The end of the TMS era

Take one sentence from your instruction manual: “Press the emergency stop before you open the cover.” Your company publishes it in more languages than anyone on your team can read. For years, the sentence came from a translation memory, a translator checked it, and often a reviewer read it again.

Now AI translates it in less than a second. How your company checks that translation decides two things. Can you still trust what you publish? And what proof do you have if someone opens that cover without pressing the emergency stop?

In the TMS era, quality was a promise

In the TMS era, your team exported the manual as a file. You sent the file to an agency or loaded it into a translation management system (TMS). A translator read every line, often a reviewer read it again, and the file came back.

You got an invoice for the translation, but you could not see the check. You trusted that someone had translated “emergency stop” correctly.

Your agency was a black box. You could not see who translated your file. You could not see which terms applied to your texts, whether translators used them consistently, or which system held them.

Ask yourself who holds your terminology and translation memory today: your company or your agency?

This process cost time and money. You paid per word for every line. You waited for the whole file, whether it held a safety warning or an internal note.

AI took over translation. The promise stayed.

AI changes this process from two sides.

First, your agency noticed. It most likely translates your content with AI now and has a translator check the result. So a translator now proofreads AI output, whether you asked for this or not.

You still cannot see the check. Is the new process faster and more cost-effective for you? Probably, but only marginally.

Second, your own teams now translate with AI tools.

Back to the safety warning. Your documentation team translates “emergency stop” into German with AI and gets “Not-Halt”. Your service team pastes the same warning into another AI tool and gets “Not-Aus”.

Both versions read well. But German safety standards define them as two different functions. Not-Halt stops a dangerous movement, and Not-Aus switches off the electrical power supply.

One warning now gives two different instructions, and nobody compares them.

So do you now trust AI instead of the agency? Or do you stay with your agency as before, while your teams keep using AI tools?

At TextUnited, we think that AI ended the era of the TMS and the agency. Your company will still need translations. But they must be trustworthy, fast and cost-effective, and they must work with your company’s AI.

Translations also produce something that the old era overlooked: multilingual data. If data is the new gold, multilingual data is the new platinum. Your company must control this data, and the data must work for your business, not stay buried on your agency’s hard drives.

Two answers that do not work

The first answer is an AI-enabled agency with its TMS. It is the old approach with a new label, with few benefits for you and all the old drawbacks. The biggest drawback is that your company does not control the language data that its AI needs.

The second answer is to trust the newest LLM. You only need a large enough context window and enough tokens in your account. But nothing checks the result against the source and your approved terms.

Nothing sends the safety warning to a translator, and nothing leaves a record.

The LLM translates. It does not take responsibility.

Neither answer fits a company whose teams write and translate with AI.

What comes next: a language layer

A language layer replaces the TMS. It connects to the places where your people write, including their AI tools.

An enterprise language layer holds your language data: approved terms, style guides and translation memory. It applies this data to every text that passes through it.

For every text, it matches the check to the risk level that your people assign. It records every decision. Your language data belongs to your company, not to your provider.

Companies have seen this before, with data and with security. When everyone could create and share content, the rules and the records moved to where people work. Language is next.

Six principles for the language you publish

What this does not mean

This does not mean the end of translators. Translators spend their time on the texts where an error costs the most, and on the language data that every later translation uses.

This does not mean that you can publish AI translations without a check. Every translation needs a check against its source. Some content, like the safety warning, will always need a full check by a person.

This does not mean a guarantee. A language layer gives your company control and proof of its process. The quality still depends on your language data, and an old error in that data repeats until someone fixes it.

Where to start

Start with one type of content, next to your current process. Keep your TMS or your agency for everything else. After the first jobs, compare terminology consistency, rework and time to publish.

At TextUnited, we build an enterprise language layer. TextUnited matches the method of every translation to the risk level your team assigns, from AI translation to a translator and a reviewer. It enforces your approved terms and style guides in every job.

TextUnited also records the quality level of each translation. Your team can work in our web app or connect TextUnited through the API or an n8n automation. Your people can also use it in AI assistants through our MCP server.