Our Head of Product, Andreas Ljungström, recently sat down with us to talk about where the industry is heading: what’s happening to the TMS, disruptions we need to watch out, what AI in localization actually means, and lots more. Keep reading to hear Andreas’ expert take.
Honestly? I don't think we'll be calling it a TMS for much longer.
Frankly, most of what gets called "AI localization" today is just an MT engine bolted onto a workflow that hasn't changed in fifteen years. That's not transformation, that's decoration.
Right now people think of it as software you log into to set up a project, push some files through, wait for them to come back. That's disappearing. What's replacing it is more like an orchestration layer that just sits inside your content pipeline and runs on its own, with the right governance built in, so content moves from source to global delivery without someone babysitting it.
That's why we're pushing XTM Cloud to be headless wherever it makes sense.
And when people do open up the interface, they're not going to be doing project management anymore. They'll be checking AI output and steering it, which is a pretty different job than the one localization PMs have today.
That means moving beyond localization as a downstream service and embedding it directly into how organizations create, adapt, and deliver content globally.
AI gives us speed and scale, but the real focus is control and confidence, helping teams move faster without sacrificing quality, brand integrity, or governance as content volumes and complexity keep growing.
What I mean by it is pretty specific. Every bit of content that comes in gets looked at in real time, what kind of content is this, how much risk is there if we get it wrong, and the platform decides where it goes based on that. Routine, low-risk stuff just flows through AI. Anything higher stakes gets flagged for a person.
Nobody's manually sorting that anymore. And that's really the point of it, your best people stop spending their week on stuff that doesn't need their brain and only get pulled in when it actually matters.
We talk about this as the "intelligence layer" doing the sorting so humans can do the judgment.
Our intelligence layer is called XTM IQ. And it's already proving out at scale: one enterprise customer saw 61.7% of their entire AI translation output, 9.5 million words, go live with zero edits.
Easy one. Shadow AI.
You're not going to solve that by telling people not to do it. People take the path of least resistance, always have. The fix is making the approved path the fast one.
Give people instant, self-service translation, but route it through something that quietly enforces your security and brand rules in the background. Once the sanctioned option is also the easy option, shadow AI mostly takes care of itself.
XTM recently won the Process Innovation Challenge at LocWorld55 in Dublin with XTM Go, an innovation that addresses exactly this challenge for enterprise businesses. You can read more about it here.
Manual project scoping, that's the first one. All the admin around setting up a project, deciding what goes where, chasing people for handoffs. That's going to be handled automatically based on risk and content type, and I don't think anyone's going to miss it.
I do want to say, this isn't "people disappear." It's the repetitive, low-value parts of the job disappearing. The actual hard calls, the cultural nuance, the judgment work, that stuff gets more valuable, not less.
You need something that can classify all of that by volume and by risk, at roughly the same speed the content's being created.
The volume part is the easier one to accept. There's just going to be more content, made faster, than any human review process was ever designed for. If the plan is "hire more reviewers," that doesn't scale, and honestly it was never going to.
Risk is the part that actually matters. Not every AI-generated asset is equally dangerous if it's wrong. So you want a layer that looks at each piece as it lands and decides how much scrutiny it actually needs, then applies guardrails and quality gates automatically instead of someone eyeballing it case by case.
That's a lot of what we’re building XTM Go for. It's designed for that instant, risk-aware triage, so people get self-service speed without losing the guardrails underneath. You end up handling a lot more content without needing a lot more people.
Governance. But governance as something that actually helps you move faster, not something that gets in the way. I can't say this one enough.
Start with what I'd call your brand memory, your TM, your glossaries, your style guides. That's your contextual fabric, and it's part of what keeps your AI sounding like you instead of sounding generic.
But we're pushing further than that.
The second part doesn't cost anything, it's just visibility.
Don't wait to be asked. Go raise awareness of what your localization program actually does, and put yourself forward as the champion of AI governance inside your own company. You already know how to manage quality and risk across languages at scale, and that's exactly the skill every other department is about to need.
Honestly, this is the moment to show the rest of the business what the loc team is actually worth.
Stop treating translation like a project you kick off, and start treating it like a service that's always running.
If it's a micro-service sitting in your pipeline, content gets translated the second something changes. No batching, no waiting around for someone to start a job. You connect the platform straight to your content sources and updates just trigger localization on their own.
People used to assume faster meant lower quality. That's not really true anymore. Risk-aware triage means the routine stuff can move at full speed while your actual human reviewers only touch the handful of things that genuinely need them. And having one platform as your single source of truth just makes the whole thing move faster too.
This is genuinely the part I get most excited about, because it flips the whole model around.
Right now, most teams find out about quality problems after they've already happened. Translate it, review it, spot the issues, fix them. That's reactive, and it means someone's checking everything just in case something's wrong.
Prediction works the other way.
And that threshold isn't something we set for you, you define what "good" looks like for your content, your brand, your risk appetite, and TQI scores against that.
Review stops being a blanket check on everything and becomes something targeted and driven by data. Same people, way better use of their time.
It's the platform making the routing, quality, and workflow decisions on its own, instead of a person necessarily deciding case by case, or a fixed rule set deciding it once and being left alone. But the human always has the final say. We're automating and assisting the decision-making, not replacing the person who makes it. The human stays at the center of this, full stop.
That's the actual difference from regular automation. Automation is "do this fixed thing without a human." Autonomous orchestration is a system that watches its own outcomes and gets better at the decision over time, without anyone sitting down and changing the settings.
So the goal was never hitting a target split and holding steady. It's building something that keeps getting sharper at deciding where content should go, on its own, the longer it runs.
And that's really the dividing line coming for this industry. The localization teams that matter in three years will be the ones who step up and become the AI governance layer for their entire enterprise. Everyone else gets quietly absorbed into marketing ops.