Yves Van Goethem 
@yvg@indieweb.social
Been experimenting with #zed + #claude a bit, a few things are very odd:
1. The quality of interactions I'm having with Claude inside Zed is of lower quality than in VSCode + Claude or in Cursor. It seems "dumber", assumes more, doesn't verify claims, just generally more nonsensical answers and needs more hand-holding.
2. It seems to consume a lot more tokens, I'm not sure why, I suspect they don't properly cache, don't compact, don't optimise input/outputs, etc. This is horrific.

