Translation Is A Workflow

Good translation is not a prompt.

I learned that the hard way.

CADE supports a Ukrainian training audience. They needed native Ukrainian support documents, not English with a translation bolted on at the end.

My first instinct was the obvious one. Paste the English. Ask the model for Ukrainian. Done.

It was not done.

A training document is not just words. It carries doctrine, intent, tone, and terms that have to land correctly for the people using it. A translation can be technically correct and still wrong for the purpose.

That gap matters most exactly when the stakes are highest: when someone is supposed to learn from the document.

So I stopped treating translation as a single step.

I talked with a Ukrainian translator trained in philology, the deep study of language and how it actually works. That conversation changed how I approached the whole thing.

Then I built a translation skill around the process.

Not one prompt. A workflow, with the steps, standards, and checks that proper translation actually needs.

It still needs human review. Always. The skill does not replace the translator.

What it does is save time, and hand the translator a far better starting point than raw machine output.

That is the pattern under all of this.

When AI output has to meet a real professional standard, the prompt is the start.

It is not the product.

The workflow is.