Every few years something promises to "solve" translation. This time the tools are genuinely impressive — and genuinely changing how the work gets done. But the headline ("AI replaces translators") keeps missing what's actually happening on the ground, which is more interesting and more useful.

What AI is genuinely good at now

  • Stronger first drafts. LLM-based engines produce more fluent, context-aware drafts than the previous generation, especially for general content.
  • Following instructions. You can now feed an engine a glossary, a tone guide and surrounding context, and it largely respects them — something older MT couldn't do.
  • Assisting QA. AI can flag likely errors, inconsistencies and terminology drift for a human to confirm — a faster first pass over huge volumes.
  • Filling in context. Tools increasingly auto-suggest screenshots, descriptions and references that used to be manual.

What it still gets wrong

Confidently. That's the catch. AI output reads fluently even when it's subtly off — a mistranslated negation, a wrong honorific level in Japanese, a brand term quietly "improved", a cultural reference that doesn't land. The errors are fewer than before but harder to spot, precisely because the surface is so smooth.

The danger of great AND translation isn't that it's wrong. It's that it's wrong in a way that reads completely right.

The model that's actually winning

Not "AI instead of humans". Not "humans ignore AI". The teams getting real results in 2026 run a human-in-the-loop flow: AI drafts and flags, a qualified linguist decides. The human's job shifts from typing every word to judging, fixing and owning the result — which, for high-stakes or culturally loaded content, is exactly where human judgment is worth most.

Where you should and shouldn't lean on it

  • Lean in: high-volume support content, internal docs, first drafts you'll review, QA triage at scale.
  • Stay cautious: marketing and brand voice, legal/medical, anything with heavy cultural nuance or Asian formality systems — here AI assists, but a human owns it.

The short version

AI has made the draft better and the QA faster — a real productivity shift, used well. It hasn't removed the need for someone who can tell when a fluent sentence is quietly wrong. The smart move in 2026 isn't choosing AI or humans; it's wiring them together and keeping human judgment exactly where it changes the reader's trust.