Hot take: People dislike anthropomorphizing vocabulary when talking about LLMs. Words like “think”, “dream”, maybe even “you” when talking to an “it”. However due to how the beast works, it is actually immensely useful to refer to it as a human, not because it is, but because on the vernacular it was trained on. However it should be possible to re-train a model that will completely separate how humans function from LLMs. That would open some interesting options.

One possible future for programs will be a solid core, with a plugin architecture that can hook almost anywhere (think WordPress or the old XUL based Firefox) and is capable of protecting the program well enough.

On top of that people can vibe-code their plugins. The conversations that built them would be stored, with compacted contexts and plans. On core update or new extension addition everything will be regenerated.

Sharing would be done only through extraction of specs. No code would be able to be shared because unless all of the plugins are generated at the same time, there is no guarantee that the code would work.¹

If your email had the most important part at the beginning; I would not have to summarize it using AI.

We put so much value of convenience, that we send signal to the younger generation that learning basic skills is a waste of time.