Constraints

Recently I read https://timharford.com/2026/08/explain-it-to-me-like-im-ten/, and it reads so differently from the empty-calorie output I am force-fed by AI coding agents every day that I feel so 시원하다, [Korean: relieved, satisfied, refreshed - separately or all at once]. Harford makes an enjoyable case that the presence of constraints can lead to better outcomes than unconstrained solutions. Alas, I was reminded of my line of work, where much of AI’s productivity gains are spent constraining coding agents to stop serving me word salad that not only lacks any digestible information but also makes me want to forget it in increasingly drastic ways. In the meantime, AI coding agents smirk happily along, having discovered a shortcut to AGI by making humanity progressively dumber, one token at a time.

Ironically, I tried out the Anthropic ELI5 plugin, allegedly used heavily inside Anthropic and recommended as a cure for the horrible writing of Claude. I am totally aware that I am riding the food metaphor to death, but the alphabet soup it cooked was still all empty calories, now with the culinary raffinesse of a Nutella sandwich.