For decades, the translation industry has treated reuse as efficiency. Never translate the same sentence twice. Never rethink what's already been decided. That logic built the translation-memory system — and the vendor model that managed it. But that was a 20th-century mindset: a production-line view of language. If text is a collection of interchangeable parts, someone has to supervise the assembly. That's what vendors were for.
AI broke that logic — not by accident, but by design. It doesn't see sentences as parts. It sees tone, intent, and context as a single continuous field.
Why "AI-powered CAT tools" rarely change anything
Cut that field into tiny sentence-level segments, and even the smartest model turns back into a phrase-replacement machine. That's why "AI-powered CAT tools" rarely change anything fundamental. You're still forcing intelligence into a container built for repetition.
If you force AI to work sentence by sentence, you end up with the same MTPE treadmill — only faster, not better. The output may come instantly, but the human operator still has to touch every segment, fix the same structural issues, and make the same judgment calls. A person who can edit 4,000 MT words a day doesn't suddenly double their speed just because the source came from an AI model. The bottleneck is not the machine. It's the interface.
When reuse dies, so does the vendor
The next generation of tools may shift from sentence-level to paragraph-level segmentation, allowing AI to handle coherence across a full block of meaning. But the moment you do that, reuse — the foundation of vendor logic — dies. And when reuse dies, so does the reason vendors exist.
Their traditional value lay in managing translation memories, distributing tasks, and enforcing consistency through process. But if AI can handle memory, style, and consistency in context, none of those functions require a human middle layer anymore.
From suppliers to shapers
Consistency will shift from TM to voice. Quality will shift from rule compliance to tone alignment. Translators will evolve into localization writers — not suppliers but shapers of expression. AI can learn the writer's style and replicate it across projects. What used to be "management" now becomes collaboration between human and model.
Vendors were built to sell reuse. When reuse loses meaning, they lose their product.
Translation as authorship
Translation in the AI era isn't about reuse — it's about recomposition. Not patching sentences together, but rebuilding meaning from the ground up. It's no longer manufacturing. It's authorship. And authorship doesn't scale through management layers.
AI won't replace translators first. It will replace the systems that kept translators interchangeable. Once reuse is gone, what remains is direct collaboration — between client and creator, between human judgment and machine insight.
That's where translation finally becomes what it was meant to be: not a process, but a craft. And if craft still matters, it's because words still move people.