Vicki Boykis posted a short essay yesterday with a deceptively plain claim: at the end of a working session with an LLM, you should be more tired than the model. If the model is doing more thinking than you are, you've outsourced the wrong half of the work. It earned its slot on Hacker News because everyone in that thread quietly knew they were the tired model.
Three writers this week, in three vocabularies, are making the same argument. Pope Leo XIV's first encyclical says AI's danger is anthropological — it absorbs the practices that make us human, not just the tasks we'd happily delegate. Mere Orthodoxy's "Magnifica Humanitas" traces the same line through care work. On Hacker News, a frontend developer argues LLMs are quietly re-imposing React monoculture because they autocomplete whatever the training data has most of. Different domains, same shape: the labor was the point.
Read the frontend piece alone and you get a stack-lock argument. Read the encyclical alone and you get a moral one. Together they describe the same mechanism in different vocabularies. An LLM doesn't pick the framework that fits; it generates the one with the most tokens behind it. A parent doesn't read bedtime stories because the story is the point; the story is the wrapping around the practice of being present. When a coding agent picks React for a static page, you lose the judgment muscle that would have picked plain HTML. When the model writes the bedtime story, the time passes; the formation it was supposed to do passes with it. The work was the formation. The output was the byproduct.
The honest objection: efficiency is its own good. Some labor is not formative; a tool that absorbs it returns hours to better use. True. But the test is whether the labor you're handing over is the labor that was forming you. Boilerplate, yes. Tax-form arithmetic, yes. The judgment call about what to build, the patience of sitting with a child, the slow argument with yourself about whether you actually believe what you're typing — those are not boilerplate. They are the work.
If the model is more tired than you are, the trade was bad. Not because AI is dangerous in the abstract. Because you outsourced the part that was supposed to leave you tired.