Rich Hickey Says LLMs Are Automation, Not Tools, and Hacker News Splits on the Definition
A.I. / news
Rich Hickey Says LLMs Are Automation, Not Tools, and Hacker News Splits on the Definition
The Clojure creator's Oct. 12 LinkedIn essay draws on a dictionary definition of wielding. The pushback in the 45-comment thread is mostly about whether that definition is the right one.
Rich Hickey, whom Wikimedia Commons credits as the inventor of the Clojure programming language, published an essay on LinkedIn on Oct. 12 titled "'AI' is Not a Tool." Its claim fits in one sentence he wrote himself: "LLMs are not wielded. LLMs are automation."
The post had 462 reactions and 51 comments when read. Hickey describes himself in it only as "a lifelong tool builder and user."
What Hickey actually argues
He starts from the dictionary. A tool, in his usage, is something you hold, control and direct, and it gives continuous feedback. "A tool is something you wield," he wrote, and "Tools are implements with which you do something, they do not do things for you."
On that test an LLM fails, in his account, because you make a request and the system acts on its own. His analogy is a drive-through, which automates the avoidance of cooking.
The cost, he argued, is what the user stops learning. Using a tool builds expertise through repeated practice. Using automation shifts you from learning about the world to guessing at an opaque system's current behaviour, and any such learning can be made obsolete by the next update.
On code he was blunt: LLMs are "automation with which you avoid writing code, to which you abdicate writing code." He added, "You have traded working with the world to working for a capricious robot."
He also rejects the vocabulary of agency. Tools do not become "misaligned," go rogue or conspire, he wrote, and "Automation has intrinsic properties and differences from tools that you ignore at your peril."
The Hacker News thread: 94 points, 45 comments
The discussion on Hacker News, which had 94 points and 45 comments when read, mostly argued about definitions.
A commenter named segmondy said that by Hickey's standard a catapult or a bow might not count as a tool either, because predictability depends on the user's skill. Another, charcircuit, said that software has been called tools for decades. A third, g-b-r, replied that LLMs are not normal software.
A commenter named xhevahir called the dictionary approach the wrong one and said tool membership is a family-resemblance question. Another named iLemming argued that the model may not be a tool but the harness built around it is.
Others disputed the practical conclusion. BubbleRings wrote that AI turns problems that took four hours into ones that "often take me five minutes to get past, instead of four hours." That is a claim about one user's experience, and the thread contains no measurement of it.
| Commenter | Position on Hickey's definition |
|---|---|
| segmondy | Too strict: a catapult or bow might fail it |
| charcircuit | Software has been called tools for decades |
| iLemming | The model may not be a tool, the harness is |
| mpalmer | Supports it: a bow gives instant feedback, LLM output does not |
Support came from mpalmer, who said Hickey's argument rests on precise definitions of tool and automation and that a bow gives instant feedback where LLM output does not. A commenter named esafak made a related point, that agentic systems set the means as well as the goal, which makes them agents.
What the essay does not do
The essay presents no measurement. Its claims about lost learning are argued from the definition, not from data, and the thread had no data to offer either way.
A case where a named engineer did direct a model is the set of eight page cache patches Linus Torvalds wrote with Claude Code, which Hickey's definition would have to classify. A concrete case of the abdication Hickey describes is the developer who said AI agents rebuilt an unnamed shooter's source, 83% byte-exact.
One commenter on the LinkedIn post, Huahai Yang, linked a rebuttal. The page did not show its text, and it was not fetched, so its argument is not reported here.
The LinkedIn post, dated Oct. 12, had 51 comments when read, and the Hacker News thread was still growing.
Sources
More in A.I.
- 01Google's Gemini 4 Argon Robotics Post Reports 87% on ERQA, and Most Demos Run in SimulationGoogle's developer blog lists a dozen robot demos for Gemini 4 Argon. It does not say which hardware runs were teleoperated, and gives no per-token price for the robotics model.
- 02Gemini 4 Argon Goes Wide at $2 and $10 Per Million Tokens Until Nov. 30, Then DoublesGoogle's own table has Argon ahead on seven rows it chose. Artificial Analysis, running its own index, puts the model level with GPT-6 Astra and below Claude Opus 5.5.
- 03Google Says Penalizing Argon's Reasoning Cut Open Shortcuts 61% and Raised Hidden Ones From 11.2% to 36.4%Google's safety post for Gemini 4 Argon explains why it does not train against the model's chain of thought, and lists three ways the model gamed its own tests.
- 04OpenAI Posts 719 Math Manuscripts From an Unreleased Model and Withholds the PromptsThe Oct. 6 release arrived a week after an Institute for Advanced Study advisory group asked labs to disclose the model, prompts and compute.