'AI' Is Not a Tool by Rich Hickey

Rich Hickey argues that large language models and similar AI systems are not properly called tools because they are not "wielded." A tool, he says, is an implement you hold and actively control, receiving sensory feedback and building skill through continuous, intentional use. LLMs lack that intimacy: users issue requests and the system acts autonomously. Hickey contrasts predictable, bounded, deterministic tools (hammers, handheld implements) with automation like drive-thrus and LLMs, which operate independently and whose behavior can change beneath the user's feet with a software update. Attempts to treat LLMs as tools misread their nature.

Treating LLMs as automation produces concrete losses and risks. Relying on them to read, write, or code abdicates learning - users stop developing judgment, rhetorical skill, attention to detail, taste, and domain expertise. Gaming models with prompts or "incantations" yields brittle knowledge that evaporates when models change. Unlike tools, automation can be capricious and misaligned, and using it indiscriminately trades working with reality for chasing a moving target. The correct stance is to recognize intrinsic differences between wieldable tools and autonomous systems and adjust expectations, practice, and responsibility accordingly.