As someone who has spent quite a lot of time on inference, I would a add a small note:
Deployment looks very different for MoE than dense style models so I would say that it is more nuanced than "inference memory reqs remain the same". Memory can be very different for MoE style models.
I find his views on user friendliness to be very interesting. In fact, with the advent of agents this view can be challenged as the user is the agent who doesn't share the definition of user friendliness in the context of humans.
Unlike, humans being education resistant in understanding new ideas/products , agents are not like them.
tl;dr Use Claude Code in WSL inside Windows Terminal? Copying an image in Windows and pressing Ctrl+V in Claude Code doesn't work. Three things break: (1) WSL only hands Windows images to the Linux side in an old BMP format Claude Code can't read; (2) WSL also keeps quietly overwriting your fixes a moment later; (3) Windows Terminal grabs Ctrl+V before Claude Code can see it. The fix is a small Windows program that converts the image to PNG, a Linux script that puts it on the Linux clipboard (and re-asserts once after WSL overwrites it), and one extra keybinding for Claude Code so the keystroke actually reaches the program.