. When your codebase is composed of 10,000-line, randomly-generated PRs that nobody reads or understands, it's going to be full of untraceable heisenbugs and egregious security holes. What's going to fix them? More 10,000-line PRs? Is it turtles all the way down?
Don't mistake my rage at the egregious state of the industry with sadness. In a year, we've thrown much of everything we've learned about solid engineering out the window.