A reader asked Fabrizio Ferri Benedetti what skills to prioritize in tech writing now. Is it AI? Maybe coding? His answer was neither: go back to the humanities, and learn a language or two — the kind you don’t compile.

The argument is that AI has commoditized the shallow half of knowledge work, retrieving facts and assembling plausible prose on demand, which leaves the deeper half as the differentiator.

  • Study philosophy of any kind, literature, linguistics (computational is fine), math, history, anthropology — and self-study counts.
  • The skill being trained is going deep, holding stances, and conveying and defending them, not collecting credentials.
  • The useful outputs are concrete: reading a proposal closely enough to detect bullshit, building arguments that survive disagreement, separating evidence from inference, noticing when the wheel is being reinvented.
  • Knowing Java boilerplate, Bash tricks, Markdown or DITA is not what makes a professional great. Learning fast, thinking hard, and picturing whole systems — including their hairy human factors — is.

He pushes back on the ivory-tower objection. Philosophy of mind and philosophy of science, computational linguistics and cognitive science supplied the conceptual foundations of the AI systems now being built. Logic and ontology run through database design and information architecture. UX leans on the arts and anthropology. Projects like NOPE turn clinical psychology into standards for safer AI behavior. AI labs themselves are hiring philosophers.

What he is worried about is knowledge work collapsing into “a polite exchange of statements between AI agents, mediated by that patina of human ritual that we call professionalism.” His term for the humans left in that loop is meat proxies.

He is learning Mandarin, not to work in it, but because trying to think in a radically different language drags culture and worldview along with it. The closing image updates his oldest piece of advice: knowing your way around a library becomes knowing how to read the map of the world and navigate it with whatever tools you have. The rest are implementation details.