@djsumdog @coolboymew I still think it's worth knowing how to program and always will be. I don't think any LLM is immune to model collapse so we could see a scenario where vibecoding because useless. Basically, AI generated content will have to find a way to coexist with new human generated content in order to prevent AI from feeding on it's own data, leading to worse results.
https://en.wikipedia.org/wiki/Model_collapse
When it comes to gamedev, I'm still going to program things the old fashion way because I enjoy it. I only use AI for graphics and even then, I strive to do image-to-image conversions using rough drafts I made myself, so in that regard it's more AI-assisted rather than AI-generated.
I mean, I could probably just write a design doc for my dream game and feed it into Claude right now, I don't want to because I would never be satisfied, if I'm not involved with most of the game's development.
@icedquinn @djsumdog @coolboymew
>lot of people don't really care that much about quality or maintainability
They will care if the output quality degrades to the point where we go back to 2023 level output where the code doesn't even compile half the time.
You gotta understand how the underlying thing works, at least I think you do. It becomes more essential when you have really complex systems, archaic old IBM endpoints, data integration across multiple parts of your organization, decades of technical debt ... maybe it will just depend on the types of tasks.
I still think subject matter experts are going to be valued, but identifying them is going to be increasingly difficult.
No one in the industry seems to be able to answer me when I ask "How do you train a junior engineer in this environment?"