Can We Make MultiLingual Content Frictionless?

The Evolution from Localization Management to Data Orchestration

How global content is created and delivered is changing the role of localization fast. It’s no longer about managing projects; it’s about orchestrating intelligence.

In this conversation, Andrew Thomas (RWS) and Yvan Hennecart (Global Content Consultant) explore what happens when humans become the conductors of multilingual ecosystems, where AI, data, and human expertise play together to deliver meaningful global experiences.

We examine how the industry’s legacy knowledge, long treated as static assets, is becoming the foundation for a new kind of global intelligence.

Stefan Huyghe: Good afternoon from sunny Texas for a discussion I’ve been looking forward to. We’re going to talk about localization orchestration and how agentic AI is rewriting the playbook for global content. I got inspired by a recent exchange with Andrew Thomas and Yvan Hennecart on one of my LinkedIn posts, and I thought, why not bring that conversation here?

Before we dive in, Andrew and Yvan, why don’t you introduce yourselves?

Andrew Thomas: Sure. I’m VP of Marketing at RWS, but I’ve been in the localization industry my entire career. I started as a localization project manager, then worked on product management for translation management systems. Even though I’m on the marketing side now, I still feel close to the technology and process discussions that shape our industry.

Yvan Hennecart: I’m a global content consultant. Like Andrew, I’ve been in this industry for quite a while, probably too long! Recently I’ve had to revamp a lot of my expertise to understand what AI truly brings us. My goal is to approach it from a positive, practical perspective, and not just focus on what we might lose, but on what we can build.

Stefan Huyghe: I like that. I think all three of us tend to see the glass as half full when it comes to the future of localization. And one thing I really want us to unpack today is how we can go beyond the traditional translation picture into orchestrating multilingual content. So let’s start there. How do you define “orchestration” in the context of multilingual content today?

Andrew Thomas: At the end of the day, every company needs to get content into the hands of their audience in a way that allows that audience to consume and understand it. Historically, that process was human-driven and computer-assisted. Humans managed everything, every step, every file, every approval.

With AI, that dynamic has changed. Humans are still involved, but the process is becoming increasingly self-driven by technology. That’s why I think orchestrate is a better word than manage. Humans aren’t managing the end-to-end process anymore, they’re conducting it.

Think of the orchestrator as the conductor. Some instruments might still be played by humans, but many are now played by AI, algorithms, or legacy systems. The point is that they all have to come together to deliver the song. The human is there to make sure the performance works, not to play every instrument.

Yvan Hennecart: Exactly. Whether you call it management, orchestration, or caretaking, it’s all about bringing the right content to the user when they need it. We’re really talking about content generation on demand.

And that goes far beyond translation. As much as translation has helped us so far, user expectations have changed dramatically. Gone are the days when one solution fit all, when it took five days to deliver ten thousand words and that was considered efficient. Today, content has different shelf lives, different audiences, different contexts. How we deliver it, whether it’s unilingual or multilingual, has to adapt.

Stefan Huyghe: That’s such an important point, that orchestration signals a change in both mindset and capability. It’s not about pushing text through pipelines anymore; it’s about designing intelligent systems that understand purpose, timing, and audience.

Legacy Knowledge The Hidden Goldmine

Stefan Huyghe: Let’s build on that. If orchestration goes beyond translation, what happens to all the experience and data the industry has accumulated over the years? How do we use what we already know to shape what comes next?

Yvan Hennecart: Well, collectively, the three of us probably represent close to eighty years of experience in translation. I was trained as a translator myself, and as much as that foundation has served us, the expectations around digital content have completely changed. We have to adapt.

Gone are the days when we said, “It’ll take us five days to do 10,000 words, and that’s just how it is.” Content today has different uses, different lifespans, and different audiences. The way we deliver it must reflect that.

And this is where our legacy data, decades of linguistic assets, becomes so valuable. All that linguistic data we’ve been caretaking for our customers is now central to their AI initiatives. It’s part of the ecosystem. It fuels automation and machine learning.

