Years ago, William Safire wrote a weekly article in the Sunday New York Times Magazine called “On Language.” I loved it. He explored the origins of words and phrases. How they entered the language; how their meanings changed; how people used them correctly, incorrectly and sometimes differently enough that the incorrect use eventually became correct.

The column was an insightful, fun exploration of how language never stands still.

I started thinking about Safire recently while considering one of the more remarkable capabilities of AI, translation. We can now communicate almost instantly across languages. Not simply by substituting one word for another, but increasingly by translating meaning in context, tone and even some of the subtlety that once required a person fluent in both the language and the culture.

The progress is impressive and useful. It also made me wonder, do we even still need Babbel? Why spend months learning another language when an intelligent translator can travel with you and translate almost anything just in time?

Of course, something would still be lost. Learning another language is not merely learning different words for the same things. Language carries history, humor, assumptions and ways of seeing the world. To understand another language is to understand something about the people who speak it. And the language will not wait patiently for AI to master it. New words appear or old words acquire new meanings. Slang moves from one generation to another. Humor has a unique way of changingthe meaning of a phrase almost overnight. People deliberately bend language, misuse it and remake it. That was one of Safire’s enduring lessons. Language belongs to the people using it.

AI can follow those changes capably. It may recognize them faster than any individual human can. It may even accelerate them by carrying new expressions across borders and cultures.But in the end, it is still humans moving first.

Then I realized that translation is a problem I have been wrestling with quite a bit recently. Not translation between formal languages like English, Spanish and French. Translation between people who already appear to speak the same language.

AI developers and evangelists speak about models, agents, APIs, velocity and scale. They see the possibility of moving faster, eliminating old limitations and solving problems that previously seemed beyond reach. The people operating inside established legacy businesses speak a different language. They speak about exceptions, reconciliations, controls, ownership and all the peculiar things that happen between the beginning of a process and its end. Much of what they know has never been written down. It has been absorbed through years of doing the work, correcting failures and building manual routines around systems that never quite did everything the business needed them to do.

Interesting, but both groups may want the same thing. They may simply not understand one another. The people closest to the work can sound resistant when they try to explain its complexity. Developers can sound reckless when they try to move past limitations they believe technology has made obsolete. One side hears unnecessary delay. The other sees consequences that have not yet been translated.

Perhaps AI can help with that translation too. Not by deciding which side is right. And not by replacing the difficult work of understanding the business. AI can help the people who possess the operational knowledge express it in a form developers can use. It can interrogate processes, organize exceptions, identify patterns and turn decades of accumulated workarounds into requirements that can finally be examined.

It also helps ‘regular’ people learn the language of the developers. The argument does not have to be “slow down.” It can be exactly the opposite. Please move faster, and please use that speed to go deeper. Do not use AI merely to construct a more impressive front end over the same neglected foundation. Use it to understand the business deeply enough to remanufacture the systems underneath it. The true competitive advantage awaits in the deeper dive.

That may require a different kind of prompt. It is not simply asking how can we automate what these people are doing. We ask why they are doing it; what they know that the system does not; which manual steps are waste and which ones quietly prevent failure; and what would have to be true for the entire process to be rebuilt rather than merely accelerated.

The greatest AI opportunity may not be translating one language into another. It may be translating human knowledge into something from which a lasting system can finally be built.

William Safire taught his readers that words carry histories and meanings that are easily overlooked. The same is true of old business processes. Their language can appear archaic. Their practices can look inefficient. But buried within them is the history of what went wrong, what had to be corrected and what the organization learned.

Before we discard that language, we should translate it. Perhaps AI’s most important language breakthrough will not be allowing two strangers from different countries to understand one another. Perhaps it will be allowing two people in the same meeting to finally do so.