From Autocomplete to Auto-Disaster
By the Editor · Opinion
I am currently experimenting a lot with AI coding and vibe coding in general. I am kinda astonished how far it has gotten and how much easier it is to write boilerplate code. Tasks that took me an hour are now finished in 5 minutes. That is a crazy amount of time saved, one would think.
But when I take a step back and look at it, my feelings change. The diffs in between me accepting answer all looked pretty good and yet the end results is a hot mess. My tech debt regularly increases by leaps and bounds with no end in sight. Even with prompts reduce the tech debt and prompts for better coding quality only help superficially. What was a good and helpful tool in the beginning begins to struggle as soon as the project has more than 50 files. Architecture that you carefully prepare is often ignored and completely circumvented for a quick fix or implementation. Even a planning mode never catches most obvious edge cases. Turns out you don't need bad architecture to get bad architecture vibe coded.
I am re-thinking my approach in working with AI in general and in what manner I should use it. A localized approach is probably the way to go for a simple feature.