Just a short one – since the recent release of Opus 5 I noticed that all of my sloperation seem to have even more comments in the code now than ever.
On one hand is kind of good for future generations [of agents] to figure out this or that thing, but it seems to be very excessive and deliberate.
My theory is that the model is incentivized to do that by either claude code itself or it’s baked in the model weights to produce more output tokens.
I don’t have other good explanation for this where the task was to just swap some translation strings:
Also my colleague shared this recent snippet of his work:
