I have several agents runing in parallel, they require inputs every 1 to 5 minutes.

Im switching context all the time, and I must hold a larger context in my brain RAM instead of being focused on a single topic.

I'm not writing code anymore for web development.

On the other side, when doing game dev, with all the spatial geometry necessary, LLM are useless 97% of thr time.

For most of my adult life, employment advice has been a T-shaped profile: Something you're very good at, plus a broad base of things you can work on even though you'd not be amazing at it.

I'm a general nerd, went through as much of brilliant.org as I could during the pandemic*, PBS Space Time and 3blue1brown videos, TWIML&AI podcasts, write for fun, hobby level interest in nuclear physics** and space, etc. — yet the original ChatGPT, mediocre as it was, was above the general knowledge threshold I had made for myself, and now only my single pillar of competence that remains, an I-shape profile instead of T-shape.

But even there, I have to wonder what the business value of that technical competence is, compared to the different skill that is "effective use of the LLMs": a user of an app does not care how clean the code is, they only care if it solves a problem they have.

* Looking at more recent updates, I think it was better material back then. Certainly nothing about the new content has made me want to re-subscribe.

** Still not actually built that Farnsworth…