When it inevitably all comes crashing down because there was no actual software architecture or understanding of the code, someone will have to come in to make the actual product.
Hopefully by then we will have realistic expectations for the LLM, have skilled up, and we as a community treat them as just another feature in the IDE.
But beyond that, I'm really not looking forward to trying to discover new good libraries, tools, and such in 5 years time. The signal to noise is surely dropping.
My experience has been more that they expect you to fix the broken mess, not rebuild it properly.
With the ambiguity in the meaning of "VC" being intentional.
The idea that code gluers understand the code better than a chatbot (that admittedly totally lacks the ability to understand) isn't clear.
Just like an LLM they can only create very ill-thought-out ripoffs of software from the 70s and 80s (they have yet to reach the level of ripping off 90s Microsoft.)
Why would a window maker be against breaking windows?
More realistic: AI assisted tooling will continue to improve as it has, the average code quality will rise as conventions and workflows improve and those who wait to be called in to clean up slop or whatever will wait forever, pushed by the wayside by those who can deliver great quality with the help of these new tools.
That sounded a lot like the "have fun staying poor" argument from the peak cryptocurrency days.
Current gen AI has a ton of issues, but it nevertheless enables vast amounts of use cases today, right now.
And hoping that slop that is created today will provide work for the artisanal craftsman in the future is wishful thinking at best.
You set up your prompt, CLAUDE.MD, include the source files and let it rip. It gets some things right, some things wrong. You fix some things manually, /clean and go again. Sometimes you gotta throw it out and start over. Feels like "most players stop just before striking it big".