Nice writeup! Been experimenting a lot with skills, agents and tools like GSD and honestly the biggest lesson for me has been: keep things short and simple.
Not just CLAUDE.md but everything, the prompts you type in a session, skill descriptions, agent configs, all of it. More instructions does not mean better results. Claude actually performs noticeably better with brief focused input. The moment you start over-specifying things quality drops.
Just my observation from using it daily but it's been consistent enough that I figured it's worth sharing.
There are drops every day, sometimes for quite a while. It's pretty annoying, especially when you've paid for the MAX plan and there's no compensation, or at least I'm not aware of any.
Honestly I think it's more speculation than actual shortage. TSMC is a bottleneck sure, but the bigger issue is demand side FOMO. Companies hoarding GPUs without clear ROI, hyperscalers building for projected demand that may never come. Feels a lot like the fiber optics overbuild in the early 2000s.
Fair point about set and setting, taking a big dose when you're in a bad place is asking for trouble. But the research and most microdosing protocols are talking about way smaller amounts, like 0.15-0.3g every other day. That's well below the 1.5-2g where you actually start tripping. Personally microdosing in that range has helped me a lot with focus and creativity, no trip, no weirdness, just a subtle shift. It's a completely different thing from what people usually imagine when they hear mushrooms.
The psilocybin data tells a similar story. One or two sessions outperforming months of conventional meds. The fact that both LSD and psilocybin work on neuroplasticity probably isn't a coincidence. Hope regulators don't drag their feet on this for another decade.
The data here matches what I've seen firsthand. Working remote I get roughly 30% more actual work done, mostly because deep focus is way easier to protect. The productivity argument for RTO always felt backwards to me.
Not just CLAUDE.md but everything, the prompts you type in a session, skill descriptions, agent configs, all of it. More instructions does not mean better results. Claude actually performs noticeably better with brief focused input. The moment you start over-specifying things quality drops.
Just my observation from using it daily but it's been consistent enough that I figured it's worth sharing.