Rich Hickey's 'AI Is Not a Tool' Draws a Same-Day Reply Arguing His Own Definition Allows It
A.I. / news
Rich Hickey's 'AI Is Not a Tool' Draws a Same-Day Reply Arguing His Own Definition Allows It
Hickey says an LLM is automation, like a drive-thru. Huahai answers that by that definition a person can still wield one.

Rich Hickey wrote on October 12 that large language models are automation rather than tools, and a rebuttal posted the same day argues that his own definition of a tool lets a person wield one.
Hickey's essay, titled "'AI' is Not a Tool," appeared on LinkedIn and was mirrored as a GitHub gist. InfoQ introduced him in a 2012 interview as the author of Clojure and designer of Datomic. The Hacker News thread on the essay stood at 303 points and 180 comments on Wednesday.

What Hickey's essay claims
Hickey starts from the dictionary. He quotes one definition of a tool as "a device or implement, especially one held in the hand, used to carry out a particular function," and defines automation as "working by itself with little or no direct human control." His own preference is that a tool is something you wield, meaning you hold it, feel its feedback and command its course.
From that he draws a flat conclusion: "LLMs are not tools." and "LLMs are automation." His example is food. "A drive-thru is not a tool for cooking, it is automation for avoiding cooking."
The essay's central cost is learning. A person who makes a request and waits is not building skill at the task, Hickey writes, but guessing at the current behaviour of a system that a software update can change. He gives a custom burger order and a joke file called "voodoo.md" full of incantations as examples. "You have traded working with the world to working for a capricious robot."
He applies the same test to reading. A Kindle is wielded; an LLM summary is not. "It read the papers, not you."
He also objects to the vocabulary. "Tools don't get 'misaligned'," Hickey writes, and they do not go rogue.

The October 12 reply on yyhh.org
The rebuttal, "AI is a tool and using it well takes skill," is dated Monday, October 12 and is credited to Huahai, an author on the yyhh.org blog. The page gives no job title. It reached the top of Hacker News with 676 points.
Huahai accepts the dictionary and turns it around. "Even by his definition, an LLM can be something a human wields." The post argues that predictability and reliability are ways to rate a tool, not conditions for being one, and that compilers and spreadsheets automate work and are still tools.
On skill, the reply is blunt. "Picking up a saw does not guarantee a straight cut." And: "Tools can amplify poor judgment as readily as good judgment."
Huahai concedes one point. Hickey's worry about reading is fair when a summary replaces the paper, the post says, though an LLM can also support a careful read. The line it wants drawn runs between "wielding an instrument with skill and delegating work without adequate understanding or oversight."
Where the 303-point thread split
| Position | Who | Argument |
|---|---|---|
| Automation, not a tool | BatchJob, Barrin92 | An indeterministic system cannot offer mastery, and delegation erodes skill |
| Toolness is a spectrum | didibus, atleastoptimal | Hickey replied "Toolness is a spectrum," per didibus; use sets the position |
| Depends on the harness | iLemming, NateEag | The surrounding harness is the tool; Claude needs it to call rm |
| Narrow private definition | xhevahir | "Tool" is a family-resemblance word and Hickey has picked a narrow sense |
Commenter dasil003 argued that the label has a political edge: calling AI a tool keeps a human accountable for what it does.
What neither Hickey nor Huahai measures
Both pieces argue from definitions and analogies. Neither cites a study or a measurement of skill loss, output quality or time saved, so the central question of whether delegation costs people ability stays a claim on both sides.
Software projects are already writing policy on that question. QGIS merged a closed-door rule on AI-written contributions, and Anthropic moved its monorepo off Git as machine-driven merges outgrew the old tool.
One fact separates the sources. Huahai dates Hickey's essay October 11, while the LinkedIn page and the gist mirror both show October 12. Hickey's own reply on the toolness point appears only in the Hacker News thread, quoted by a commenter, so it is not independently confirmed.
Sources
More in A.I.
- 01Torvalds Posts a Kernel Patch Series Written With Claude Code, and Sources Disagree on Its Size and SpeedupLWN counts ten patches and an order-of-magnitude gain, while It's FOSS counts eight patches and a threefold gain on one laptop.
- 02Parsimony Ranks Coding Agents by How Much Code Their SWE-bench Fixes Add, and Its Own Page Warns Smaller Is Not BetterThe MIT-licensed benchmark counts coding units in published patches without rerunning a single test, and reports that code volume tracks solve rate.
- 03OpenBMB's MiniCPM-V 4.7 Went Up With No Model Card on Oct. 6, and Now Reads Apache-2.0 With No Benchmark NumbersThe 35-billion-parameter mixture-of-experts vision model is 27 times larger than its predecessor, and its card claims a win over Qwen3.6 35B-A3B without a score.
- 04Design Creativity Bench Scores GPT-6 Astra 65.5 Against Humans' 92.3, and the Paper Never Names Its JudgeA post using Simon Willison's pelican SVGs to argue that models converge on one image cites a leaderboard whose method the arXiv page does not describe.