By the way, during the hackathon materials, I came across a very sensible slide. I can already see a huge difference in working with vibecoding (here they call it AI-DLC (AI-Driven Software Development Lifecycle)) between developers and non-developers, and consequently, general dissatisfaction.
Yesterday, I watched a stream of an entrepreneur who says that models are getting worse and worse at handling his project tasks. The project is expanding, and he attributes this to a conspiracy theory, claiming that models are intentionally being made dumber, while his partner (a developer), on the contrary, says he sees progress.
Listening to how they work, it quickly became clear that the entrepreneur simply throws a poorly described task into the AI and forces it to work for 16 hours straight, expecting a good result. Naturally, over time, the code increases, and the result worsens.
Meanwhile, the developer is satisfied because he decomposes the task into blocks, controls these blocks, understands where everything is heading, and timely "cleans the code," etc.
#ai@rvnikita_blog #coding@rvnikita_blog #vibecoding@rvnikita_blog #llm@rvnikita_blog
