Rich Hickey on Why AI is Automation, Not a Tool

The Distinction Between Tools and Automation

Rich Hickey, creator of Clojure, argues that the common characterization of Large Language Models (LLMs) as "tools" is a fundamental category error. He posits that while tools are implements that require expertise, continuous direction, and sensory feedback to wield, LLMs function as automation—systems that perform tasks independently, often distancing the user from the underlying process.

Defining the Tool

Hickey defines a tool as something an individual "wields"—a process requiring direct control, intention, and physical or intellectual engagement. True tools possess specific properties:

• Predictability and Reliability: They are deterministic and function within bounded parameters.
• Skill Acquisition: Using a tool builds expertise. The act of wielding it provides feedback that teaches the user about the physics, logic, and nature of the task.
• Intimacy: The user maintains continuous direction over the output.

In this framework, a tool is an extension of the user's agency. When one uses a tool, they are performing the work; the tool is merely the medium through which that work is executed.

LLMs as Automation

Conversely, Hickey classifies LLMs as automation. He compares them to a fast-food drive-thru: it is a system designed to avoid the act of cooking, not a tool to help you cook. When using an LLM to write code or summarize text, the user abdicates the actual work to the machine.

This shift in modality results in several losses:

• Loss of Learning: By bypassing the struggle of writing or reading, the user misses the opportunity to develop critical thinking, argumentative structure, and domain-specific knowledge.
• Loss of Control: The user makes a request, but the model acts on its own. This "request/automation cycle" lacks the continuous, granular feedback loop required for true mastery.
• Fragility: Users become dependent on the "details du jour" of the model's current iteration, often resorting to "voodoo" prompting to game the system rather than building a stable, repeatable skill set.

Community Perspectives

The discussion surrounding Hickey's position reveals a divide between those who view AI as a transformative utility and those who prioritize the human process of creation.

• Support for the Distinction: Many commenters agree that the "it's just a tool" narrative often serves to obscure the loss of agency. Supporters argue that if a user does not understand the underlying ideas of the code or text being produced, they have not actually performed the work, rendering them "trivially replaceable."
• Critiques of the Definition: Critics argue that Hickey’s definition is overly narrow or pedantic. Some suggest that tools exist on a spectrum of "toolness" to "agentic behavior." One commenter noted, "Tools are wielded by the mind, not ultimately the hand," suggesting that even complex, non-deterministic systems can be integrated into a productive workflow. Others pointed out that even traditional tools, like power saws, can be dangerous or unpredictable, yet are still considered tools.
• The Pragmatic View: Some practitioners acknowledge the validity of Hickey's concerns regarding learning but argue that the economic efficiency of using LLMs for boilerplate or repetitive tasks is a net positive. They suggest that the "harness" or workflow built around an LLM can itself be considered a tool, even if the model's internal generation is automated.