From translation to AI trust: Introducing XTM IQ

AI has all but solved the capacity problem in localization. Anyone in your organization is now one translate button away from publishing in forty languages, and the raw cost of producing a translation has collapsed.

But the cost didn't disappear. It moved downstream, into review, rework, and risk, where most enterprises have no way of measuring it. The question that matters now isn't how fast you can translate. It's what you can trust to ship.

I spend a lot of my time visiting and talking with our customers, and this is the shift they keep circling back to. It's reshaping what enterprises need from their language technology, and it's reshaping how we're building XTM.

Let me set out how we see the category changing, and the investment we're making with XTM IQ, the intelligence layer of our AI globalization platform.

None of this shrinks the human role

First, let me be clear about what these changes don’t mean.

There remains a distinct and irreplaceable place for expert human translation. High-stakes, culturally nuanced,regulated and premium brand content will always demand a deeply human touch, and no threshold we build will change that.

But for a huge volume of enterprise content, the human role is shifting from operating the machinery to defining the strategy, policy, and context the machinery follows. Humans set the quality bar. Humans decide which content earns their attention. Humans own the guardrails.

That's not a diminished role. It's a bigger one. And for the teams doing this work, stepping into true oversight and creative control is exactly where the next few years should take them.

The enterprises that win globally won't be the ones with the most automation. They'll be the ones who can answer for everything they ship, in every language, at any volume.

That's the standard we're building toward, and the demands of this category will keep changing. Our job is to keep adapting to meet them, and to stay a step ahead of where they're going.

Moving quality upstream and building an agentic layer

Where we are heading is a model that stops treating every word the same.

A regulatory filing and an internal FAQ don't carry the same stakes, so they shouldn't move through the same process. When the platform understands the content and the risk it carries, low-risk work flows straight through, and human attention concentrates where it actually changes the outcome.

Two shifts make it work.

The first is moving quality upstream. Instead of catching problems at the end, in review, you build quality in from the first draft. Translation grounded in your own memory, terminology, and style comes back on-brand, so there's far less to fix later. Get it right earlier and the whole pipeline costs less.

The second is changing how people interact with the system. The future isn't someone configuring a workflow inside a portal. It's someone saying what they need in plain language, from inside Slack, Teams, or wherever they already work, and the platform orchestrating the rest.

Through our Model Context Protocol (MCP) layer, that kind of agentic interaction is already live. You ask for the outcome, and the system works out the route.

What if capacity was never the real question?

I came to localization from outside the industry, what struck me first was the parallel with software engineering. This disruption isn't unique to localization. We're seeing dramatic shifts across the board, and the parallels with what's reshaping software engineering are hard to miss.

When continuous deployment made shipping code trivial, the hard questions stopped being about throughput and became questions of governance. What gets released automatically, what needs a human sign-off, and who answers when something goes wrong in production.

Localization has arrived at the same moment.

When everyone can publish in every language, capacity stops being the bottleneck. Governance and risk management become the problem. The questions that matter are about sovereignty and accountability.

• Where does your data live?
• Which rules apply to which content?
• Whose models process your most sensitive material?
• And when something ships, who answers for it?

These are the decisions most AI translation tools were never built to make. A standalone MT feature can produce words. But it can't tell you whether those words meet your bar, route them accordingly, or leave an audit trail behind.

This issue scales with your volume of content. Our customers translate billions of words a year, much of it in regulated industries like life sciences, financial services, and legal, where any word may have to be audited later. At that scale, the gap between producing content and governing it is where the whole thing is won or lost. It's the gap we've spent the last two years closing.

Our bet: use human judgment better, not less

Much of the industry is racing toward AI that doesn't need human review at all. We've taken the opposite bet: We're building a system that uses human judgment so precisely that you need less of it each time. I think that's the more honest path to autonomy, because it earns trust segment by segment instead of asking you to grant it upfront.

That's the thinking behind XTM IQ, the AI intelligence layer that runs across our entire platform: XTM Cloud, Transifex, and beyond. It's the layer that makes every tool smarter, built as a direct response to how the category's demands have changed. Each shift we see in the market maps to a capability we've built:

1. Generic AI output creates more review work, not less, so translation starts from your context. XTM IQ uses your translation memory, terminology, and style rules, and checks itself across more than one LLM model before handing over the translation. Get the first draft on-brand and there's far less to clean up later.
2. Quality judgments can't stay subjective at scale, so every output gets an objective score with the reason attached. XTM IQ scores every translation against the rules that you set. And if something isn't perfect, XTM IQ tells you what's wrong and how to fix it. Nobody should be rereading a whole document to find the one thing that slipped.
3. Routine corrections are swallowing expert time, so they're handled before review. XTM IQ fixes terminology, tone, and formatting automatically, wherever the text came from, so your linguists spend their judgment on the calls that genuinely need a human.
4. Not every word carries the same risk, so risk decides the route. XTM IQ sends content above your quality threshold straight to publish, with its score as the audit trail. Only content below the threshold reaches a reviewer, already flagged and scored. The key thing here is that you set the threshold by content type.
5. Work shouldn't live inside a portal, so the platform comes to your team. Through our Model Context Protocol (MCP) layer, someone can ask for a translation in plain language from Claude Desktop or Cursor, and XTM orchestrates the routing, quality, and review in the background. This is where agentic localization gets real. Your team describes the outcome it wants, and the platform coordinates the work to deliver it.
6. Enterprises won't surrender their data to adopt AI, so they don't have to. You shouldn't have to hand over your data to get the benefit of AI. So bring your own key, bring your own model. Everything can run through your own contracts, inside your own compliance perimeter, and your content is never used to train anyone's model. Adopting AI and giving up control were never the same decision, whatever anyone tells you.

And because every approved translation syncs back to enrich the context, the system compounds. Review shifts from a recurring cost line to a shrinking one, quarter over quarter. That compounding is what separates a governed platform from a point solution, and it's why we describe XTM as an open, composable AI globalization platform rather than a translation tool.

The journey we're all on

I keep coming back to the fact that nobody has this fully figured out, us included. The pace of change is relentless, and every enterprise I sit with is working through the same questions in real time. We're all on this journey together.

Our job at XTM is the one it's always been: keep leading the industry we helped build, and put this technology into our customers' hands in a way they can actually trust. That legacy of innovation is the bar we hold ourselves to.

The TMS itself is becoming something new. Less a system you log into to manage translation, more a harness for AI across the entire globalization lifecycle. A layer that governs the models, orchestrates the work, and keeps a human accountable for what ships.

I'll leave you with one moment I'm proud of. At LocWorld55 in Dublin this year, XTM won the Process Innovation Challenge, the industry's live vote on the most promising new idea in localization.

Having the room decide where this should go next, and pointing at us, is about the best signal we could ask for that we're building the right thing. We intend to keep earning it.

See the AI layer behind smarter localization

XTM IQ translates, scores, corrects, and routes your content automatically, with your brand and business context built into every decision.

Explore XTM IQ: https://xtm.ai/xtm-iq