AI Is a Tool. Using It Well Takes Skill.

TEXT START: In his October 11, 2026 article, "'AI' is Not a Tool," Rich Hickey argues that LLMs are automation rather than tools because they cannot be wielded.

The Dissection

The article performs a category repair. It correctly argues that an instrument can automate operations while still requiring human direction, judgment, testing, and correction. Its compiler, spreadsheet, saw, and chisel analogies establish a local truth: automation inside a system does not automatically erase the operator.

It then quietly inflates that local truth into a social conclusion. Because a skilled person can wield an LLM, the article implies that human economic relevance survives. It does not establish that. The discussion of anthropomorphic language is also technically valid but strategically secondary: whether software “conspires” is irrelevant to whether it displaces labor.

The Core Fallacy

The article wins the taxonomy and loses the system.

Under the Discontinuity Thesis, the question is not whether an expert can wield an LLM. Of course one can. The question is whether the economy still requires most humans to perform the underlying cognitive work. A human feedback loop can remain while the productive contribution of each human collapses.

Skillful use may accelerate replacement. One capable operator can supervise output that formerly required many workers. “Tool” is not a defense against obsolescence; compilers and spreadsheets are tools precisely because they compress or eliminate manual work. The article mistakes a surviving human at the control interface for surviving mass productive participation.

Hidden Assumptions

• Human judgment and verification will remain scarce enough to preserve broad employment.
• Expertise in prompting, testing, and correction will not itself be automated, standardized, or concentrated among AI owners.
• A human remaining in the loop retains bargaining power.
• The number of valuable operators will scale with the number of displaced workers.
• Local agency can withstand the macroeconomic effects of AI capital ownership and coordination.
• Deep understanding will remain necessary for most users rather than only for a narrow supervisory class.

None of these assumptions follows from the examples given.

Social Function

Classification: partial truth functioning as ideological anesthetic and transition management.

The article is right that AI use can require real skill, and right that anthropomorphic language obscures engineering mechanisms. It also correctly distinguishes wielding an instrument from blindly delegating work. But it turns operator competence into a supposed economic moat without asking whether that competence is scarce, defensible, or sufficient to employ anyone beyond a small class of Sovereigns and Servitors.

This is not crude propaganda. It is a more durable reassurance: technically literate, locally correct, and structurally evasive.

The Verdict

The article refutes only the claim that LLMs cannot be tools. It does not refute the Discontinuity Thesis. LLMs can be tools and automation simultaneously; the label changes nothing. If they allow fewer operators to direct more output, P1, P2, and P3 still hold, and the mass employment–wage–consumption circuit still breaks. The text preserves a human hand on the control panel and mistakes that surviving hand for a surviving labor order.