I've been reducing the amount of dependencies that my work projects have every year. I plan on continuing to reduce the amount of dependencies we have. There isn't a world where I'm going to just start having an LLM fabricate them for me instead. I'm just reducing the need for dependencies. I don't want to suddenly have to maintain them myself.
Even if you generate your dependencies by telling an LLM to reproduce the API of some popular library you now have an imitation of that library. Maybe it's perfect. Maybe it's not. Even if it's a perfect replica, if a CVE is discovered in the dependency you copied you're not going to be alerted, and you're not going to be able to just bump the version you depend on to get the fix.
Your dependencies will also stagnate. You're not focused on developing them, you're focused on developing your product. You're not going to benefit from the new things added to future versions of the dependency that you just copied. You may miss out on new features that make your codebase easier to use, more performant, or easier to maintain.
@fancysandwiches This is worse than when I heard people were deleting unit tests because, "They passed, so I don't need them anymore."
@Azuaron lmao wut
@fancysandwiches I can't find it now, but some vibe coder had the system functionally where they wanted it, but the build was taking too long, so they just... deleted all the passing unit tests.
Mar 6, 2026, 12:31 AM@Azuaron
@fancysandwiches
When I am forced to review LLM code, the odds that any random change will also include "delete all the tests" is shockingly high
Mar 6, 2026, 12:33 AM@fancysandwiches
Why on earth would I want to make my own numpy when there is a perfectly good numpy right there for free. That sounds like sucker behavior