Raw LLM Responses

It wont get that far. Well, not *really.* It might get that far from a economic standpoint because Executives are all about cost cutting, but they will end up very rapidly finding they need to hire more people when their whole infrastructure begins to collapse after a few weeks because of linguistic inertia. The main reason is this: "*only need debugging,*" That only is an absurd time sink. The AI is way better at debugging human code than it is at debugging its own, as there is a sort of feedback loop that happened when it debugs its own code. It is using its own production as its input to generate new prompts, which means patterns in the data have a sort of in-breeding problem. At the moment productivity gains have been pretty minimal. In my case code that I normally finished in 15 minutes has taken ~6-7 hours for the AI to figure out how to do, despite it having complete access to the source code. The initial plan, the first prototyping stage, and the structure of the project happened in about 30 seconds. Then the iteration started and burned a full day before it was able to figure out how to get it working. Importantly, this was done by me largely out of personal curiosity, so I limited myself to vibe-coding on purpose, only intervening to provide it with clarification and instruction in plain English, as well as to give it sources for it to read to learn about what it needed to do. If I had taken a more active role in the process it probably would have taken about an hour to get everything working. Which is still slower, but well within the realm of possibility that improvements in my workflow could get it down to my normal pace. But importantly, that would require me to be *active* which means I need to be there, working, the whole time.