The Future of Language Tech is Platformized, Not Tool-Based
Tech, localization, and global strategy - decoded.
I honestly had so much fun writing this article, I was grinning ear to ear the whole time. WHAT A COOL THING THAT I GET TO TALK ABOUT, and share with you, and be a nerd about! What is life?! And the whole time I was writing it, I had this image of compounding layers in my head. So I decided that this article should feature art that reminds me of geological strata. I hope you enjoy!
Matter of movement 2/3, 2015, by Pernille Snedker
Ok, let’s get into the good stuff, I’m going to say something that will sound obvious in a few years (and I’m sure many of you already know):
Language will be a core infrastructure layer.
Not a service you buy. Not a tool you implement. Not a vendor you negotiate with every quarter. Nope. No more of that.
What I’m talking about here, is a core foundational capability, built into the company’s infrastructure, just like authentication, or observability, or CI/CD, that sits underneath everything the company builds.
We’re Still Thinking About Language Wrong
Right now, when companies talk about “scaling internationally,” the conversation immediately jumps to implementation questions like these:
Which translation vendor should we use?
Should we build our own MT workflow?
How do we integrate this LLM into our product flow?
Can we just add GPT-4 calls to our codebase?
These questions aren’t wrong. But in my experience of what’s happening right now (and where I see us going), these questions are incomplete.
It’s like asking “which CMS should we use?” without asking “how does content need to flow through our organization?” or “which database should we use?” without first asking “what does our data infrastructure need to look like?”
The crux of the issue is actually about (drumroll) architecture!
If you’ve been following along with me (first of all thank you!), then you know that I’ve talked a lot about architecture in the last few months, and this is because it’s truly where I see the industry going. Embedded architecture that is core and foundational. I’m going to walk you through my logic step by step to show you how I got here, and what I mean.
Strata-47B, by Mikael Hallstrøm (© 2025 Mikael Hallstrøm)
From Translation Service to Control Layer
Here’s what I see happening beneath the surface of the “AI translation” hype:
Multilingual AI, and localization ops in general, is evolving from a translation service into a control layer.
A control layer that orchestrates engines, manages prompts, handles routing logic, and enforces quality standards across every product, every market, every modality.
Think about what happened with customer support. Fifteen years ago, every team had their own ticketing setup. Email folders, spreadsheets, customized responses in spreadsheets, tracking systems on spreadheets, suuuuper fragile, and each set up was slightly different.
And now it’s so easy, you plug into Zendesk or Salesforce or whatever, and you get centralized ticket routing, automated workflows, unified customer history, and it even already comes with consistent SLA tracking. Of course, different teams can still customize their response templates and escalation rules, but the foundation is very unified.
A more technical example (and one you can use with your engineering teams) is identity management. Not that long year ago, every team built their own login systems. Password hashing, session management, password resets, it was all custom, all fragile, and of course all slightly different.
Then companies realized: identity isn’t a feature. It’s infrastructure.
Now? You plug into Okta or Auth0 or whatever, and you get centralized identity management with SSO, MFA, audit logs, and role-based access control. Different teams can still customize their auth flows, but the foundation is unified.
And that’s where language is heading.
What Platformized Language Actually Looks Like
When I talk about language as a core infrastructure layer, I’m talking about something fundamentally different from “we have a TMS” or “we use this translation API.”
I’m talking about a unified linguistic intelligence layer that provides things like:
Standardized prompting scaffolds that every team has access to and can customize. Right now you may have product, marketing, and support all writing their own system prompts and discovering the same pitfalls separately, and then sending them to localization once they’ve selected the best output in English. (Sigh) If language were a core infrastructure layer, you would get centrally maintained templates that encode institutional + linguistic knowledge. Teams could still adapt them but they’re adapting from a strong foundation that was laid by the experts, by people with linguistic expertise, and they’re not starting from scratch. So stakeholders don’t have to wait for weeks for language teams to build custom solutions, and language experts don’t get bypassed.
Here’s another example: Unified quality scoring that works the same way whether you’re translating UI strings, marketing copy, support docs, or generated AI responses. Company-wide, linguistically driven and defined definitions of “acceptable quality” that adjusts by content type and language pair but stays consistent and automated in its methodology.
And let us not forget, routing logic that actually makes sense. High-stakes legal content goes through one pipeline. Marketing campaigns go through another. Auto-generated tooltips go through a third. Not because someone manually decided each time, but because the core infrastructure language layer knows the rules and enforces them automatically.
This isn’t a vendor pitch (but maybe it should be? Who wants to build a company with me?). NO, this isn’t a pitch dangit! This is what I’m actually seeing, and I want to share it with you, because I believe that the faster our industry grasps this, the faster we will be able to move strategically.
Bush Leaves & Seeds, 2023, By Rita Pula Loy
Why This Matters Now
AI-native companies are about to hit a wall. Being AI native isn’t going to be enough, let me tell you why.
