Using AI well is a skill, and the line that matters runs between wielding it and delegating to it

A post on yyhh.org this week answered Rich Hickey directly. In his October 11, 2026 article, "'AI' is Not a Tool", Hickey argued that LLMs are automation rather than tools, because they cannot be wielded. The yyhh.org author accepts part of the case: outsourcing work can cost you opportunities to learn. The conclusion is where they part ways. Even by Hickey's own definition, the post argues, an LLM can be something a human wields, and "using it well takes skill, just as using other tools does." The disagreement sounds academic. It settles who gets credit for the work.

The judgement stays with the operator

The post starts in the workshop. Picking up a saw does not guarantee a straight cut, and holding a chisel does not make anyone a sculptor. A tool amplifies poor judgement as readily as good judgement. Its usefulness depends on the person directing it, reading the feedback, spotting mistakes and correcting them.

Those requirements carry straight over to LLMs. The author's example is a programmer who asks a model to explain unfamiliar code, challenges the explanation, inspects the source, tests a proposed change, rejects it and refines the request. That is a feedback loop, and a human with a purpose is running it. Doing it well means knowing what to ask, which evidence matters and when the answer is wrong.

The post also takes apart the split between tools and automation. A compiler does substantial work without the programmer directing each operation. A spreadsheet recalculates without anyone working through every sum by hand. People still wield both: they specify the inputs, examine the results and decide what happens next.

A colleague of the author's put it in four words: "AI is a mirror." It reflects the curiosity, judgement and habits you bring to it.

This is the War Room position, argued by a practitioner. AI does the routine work. You do the thinking. The value sits in the loop of directing, inspecting and correcting, and the loop belongs to whoever runs it.

Where the line actually runs

The post closes on its sharpest sentence: "The useful distinction is between wielding an instrument with skill and delegating work without adequate understanding or oversight."

Reading is the test case. The author agrees with Hickey that taking a summary in place of a paper gives up much of what reading teaches. Someone else can read the paper and then use an LLM to question their interpretation, explore an objection or find a passage worth a closer look. The same software sits on both desks. In the author's words, "The relevant question is what the person actually does."

The post also flags the language around these systems. Phrases such as "go rogue" and "conspire" give software human motives. The real failures are engineering problems: unexpected outputs, ignored instructions, harm once a system has access to consequential actions. Naming them plainly keeps the human in the role of the one who decides.

For an operator, this changes what counts as competence. Two people can log the same hours in the same model. One has added judgement to the work and the other has handed it off. If anyone asks, only the first can explain what they rejected and why.

Your Next Move

• Sort last week's AI use into two columns. Wielded means you set the task, inspected the output and corrected it. Delegated means you accepted it without checking. Move one recurring task from the second column to the first this week.
• Keep a log of your rejections. Each time you turn down or rework a model's output, write one line on what was wrong and how you spotted it. That record is your evidence of judgement, and it is what to bring when your employer starts weighing AI use. Our guide to showing impact when your employer grades your AI use explains how to present it before your next review.
• Read the primary source on the one document that matters most this week. Read it yourself first. Then use the model to argue with your reading, and keep the summary for the documents that matter less.