The word is a joke that turned out to be useful, which is the best kind.
Technomancy is the domain covering computing, electronics, engineering, machine learning, security and making. It is named the way it is because the alternative — technology — has been worn smooth by marketing until it means roughly nothing, and because the older word carries the right implication: that these are systems which do exactly what you tell them, rather than what you meant, and will hold you to the difference.
Anyone who has spent an evening staring at a configuration file that is correct in every respect except one will recognise the genre.
The literacy gap
Something has changed in the last two decades and it is worth naming without panic. The devices around us have become both more capable and markedly less legible. A radio could be opened. A phone cannot, practically speaking, and would not tell you much if it could. The systems that route your money, rank your search results and decide whether your insurance claim is unusual are not merely closed; for most people they are not even conceptualised.
This is not a call to return to soldering irons, though soldering irons remain excellent. It is an argument that the gap between using a system and having any model of it at all has widened to the point where it affects ordinary judgement.
What the practice actually looks like
Technomancy is not a qualification. In practice it tends to look like a handful of unglamorous habits:
- Opening the thing. Physically, or by reading the source, or by watching what it sends over the network.
- Building a small, wrong model of how it works, then finding out where the model breaks.
- Being able to say what a system is optimising for, and who chose that.
- Knowing enough about failure to be appropriately, rather than generally, nervous.
The last one matters more than it sounds. Undirected anxiety about technology is common and does very little. Knowing specifically which part of a system you would not trust with your data is worth considerably more than a general sense that something is wrong.
The machines that answer back
Language models complicate this domain rather than transforming it. They are, at last, systems that respond in fluent sentences — which makes it far easier to believe you understand them and far harder to notice when you do not.
The technomantic position is neither reverence nor refusal. It is the same position taken toward any other instrument: find out what it does well, find out how it fails, calibrate accordingly, and never let a confident tone stand in for a checked result. That posture is not new. It is the one every discipline eventually adopts toward its own most useful and most misleading tools.