> Anthropic's more technical users inherently understand how LLMs work.

Yes, I too imagine these "more technical users" spamming rocketship and confetti emojis absolutely _celebrating_ the most toxic code contributions imaginable to some of the most important software out there in the world. Claude is the exact kind of engineer (by default) you don't want in your company. Whatever little reinforcement learning system/simulation they used to fine-tune their model is a mockery of what real software engineering is.