Can you give us a brief introduction about your role at Lokalise?
As product lead for the AI Translations domain at Lokalise, I mainly focus on innovation with AI-powered solutions like custom models, translation quality evaluation with scoring, smart routing between different LLMs, and what we call AI orchestration. These are the key areas I focus on at Lokalise.
How did you get into developing innovations for AI?
My background is mostly in launching disruptive products at different startups and scale-ups. I was always excited about disruption, transformative technology, and how technology can really change industries. When the GPT moment happened, when LLMs scaled and became available for broader use, I joined Lokalise because I saw a huge opportunity to apply AI in a truly transformative way in the localization space.
I really believe that with AI we can have a great, tangible, measurable impact in localization. We are already transforming the industry itself, really deeply changing how it works. That’s how I got here and why I’m here.
Can you tell us a little bit more about what you’re most excited about in AI at the moment?
Right now, I’m most excited about the shift from general-purpose LLMs to more tailored, domain-specific augmented systems that leverage context. In our industry, there’s been a shift since the GPT moment three years ago toward more tailored experiences.
Seeing this materialize—not just the promise of it, but seeing it really happen—is very exciting. At Lokalise, it’s coming to life right now with custom models, which we have in beta.
With our custom models, we’re leveraging context, especially from past translations, to deliver personalized translation for every customer. Seeing the results and impact is really exciting.
We’re delivering human-like translation quality, and seeing it work on real customers with real translations is amazing.
It feels like true disruption because we’re finally at the point where we can deliver high-quality translation at an absolute fraction of the cost, which is transformational.
What’s one trend in AI that’s overhyped and one that’s not getting enough love?
So for overhyped, I think there’s a difference in the stages of new things that come with AI.
There’s a boom, something new, hype, lots of promise, but not the actual impact yet. That’s where we are with agents or agentic workflows. It’s everywhere. Everybody’s talking about it. It’s on every conference as the main topic.
The potential is absolutely there. It’s incredible what it can do. But the true solutions aren’t there yet, especially in localization. There aren’t truly reliable autonomous decision-making solutions. The real-world impact isn’t there yet.
Anything around AI in the last three years becomes a buzzword. That’s what happens when something gets traction. There’s also a lot of buzz that comes with it. It’s still good because it means more investment in AI and in companies that really drive the revolution.
But, as a product manager, I need to distinguish if something is just buzz, versus a really impactful solution that delivers value to customers.
It’s not about doing AI because everyone does AI, or doing agents because everyone talks about agents. It’s about using the right solution for the problem that really delivers impact. For every product manager today, this is a relatively challenging environment because there’s a lot of buzz, but that’s different from real-world impact.
What do you think’s not getting enough love or attention in AI?
RAG (Retrieval-Augmented Generation). That was really hyped two years ago, where agentic workflows are today. It was the main topic at most conferences and in expert discussions. It’s not that much today. But in reality:
Retrieval Augmented Generation is the solution that actually delivers the most impact right now.
What it does is retrieve or provide additional context to AI to provide more tailored output. In our case, more personalized translation. We’re relying heavily on RAG at Lokalise, and it has amazing results.
This is something that’s not getting much attention at the moment, but in reality it’s incredibly impactful, not just for localization but overall. It can be considered ‘old school’ in terms of buzz and hype, but it’s very, very impactful.
What are some other real-world applications of RAG applications?
In customer support. For example, if a customer makes a request for help, thanks to RAG, the system can retrieve the customer’s history, context about what plan they’re on, what features they’ve used, and connect it with the knowledge base for customer support, like technical documentation.
With this information provided to the large language model, the answer is much more precise, much more targeted, much more personalized. In many cases, it’s beating human responses. That’s another typical example where RAG delivers real impact.
What do you think is the next big trend for AI-powered localization?
I see two major areas in localization where AI can drive truly transformative innovation. The first is delivering the translation itself. We use AI to translate content at really high quality, at minimal cost, instantly. This is where most of the focus is today in our industry, and for good reason. It’s the most valuable use case because human translation or human review is what costs the most money in localization.
Making this much cheaper, more scalable, and faster is super valuable. We already have the right solutions in place: RAG and using context from translation memory, glossary, style guide, screenshots, descriptions. The solutions are there. The next step is to configure it right and scale it properly so any customer can get real value. That’s connected with the right setup or configuration: custom AI profiles or custom models.
The second area is localization management itself. Managing the whole localization workflow, aligning stakeholders, managing budgets, control mechanisms, project management to a large extent. That’s an area where there aren’t really breakthroughs at the moment, but I believe there could be. Agentic workflows can be super impactful here.
I believe this will be the next big trend in maybe one or two years, but it will take time to materialize.
We’ll make it much easier for localization managers to put the whole process on autopilot so they can focus on strategic things.
The technology is there, but it’s about how we use it. That’s what AI and localization will be about in the next one to two years.
⚡Lightning round
Great, Adam. Let’s wrap up with a few rapid-fire questions:
What’s one thing that AI still can’t do?
I would say detect sarcasm and irony. It’s getting better but like the dry human humor is still challenging.
AI, friend or foe?
Definitely a friend, but a very, very powerful one that we should keep an eye on.
How do you imagine AI in daily life 10 years from now?
It’s going to be like electricity or internet, so it’s going to be invisible, but it will be everywhere. We will just not notice it anymore in the same way as we do not think about the existence of internet electricity, which it’s just there. We take it for granted.
If you could give AI a voice, what celebrity would it be?
It would be Emma Thompson. I love her accent, she’s very elegant, thoughtful, articulate.
If AI could learn one human skill overnight, what would it be?
It should be empathy, but like the real one, not just the mimicked one.
And we’re done! Thank you, Adam.