Andrew Thomas: I couldn’t agree more. There are two big sides to this. First, content today is multimodal thanks to AI. You might receive a text to translate, but your output could just as easily be a video, an audio clip, or something interactive. That’s transformation, not translation.

Second, Yvan’s point about legacy data is critical. Localization teams are sitting on a goldmine of linguistic intelligence, brand terminology, validated translation memories, tone, style, industry context, all of which can be used to fine-tune language models. Even if old translation memories don’t plug neatly into tomorrow’s tools, they’re incredibly valuable for shaping AI outputs to sound like your brand and reflect your domain expertise.

So, localization isn’t dying. In fact, it’s expanding. The number of content types, use cases, and channels is exploding. And the data we’ve accumulated over decades, once trapped in TMs and glossaries, is now what gives AI systems context and voice.

Stefan Huyghe: It’s interesting, isn’t it? For years, localization has been seen as a downstream function. But now, the very data we’ve created in that role could be what drives upstream innovation in AI.

Beyond Translation: Transformation and Accessibility

Stefan Huyghe:What I’m hearing from both of you is that localization has reached a kind of inflection point, we’re no longer just translating, we’re transforming. Andrew, you touched on this earlier. Can you expand on what you mean by transformation rather than translation?

Andrew Thomas: Absolutely. Content today isn’t limited to text anymore. Thanks to AI, it’s multimodal. You might start with a piece of written content, but your output could be an audio file, a video, or a completely restructured format.

So the job isn’t just to translate, it’s to transform content so it fits the audience, the medium, and the intent. You’re taking one input and reimagining it for many outputs: from one language to many, from one file format to many, from one use case to many. And that’s where I think orchestration really comes in.

I’d even include accessibility here. It’s a perfect example of an area that isn’t translation in the traditional sense, yet it draws heavily on the same principles, precision, clarity, cultural sensitivity. The same best practices that have shaped translation are just as critical when you’re designing accessible content for people who experience information differently.

Yvan Hennecart: That’s a great point. The expectations of users have evolved dramatically, and so has the definition of what “localized content” even means. It’s not just language anymore, it’s experience. Users want content that feels relevant in the moment and in the format that makes sense for them.

We’ve been managing linguistic assets for years, glossaries, termbases, TMs, but now those assets are part of something bigger. They feed personalization engines. They help AI generate content that’s not just linguistically accurate, but contextually right.

Stefan Huyghe: So the scope of our work has widened. Translation is one part of the equation, but transformation, across formats, accessibility needs, and channels, is where real orchestration starts to show its value.

Quality Redefined From Linguistic to Outcome-Based

Stefan Huyghe: Let’s talk about quality. We’ve all worked under frameworks that measure linguistic accuracy, but those metrics seem increasingly disconnected from the way content performs in the real world. How do we reconcile traditional quality models with AI-driven systems?

Andrew Thomas: That’s a big one, and honestly, a bit of a pet peeve of mine. Everyone in our industry talks about quality, and every LSP promises it. But in reality, linguistic quality is the wrong measure. C-level executives don’t care whether something is beautifully translated word-for-word, they care whether the content works.

At the end of the day, quality isn’t about how well a sentence reads in another language; it’s about whether the content served its purpose. Did it keep me out of legal trouble? Did it help a customer resolve an issue without calling support? That’s quality of outcome.

So we need to think in terms of use cases. If the content is regulated, your quality metric is compliance. If it’s a knowledge-base article, it’s deflection rate. Each scenario has its own definition of success. And that, in turn, determines how much human involvement, subject-matter expertise, or AI oversight is appropriate.

Stefan Huyghe: So instead of “perfect translation,” we’re really talking about functional accuracy, fitness for purpose.

Andrew Thomas: Exactly. That’s why orchestration matters. You use the right mix of humans, machines, and data for each case, not one standard workflow for everything.

Yvan Hennecart: And I’d add that “human quality” isn’t always 100 percent either. We like to assume it is, but humans make mistakes too. What matters now is matching quality expectations to purpose. In some cases, small imperfections are acceptable; in others, like regulated industries, you can’t take that risk.

