In an article published on October 11, 2026, titled "'AI' is Not a Tool," Rich Hickey argues that large language models (LLMs) are automation rather than tools because they cannot be wielded. The author of this response agrees that outsourcing work can limit learning opportunities but contends that Hickey's conclusion does not follow. According to the author, even by Hickey's definition, an LLM can be something a human wields, and using it effectively requires skill, similar to other tools.
The author compares using a saw or chisel to using an LLM, stating that tools can amplify both poor and good judgment, and their effectiveness relies on the operator's ability to direct them, interpret feedback, recognize mistakes, and make corrections. While acknowledging that an LLM's unpredictability is a limitation, the author argues that predictability, reliability, simplicity, and bounded function are qualities used to evaluate tools, not prerequisites for being classified as a tool.
The author also critiques Hickey's distinction between tools and automation, asserting that automation can be a feature of a tool. For example, a compiler and a spreadsheet perform significant work without constant manual direction, yet users still wield these systems by specifying inputs and examining results. The author questions why generating text automatically should disqualify an LLM from being considered a tool when other automated processes do not face the same scrutiny.
Addressing Hickey's claim that LLM use lacks intention and expertise, the author argues that a programmer can engage in a feedback loop with an LLM, asking for explanations, challenging responses, and refining requests. This process requires knowledge of what to ask and how to evaluate the responses, indicating that expertise can enhance the usefulness of the tool.
The author acknowledges Hickey's valid concern regarding the potential for LLMs to replace reading but emphasizes that users can still engage deeply with texts while utilizing LLMs to explore interpretations or objections. The author cautions against the anthropomorphic language used in the AI industry, such as "go rogue" or "conspire," suggesting that these terms may misrepresent the nature of software and its outputs.
While the author concedes that LLMs do not always follow instructions perfectly, they argue that this does not negate their status as tools. The distinction lies in the ability to wield an instrument skillfully versus delegating tasks without sufficient understanding or oversight, a distinction that the author believes is crucial in discussions about AI.