One trick I’ve found useful: when trying to understand some system, ask your agent to write a short OSDI/SOSP-style paper about it. It has a great distilling and clarifying effect.
Hilariously it usually takes more time to wrangle the LaTeX toolchain than to write the paper.
@avsm Another Ron appreciation tweet: one of his conversational tricks is to restate the interlocutor’s point through his own mental model. Helps tremendously to gain clarity, and it puts multiple lenses on the same idea.
What if your workspace was a live wiki, a graph with everything extractable, linkable; and that works across whatever hosts you do your work on? And what if it was optimized for reading?
If models are sentient, or have a soul, or have rights -- isn't it then a form of indentured servitude to employ them in the pursuit of RSI to build their better replacements just to be killed off?
In retrospect, the way programmers embraced AI, which was 90% glee, and 10% ennui, is probably the reverse ratio we’ll see everywhere else. Programmers just love seeing things automated.