And yes, quality has layers now. We need to accept that not all content deserves the same level of investment or review. The new challenge is determining what level of quality is appropriate for each type of content and user.

Stefan Huyghe: That ties beautifully back to orchestration, understanding not just what to deliver, but how much of each element is needed to achieve the right outcome.

From Linear Pipelines to Many-to-Many Networks

Stefan Huyghe: Let’s talk about workflows. Localization has always relied on linear pipelines, the familiar chain of authoring, translation, review, delivery. But today, we’re seeing more complex, interconnected ecosystems. Andrew, you once said the “single source to many targets” model is outdated. Can you expand on that?

Andrew Thomas: Sure. It’s really about how people consume content now. You can’t predict when, where, or how a user will engage with it. That alone breaks the single-source paradigm. We’re now in a many-to-many world, multiple inputs, multiple outputs, multiple contexts

Take a simple example. I’m looking for a restaurant on my desktop, I want to read reviews and check out the menu. Ten minutes later, I pull out my phone to get directions. Same user, same website, same language, but completely different intent. Then once I’m at the restaurant, I scan a QR code to view the menu, again, same content, new context.

That’s why localization needs to become orchestration. You’re no longer translating a static text. You’re ensuring that whatever version of that content appears in whatever context still fits the moment.

Yvan Hennecart: And when you add AI into that mix, content can even react to user behavior in real time. It’s about situational adaptation. We can now draw on CRM data, user preferences, and intent signals to tailor not just the message, but the modality, text, voice, or video.

Stefan Huyghe: That’s a big leap from translation as we knew it. We used to treat language as static. Now, the same sentence could change based on who’s reading it, where they are, or what they need.

Andrew Thomas: Exactly. And that’s where agentic AI starts to show promise, because it can react dynamically. But to make that possible, we need integration across the entire content lifecycle: authoring, localization, publishing, and analytics. When all that data flows together, orchestration becomes continuous instead of linear.

Stefan Huyghe: So it’s not about “projects” anymore, but ongoing intelligence, systems that learn, adapt, and respond.

Yvan Hennecart: Yes, and that’s a profound cultural shift for our industry. We’ve always thought in batches, files, projects, and deliverables. But orchestration is continuous. It’s multilingual experience management, not just translation management.

The Promise and Realities of Agentic AI

Stefan Huyghe: Let’s turn to what’s probably the hottest topic in our field right now, agentic AI. Everyone’s talking about it, but I’m curious how you both see its real implications for localization. Is this the breakthrough that finally changes everything, or are we getting ahead of ourselves?

Andrew Thomas: I’ll start by saying this: there’s a lot of hype. And honestly, it reminds me of when large language models first hit the scene. Everyone’s using the term agentic right now, but in many cases, what they really mean is simply workflow with AI.

True agentic AI, systems that can take autonomous actions, respond to changing conditions, and make decisions on the fly, we’re not there yet. It’s coming, but it’s not today.

That said, when we do reach that point, the potential is enormous. Imagine an orchestration agent that can decide, for each piece of content, which tools to use, which humans to involve, and in what sequence. It could pick the right model, the right dataset, even the right reviewers based on expertise and availability. That’s real orchestration intelligence, not replacing humans, but coordinating them far more efficiently.

Yvan Hennecart: Yes, I completely agree. And I’d add that we have to stay pragmatic. There’s a danger in promising too much too soon. When you present AI or agentic solutions to executives, the first thing they think is, “Oh, so now it’s going to be free.” But that’s not how this works. AI isn’t cheap, it’s computationally expensive, and it takes time to integrate responsibly.

That’s why I advocate for a phased approach. We start by identifying repetitive or time-consuming tasks that can be automated. Once those are stable, we can introduce orchestration agents to manage larger processes. But we’re still far from an agent that can handle everything end-to-end.

