Writing has always depended on materials and equipment. Clay, parchment, paper, ink, keys, and screens each make some operations easier and others harder. Their effects become visible in the work: how a sentence is revised, how a text travels, and who can obtain a copy.

Early Mesopotamian writing developed through records used in administration and an expanding capacity to represent language.[^1] Its history does not begin with the modern idea of an author sitting alone to compose a book. A system made for particular needs acquired further uses as people developed it.

A medieval manuscript required the preparation of a writing surface and the labour of scribes, with additional work for decoration and binding.[^2] A surviving volume can conceal that division of labour. We see a book; its production involved materials, specialised tasks, and considerable time. Those conditions affected how many copies could exist and how they were made.

Printing changed the work of reproduction. The Chinese Diamond Sutra scroll dated 868 is the earliest dated complete printed book known to survive.[^3] The Gutenberg Bible, completed around 1455, belongs to a later development in European printing with movable metal type.[^4] These histories should be kept distinct. A European milestone cannot stand for the invention of printing everywhere.

The typewriter altered composition at the desk. The Sholes and Glidden machine entered the market in 1874.[^5] Its operator could produce regular letterforms without setting an entire text for a press. Revision still involved practical costs: crossing out, retyping, replacing a page. A word processor later made moving and replacing passages much easier.

Ease of revision changes how I work. I can write a provisional sentence, move it elsewhere, and retain earlier versions without copying an entire page by hand. I can also rearrange words endlessly while postponing a decision about the argument. The same facility can support useful revision or prolonged avoidance.

External representations sometimes change the operations available to a thinker. Kirsh’s account helps explain why a visible arrangement can support work that would be difficult to perform entirely in the head.[^6] Spreading notes across a desk lets me compare them together. Rearranging them can expose a relationship that was hard to notice when the notes were encountered separately.

Early proposals for computing took these possibilities seriously. J. C. R. Licklider’s Man-Computer Symbiosis imagined cooperation between people and computers, with complementary roles in intellectual work.[^7] Douglas Engelbart’s account of augmenting human intellect treated equipment alongside language, methods, and training.[^8] Improving the device was only part of improving the working arrangement.

That broader view matters when discussing AI. A system that generates a paragraph can influence the argument before I have decided what I want to say. Its contribution differs from storing a sentence or changing its typeface. I have to examine suggestions, decide whether their reasoning is sound, and determine how they fit the work.

Lev Tankelevitch and colleagues describe metacognitive demands in generative AI use, including the work of setting goals, evaluating outputs, and regulating reliance.[^9] Their framework gives names to decisions a user must make. A good result does not follow merely from receiving a fluent response.

Different tasks justify different uses. I may ask for an index of material I have already written, an explanation of an unfamiliar method, or possible objections to an argument. Each request needs an appropriate check. An index should match the material. An explanation needs support. An objection should actually address my claim. Treating every response as the same kind of assistance makes those checks harder to perform.

The history of tools supplies useful comparisons, but it cannot decide the value of a new tool in advance. Printing, typewriting, word processing, and text generation alter different parts of the work. We learn more by examining those parts than by declaring that every change must repeat the last one.