For several years I led a cybersafety team of more than twenty agents working across seven languages: Romanian, Croatian, Hindi, Hebrew, Persian, Swahili and Czech. I speak none of them.
That sounds like a problem, and for the first months it was. It also turned out to be the thing that forced the team to work properly, because it made one question unavoidable: what has to be identical across all seven, and what has to be local?
What has to be identical: the decision
Same behaviour, same outcome, whatever language it arrives in. If a report leads to a warning in Czech and a ban in Hindi, you do not have a policy — you have seven policies wearing the same name, and eventually someone outside the company will notice before you do.
What held that together was calibration. A fixed set of anonymised cases, everyone deciding independently, then comparing. Where people diverged, the interesting finding was almost never that an agent was wrong. It was that the policy text could be read two ways. Disagreement in calibration is the cheapest signal you will ever get about where your own wording is ambiguous, and it arrives before it costs you anything.
What has to be local: nearly everything else
Tone, examples, what registers as an insult, how a threat is actually phrased by someone who means it. A sentence that is ordinary teasing in one language is a credible threat in another. No policy document written in English is going to tell you that. Only the person who speaks it will — and only if you asked, and only if they believed the answer would change something.
The part that took me longest
I could not verify the work by reading it. So the job changed shape. Instead of checking output, I had to build the conditions in which the output was reliable: criteria clear enough to apply without me, a fast route to ask, and a team habit of surfacing a doubtful case rather than guessing at it quietly.
Managing work you cannot personally inspect is uncomfortable at first. Then it becomes the only kind of management that scales. The team was not effective because everyone was equally capable. It was effective because each person was the authority on something I was not.