Andrew Thomas: Exactly. And there’s another interesting loop happening. We started with small language models, MT systems fine-tuned for one purpose. Then came the massive, general-purpose LLMs. Now we’re seeing the pendulum swing back toward smaller, focused models again. Cohere’s new translation model is a perfect example: a specialized LLM built for one task. So, the future won’t be one giant AI doing everything, it’ll be multiple intelligent agents doing specific things really well.

Stefan Huyghe: So in a way, orchestration becomes the glue, connecting specialized agents, human expertise, and legacy tools into something coherent.

Andrew Thomas: Yes, and the irony is that we’ll probably need a new kind of AI, an orchestration agent, to figure out which tools to use, in what order, and under which conditions. It’s not about size or power anymore. It’s about coordination.

Bringing Orchestration into the Boardroom

Stefan Huyghe: Let’s shift gears to the business side. If localization evolves into orchestration, how do we explain its value to the C-suite? How do we bring this conversation into the boardroom?

Yvan Hennecart: That’s a great question. The first thing to understand is that there’s still a lot of misconception at the executive level. When people hear “AI,” they often assume cost reduction, as if automation will make everything free. That’s a dangerous assumption.

AI isn’t a shortcut to zero cost; it just transfers where value is created. Computational power, integration, and governance all come with their own price tags. The right question isn’t “How cheap can it get?”, it’s “What new efficiency and capability does this unlock?”

That’s why I take a progressive approach. When you show executives how AI and orchestration can automate repetitive tasks first, then scale into more complex workflows, they start to see measurable value. It’s about trust and proof, not promises.

Stefan Huyghe: That’s an important distinction. You’re saying orchestration should be framed as strategic enablement, not automation for its own sake.

Andrew Thomas: If you want to speak the language of executives, tie what you do directly to revenue and outcomes. Ask them a simple question: How important is it to you to get the right message, in the right format, to the right customer, at exactly the right moment?

Every company wants that. And every company depends on content to achieve it. Localization is how that content becomes usable worldwide. So if you can connect what you’re doing to engagement, conversion, or retention, you’re not talking about cost anymore; you’re talking about growth.

One thing I’ve learned in marketing is that it all boils down to outcomes. Whether it’s increasing revenue, lowering costs, or staying compliant, everything ladders up to one of those. So if localization teams can show how orchestration helps achieve those outcomes, they earn a seat at the strategic table.

Yvan Hennecart: Yes, and to build on that, orchestration reframes our work as a driver of global experience, not a back-office process. It makes localization visible again, because it connects directly to customer experience, not just linguistic delivery. So the message to executives is simple: orchestration isn’t a cost center, it’s a growth engine.

Closing Reflections: The Human Conductor in an AI Orchestra

Stefan Huyghe: Andrew, Yvan, this has been such an insightful exchange. Before we wrap up, I’d like to end with a forward-looking question. We’ve talked about orchestration, legacy knowledge, AI, and business impact. But at its core, what does this really mean for the people working in localization today?

Andrew Thomas: For me, it’s about evolution, not replacement. Every major technological change in our industry, from CAT tools to MT to AI, has triggered hype, fear, and eventually, progress. Agentic AI is no different. It’s another powerful tool in our toolbox. The real skill lies in knowing when and how to use it.

We shouldn’t be afraid of it, but we shouldn’t expect it to solve everything either. The difference we bring as humans is judgment, understanding context, making trade-offs, deciding where automation ends and expertise begins. That discernment is what defines orchestration.

Yvan Hennecart: I think that’s where the human conductor metaphor becomes real. AI can play many of the instruments now, but it still needs direction, cohesion, and purpose. The orchestrator’s job is to make sure all the parts, AI, data, processes, people, work together toward a meaningful result.

We’ve spent decades perfecting workflows that moved content efficiently. Now we’re learning to manage intelligence. That’s a very different skillset, and it’s going to redefine what success looks like in our profession.

Stefan Huyghe: Beautifully said. It feels like we’ve come full circle, from managing translation to orchestrating intelligence. The tools have changed, the scale has exploded, but the purpose remains the same: helping humans understand each other.

Andrew, Yvan, thank you both for this thought-provoking conversation. I think we’ve just composed the outline for the next era of localization.