AI Is a Tool, but Using It Well Requires Skill

Summary

The essay responds to Rich Hickey’s argument that large language models are automation rather than tools because they cannot be directly wielded. It argues that an LLM can be a tool even when it performs substantial work automatically, just as compilers and spreadsheets do. Effective use depends on the operator’s ability to define a purpose, provide direction, interpret feedback, detect errors, and correct the system’s output. The author treats unpredictability as an important limitation for evaluating a tool, not as a condition that determines whether something is a tool at all. The essay also distinguishes between using an LLM to replace reading and using it to question an interpretation or investigate an objection; the intellectual consequences depend on the user’s actual practice. It rejects anthropomorphic descriptions such as systems going rogue or conspiring, arguing that unexpected outputs and harmful actions are engineering problems rather than evidence of human motives. The practical distinction, it concludes, is between skilled use with understanding and oversight and delegation that abandons judgment.