AI Is a Tool. Using It Well Takes Skill.

You still need to develop skills to effectively use AI, just as you would with any other tool.

Rich Hickey's argument that Large Language Models (LLMs) are not tools but automation is contested. The author agrees that outsourcing tasks to LLMs can hinder learning opportunities. However, the author posits that LLMs can be wielded like tools, with their effective use requiring operator skill, interpretation of feedback, and error correction. Predictability and reliability are presented as evaluation criteria for tools, not prerequisites for their classification. The author also challenges the distinction between tools and automation, noting that automation is often a feature of tools, such as compilers and spreadsheets, which are still directed by human input and decision-making. The author argues that LLM interaction involves a human-directed feedback loop, requiring expertise to ask effective questions and evaluate outputs, similar to other complex tools. The reliance on anthropomorphic language by the AI industry is also criticized for obscuring engineering challenges with notions of independent agency.

The core debate re-frames AI interaction from passive consumption to active wielding, suggesting that the perceived "automation" of LLMs is more akin to advanced features within existing tool paradigms like compilers or spreadsheets. This perspective implies that the development of human expertise in prompt engineering and output evaluation is paramount, rather than a fundamental shift in how humans engage with technology. The argument posits that current critiques of LLMs, particularly concerning their "unpredictability," are simply characteristics of complex tools, not disqualifiers from the category.

This analysis suggests that the industry's tendency towards anthropomorphic language is a significant misdirection, obscuring real engineering challenges under the guise of artificial agency. By categorizing LLMs as tools, the focus shifts to operator skill and the feedback loops involved in their application, implying that the current focus on AI's supposed independence is misplaced. The author's stance advocates for a more grounded understanding of AI capabilities and limitations, emphasizing the human role in directing and validating AI-generated outputs.