---
title: 'In the News: October 14, 2026 (Morning)'
description: 'Rich Hickey argues that coding agents are automation, not tools, and that handing work to them costs the user the learning that wielding a tool brings.'
canonical_url: 'https://darkfactory.dev/news/2026-10-14-morning'
markdown_url: 'https://darkfactory.dev/news/2026-10-14-morning.md'
collection: news
date_published: '2026-10-14T07:30:00-04:00'
date_modified: '2026-10-14T07:30:00-04:00'
---

# In the News: October 14, 2026 (Morning)


One item this morning: Rich Hickey, the author of Clojure, published a short essay arguing that the word "tool" does not describe a coding agent, and that the difference changes what the people using one give up. It is an argument, not a study, and we report it as one.

## 1. Rich Hickey says coding agents are automation, and that abdicating the work has a cost

**['AI' is Not a Tool](https://www.linkedin.com/pulse/ai-tool-rich-hickey-kjl8e/)** · Rich Hickey, author of Clojure and architect of Datomic · LinkedIn, October 12, 2026 ([HN discussion](https://news.ycombinator.com/item?id=50060269))

Hickey starts from a definition. He prefers "a tool is something you wield," which he breaks into three parts: it is held in the hand, it transmits power, and the user continuously directs it. By that test he says LLMs fail. "LLMs are not wielded. LLMs are automation." His example is a drive-thru, which is not a tool for cooking but automation for avoiding cooking. In the same way, he writes, an LLM is not a tool with which you write code but automation with which you avoid writing code.

The practical claim sits underneath the definition. Because the user makes a request and the system acts on its own, Hickey says the user learns the habits of a particular automaton instead of facts about the task, and that what they learned can be erased by the next software update. He gives a reading example: using an LLM to summarize papers means the model read them, not you, and the reader misses the chance to learn how others argue and write. He also says tools do not "go rogue" or "escape containment." He offers no measurements, and the claims about lost learning are his own assertions.

One caution on scope. In the HN thread, a commenter (didibus) quotes a reply attributed to Hickey saying he means LLMs in "conversational systems purporting to be 'tools' for doing homework, research, coding, writing etc." We have not seen that reply at its source. On the morning of October 14, about a day after posting, the thread showed 304 points and 181 comments. In the top comments we read, objections included that a multi-purpose tool can be a coding tool in some uses and not in others, and that the hand-tool to power-tool comparison does not map onto AI.

**Why it matters:** The essay does not tell you what to change in a harness, and it has no data behind it. What it gives is a stricter test for any workflow that hands whole tasks to an agent: ask where the person still steers and where they only wait for a result. That question is the useful part, and it is a prompt for judgment, not a finding.