The assumption that artificial intelligence would simply eliminate human translators has proven both right and wrong in ways nobody quite predicted. What has actually happened is stranger: the profession has cleaved into two distinct occupations with opposite trajectories, one declining rapidly and one experiencing a quiet renaissance.
Commercial translation—the business of converting user manuals, legal contracts, marketing copy, and technical documentation—has undergone a transformation so complete that the job title itself has become a misnomer. Most practitioners now spend their days editing machine output rather than translating from scratch. The work has shifted from creation to quality control, from linguistic artistry to error detection. Rates have fallen accordingly, and the number of full-time positions has contracted sharply across every major market.
The post-editing economy
Language service providers began integrating neural machine translation into their workflows years ago, but the real shift came when output quality crossed a threshold that made human review optional for many use cases. Internal corporate communications, routine legal filings, software localization for minor markets—these now flow through automated pipelines with minimal human oversight. The translators who remain in this sector have become something closer to specialized proofreaders, paid by the hour to catch the errors that machines still make: false cognates, cultural inappropriateness, technical inaccuracies in specialized domains.
The economics are brutal. When a machine produces a passable first draft in seconds, the value of human labor shifts entirely to the margin—the difference between adequate and excellent. Many clients have decided adequate is sufficient. The translators who have survived in this space have done so by developing expertise in narrow technical fields where machine errors carry real consequences: pharmaceutical documentation, aerospace engineering manuals, financial regulatory filings.
The literary exception
Meanwhile, something unexpected has happened at the other end of the market. Literary translation—the rendering of novels, poetry, essays, and prestige journalism—has experienced a genuine resurgence of interest and compensation. Publishers have discovered that readers can detect machine-assisted prose, and that the backlash against it is fierce. The translator's name on a book cover has become a selling point rather than an afterthought.
This divergence makes sense once you understand what machines actually do. Neural translation systems excel at pattern matching and statistical likelihood; they produce text that sounds plausible. What they cannot do is make the thousands of interpretive decisions that literary translation requires: whether to preserve a pun or substitute a different one, how to render dialect without condescension, when to sacrifice literal accuracy for emotional truth. These choices require understanding not just two languages but two cultures, two literary traditions, two sets of reader expectations.
The result is a profession that now resembles acting more than clerical work. A small number of translators have become minor celebrities in literary circles, their interpretive choices discussed and debated. The vast middle tier of competent generalists has largely disappeared.
Our take
The translation industry offers a preview of how AI will reshape knowledge work more broadly: not through simple replacement, but through a brutal sorting mechanism that rewards either irreducible human judgment or willingness to accept commodity wages for supervisory labor. The middle—skilled work that requires training but follows predictable patterns—is precisely what machines do best. Professionals in law, accounting, medicine, and journalism would do well to study what happened to translators and ask themselves honestly which half of the split their own work resembles.
