In my case, I had in the previous weeks to refactor/fix a side project to solve bugs that were present since two years but never had any feedback from anyone (and some people were aware). They were serious and I had no time for fixing them with my ongoing schedule on other projects. I used a Claude code wired to a local model on a dgx spark. It helped, but frustrated me to death in this new position of “chief instructor/validator”. For the first time, developed without any pleasure, got frustrated so much to wait for long turns executions, that I was burning out, almost. I am yet to analyse if it helped me for real or took 4x time the time I would have used IF I WAS FOCUSED. Problem is that schedules are just impossible to manage and nobody gives a shit. Yet, I saw that as a code explorer tool, audit, review, it was useful. As a code editor it was almost unusable because it filled its context so easily. That’s the limits of local/privacy aware solutions, when you don’t have the time to set up the “sweet spot” for the LLM.
This time I had a preview of what my job will become if all this shit doesn’t reverse. Not only the AI bubble, but the management under AI drugs as well. My boss had trainings on the benefits of AI delivered by some lobbyists from a boss cult. They promised paradise. I envisioned purgatory.

There are three options:
1) I’m an LLM
2) I am a terrible writer
3) I’m so good that LLMs decided to emulate me