They’re shipping multilingual features faster than ever, because LLMs make it trivially easy to add language support to anything. We are about to see a complete multilingual EXPLOSION of every feature in every product. But they’re doing it in super fragmented ways.
One team is calling OpenAI directly. Another team built a wrapper around Google Translate. A third team is using Claude with a custom prompt they found on Reddit. Marketing is paying an agency. Support is using a completely different vendor.
Six months from now, when they try to evaluate quality across languages, or enforce brand consistency, or understand what’s actually being spent on translation (or on tokens!), or ensure regulatory compliance... they’re going to realize they built a house of cards.
Because you can’t scale, much less maintain, fragmented linguistic intelligence.
Not when you’re launching in twelve new markets next quarter. Not when you’re generating millions of AI responses in forty languages. Not when your content team is prompting lines of content in fifty languages. Not when regulators start asking questions about what your AI actually said to users in Germany or Japan or Brazil.
The companies that figure this out first…the ones that treat language as infrastructure rather than tooling…THEY are going to move exponentially faster than everyone else.
The Shift From “Implement” to “Plug Into”
So I hope I am pulling you along with me, and if I am, then right now you are asking me what does this look like practically. Soooo, this is what I’m envisioning:
Before: “We need to implement translation for this feature. Let’s evaluate vendors, build an integration, set up a workflow, figure out quality checks...”
After: “We need to add multilingual support. Let me call the language platform API and specify content type, quality tier, and target markets.”
That’s it! And you guys, that is infrastructure. It’s a core capability, it’s completely centralized and accessible.
With core infra like this, the platform handles everything else: routing to the right engine (or to humans) based on cost and quality requirements, applying the appropriate prompts and terminology, running automated quality checks, kicking the segments that don’t meet quality metrics to human reviewers, it logs everything for compliance, and yes, it’s also feeding results back into the training loop.
Different teams can still have different needs. Product can prioritize speed. Marketing can insist on higher quality thresholds. Legal can require human review. But they’re all working through the same foundational system, just with different configuration.
Centralized infrastructure. Modular configuration. And I want localization to own it, because we are the logical owners of this infrastructure!
Does this set up sound familiar? It should. Because it’s how every other mature platform works. This is where we need to be too!
Flow Plans, 2012, Quilt by Loretta Pettway Bennett
What This Enables (That’s Actually Impossible Today)
When language becomes a core infrastructure layer, you can do things that are basically impossible with fragmented tooling:
You can launch in five new markets so much faster (maybe even in weeks) because the entire linguistic infrastructure already exists.
You can A/B test different translation approaches at scale because you have unified measurement and consistent quality scoring.
You can guarantee brand coherency across every customer touchpoint because terminology and tone are managed centrally, not reinvented by each team.
You can prove regulatory compliance because every linguistic decision is logged, auditable, and traceable back to specific policies.
You can actually calculate the ROI of language investment because costs and quality are measured consistently instead of being buried in sixteen different invoices and spreadsheets.
You can improve systematically because training data flows back into a unified loop instead of getting trapped in vendor silos.
This isn’t incremental improvement. This is a completely different operating model.
The Uncomfortable Truth
Most companies aren’t ready for this conversation.
They’re still having the “which model should we use” debate. They’re still thinking about translation as a procurement decision rather than an infrastructure decision.
And I honestly get it, because I know from experience that infrastructure is harder to build than integrations. TRUST ME I KNOW. And when you build the infrastructure, then you have to get teams to migrate on to it, it’s not easy. Platforms require long-term thinking. Centralized systems need buy-in from every team.
But here’s the thing: you’re going to build this eventually anyway. Or someone else will, and they’ll try to sell it to you.
You’re either going to build it deliberately, with architectural rigor and stakeholder alignment, or you’re going to cobble it together reactively after the fragmented approach collapses under its own weight.
The companies that treat language as infrastructure now are going to have a two-year head start on everyone else!
Section of Liza Lou’s Desire Lines, 2019, Woven glass beads and thread on canvas
Language Is Becoming Essential Infrastructure
Authentication. Observability. Data warehousing. CI/CD. API gateways.
Nobody argues that these should be fragmented across teams anymore. Nobody suggests that each product should build its own auth system or monitoring solution.
The tech world learned those lessons. Usually the hard way.
Language is next.
Not because someone decided it should be. But because the physics of scaling AI-native products in global markets demands it.
You can’t manually manage linguistic complexity when you’re shipping features in thirty languages every month. You can’t maintain quality through heroic individual effort when you’re generating millions of AI responses daily. You can’t stay compliant through institutional memory when your team doubles every year.
You need platform infrastructure. You need centralized linguistic intelligence with modular configuration. You need language to be something you plug into, not something you implement over and over again.
I might have enjoyed reading this as much as you enjoyed writing it! It feels close to many discussions we have had at Trendyol. And the artwork is gorgeous! (I have a soft spot for rocks). While reading, I was thinking about the mindset shift teams undergo as they move from tool ownership to shared context ownership. These topics are very close to how I've been thinking about our discipline in the last couple of years. Thank you for sharing :)