The post answers an October 11, 2026 article by Rich Hickey, "'AI' is Not a Tool," which argues that LLMs are automation because they cannot be wielded. The author, Huahai, concedes one point: outsourcing work can cost people chances to learn. He rejects the conclusion, arguing that predictability and reliability are criteria for judging tools, not requirements for belonging to the category. His main argument is that automation and tool use overlap. Compilers and spreadsheets do large amounts of work without step-by-step direction, yet users still direct them by setting inputs and checking results. Hickey's essay, per one summary, worries that a model reading research papers for you means you miss learning how to think. The Hacker News thread shows how builders split. One commenter says prompting workarounds from mid-2025 are now obsolete. Another says real engineers are still needed once an application starts to scale. For founders, the dispute bears on positioning. One commenter notes that AI sellers tend to pitch it as a do-it-all system rather than a tool, while users treat it as an instrument. Products that assume skilled operators need review, feedback and error-recovery features. Products that promise delegation will be judged on whether users can still verify the output.