Intention > Automation (unworkableideas.com)

This piece argues that the reflex to automate everything — especially in AI/ML — is mistaken: automation often replaces human attention, judgement and social engagement with cold prediction. “Personalization” is frequently just prediction acting on behalf of users, not with them, and as we remove friction we also remove the small acts of attention that create trust and discernment. For practitioners and product teams, that’s a warning that more automation isn’t automatically better; it can hollow out the human capacities models are meant to augment. Technically, the essay invokes a diminishing-returns dynamic (a “Law of Return on Information”) where more data and layers of automation reduce marginal value and increase abstraction between people and work. The practical implication for AI/ML is to design human-centered systems: prefer human-in-the-loop patterns, selective automation, and deliberate preservation of productive friction. Rather than optimizing solely for efficiency or scale of motion, teams should ask what a system is for and whom it serves — choosing what to automate and what to protect so that “intention scales meaning,” not just motion.