Think Tank takeaways from LocWorld 53

Several AI Localization Think Tank members attended the LocWorld53 conference that took place in Malmö, June 3–5, 2025. In this article, they each share their insights and takeaways from the event, shaped by keynote moments, discussion panels, and the conversations happpening in the hallways.

AI Localization Think Tank members at LocWorld53

Marina Pantcheva’s takeaways

The LocWorld53 highlight for me was Brian Klaas’ keynote. His insights from Chaos Theory offered unexpected parallels with AI's impact on localization.

1. The Butterfly Effect in Localization built on AI: Taken literally, the Butterfly Effect describes how a butterfly flapping its wings in one part of the planet can ultimately cause a devastating thunderstorm thousands of miles away. The explanation behind it is that in a densely interconnected system, like the weather, the tiniest fluctuation in one point of the system can trigger massive consequences elsewhere. Chaos theorists call this “sensitive dependence on initial conditions.”

   Anyone working with AI in localization has experienced this sensitive dependence firsthand. Remove a single space in a prompt or swap a comma for a semicolon, and the AI output can change drastically. Prompt engineers operate at the edge of chaos.

   But it’s not only prompting. Our systems rely on a densely connected web of interdependent elements: cloud platforms, NMT engines, AI models, translation memories, termbases, and global talent. A small failure in one component can paralyze entire operations or lead to critical issues. This is the fragility of interconnected systems.

2. The sandpile effect in Localization: Klaas’ described this effect using the analogy of building a pile of sand by dropping one grain at a time. For a while, nothing happens; each new grain simply adds to the pile. But as the pile grows taller, it eventually reaches what scientists call a "critical state”.  At this point, adding just one more grain might do nothing, or it might trigger a massive avalanche that reshapes the entire pile.

   Our tech solutions are no different. We build taller and taller tech stacks. Their complexity and fragility grows with every additional layer. One single innovation, one unforeseen breach, one unexpected frontier AI model released by an AI lab may trigger a cascading change and reshape the entire pile. Not every shift is catastrophic, but the sense of precariousness is real.

3. Influence without control: Klaas’ most captivating message was “We control nothing, but influence everything”. He urged the audience to prioritize resilience over efficiency. This is a valuable principle to follow when integrating AI in localization.

We cannot control how AI evolves, or which technologies will eventually dominate, but we’d be wise to design resilient tech solutions, balance technology with human expertise and ensure that core skills are preserved.

My main takeaway from Brian Klaas' keynote is that we cannot control how AI evolves, or which technologies will eventually dominate, but we’d be wise to design resilient tech solutions, balance technology with human expertise and ensure that core skills are preserved.

Belén Agulló García's takeaways

The underlying goal of attending LocWorld is, as always, to stay ahead of the curve and look into the future. However, even if we try really hard, it seems that we’re always chasing that idea, that futuristic approach that makes us look very visionary. I didn’t find that in this edition, despite the great speakers and insightful talks in the program. As other attendees have mentioned, it was more of a benchmarking exercise to see where other companies are heading and where their focus is. Spoiler alert: the focus is on automation, making more with less, basically. So despite our attempts to shift localization to the left and turn what we do into something more strategic, it looks like the way of showing value to stakeholders is still to prove that you can optimize the process, and that you can save money for the company, which leave us in the eternal spot of being a cost center. I don’t think that’s a bad thing per se, but we should be able to combine different strategies to show value beyond cutting costs. Yes, we can optimize processes, but we can also influence growth, user experience, brand perception, cultural awareness, and so on.

All in all, my main learning from this LocWorld edition is that we should focus on building resilience (quoting the keynote speaker) but also empathy for our colleagues and for the end users, audiences, patients, readers, artists, and humans whom we serve with what we do.

Despite all the talk about technology, optimization, and AI, two key elements were missing in the conversation, in my opinion:

1. Ethics and impact on humans: No one mentioned the ethical implications of using AI or the impact on people. There was a specific talk where the speaker mentioned how they had optimized the process and reduced costs, and someone else from the audience asked what happened to the project managers who were part of that process. The answer was that they were not renewed. A win for the company, but not so much for the human. This is just an example, but I do think we need to add these topics to our conversations more often, even if they make us feel uncomfortable.

2. Impact on end users and quality: In some sessions, end users were left out of the conversation as well. While different speakers presented how they were incorporating more MT and more AI into their workflows, they didn’t share insights into how users were perceiving that shift, if at all. During one of the conference lunch breaks, we had a lively conversation with the Think Tank members about how difficult it is to disentangle product and user experience from localization, and therefore, how to measure the impact of any changes on quality. This is not a new conversation, but now more than ever, it is important to keep experimenting and exploring how language and UX impact user behavior, brand perception, content performance, and so on. Personally, I would love to see more studies and more real data on this matter.

All in all, my main learning from this LocWorld edition is that we should focus on building resilience (quoting the keynote speaker) but also empathy for our colleagues and for the end users, audiences, patients, readers, artists, and humans whom we serve with what we do.

Finally, I would like to highlight one quote from Cristina Anselmi, which I think perfectly encapsulates the zeitgeist of the present moment: “Before, we had to convince leadership to use MT; now we have to convince them not to use AI for everything, no matter what”.

Libor Safar's takeaways

The general mood continues to be one of fin de siècle. So much progress to celebrate. But accompanied by a sense of both crisis and hope for renewal.

1. LocWorld53 made it abundantly clear again: AI is not a differentiator, it's a feature.

2. Everyone is developing something, including heavily on the client side. So many use cases, so many custom solutions, so much fragmentation.

3. Yes, it's not about the features, it's about the credibility, more than ever.

4. Because it's about the credibility, people matter more than ever. And so does attending events like LocWorld.

5. Long, intimate conversations with AI are cool, but only by talking with other people do we get to validate or change what we believe, and how genuine ideas and connections get generated.

6. The vast majority of the exhibitors were more or less tech companies, even if tech is not what they sell.

7. Like at GALA Montreal recently, LocWorld Malmö was smaller than last time, or than Dublin last year. But it was still excellent.

8. Events are always a long game, and consistent presence works well for companies or individuals. Being visible here has an impact in terms of perception far beyond the immediate attendees.

9. The general mood continues to be one of fin de siècle. So much progress to celebrate. But accompanied by a sense of both crisis and hope for renewal.

10. Still cool to see relatively new entrants, like Gridly, combining TMS and CMS into one (my first time talking to them thanks to LocWorld), or Uber, starting a new gig as an LSP.

11. There's no definitive answer to what all this means or what the future holds. We're still waiting for the right Oracle of Delphi to appear. So it's Python before Pythia for now. Perhaps at #LocWorld54?

Gabriel Karandyšovský's takeaways

This LocWorld reinforced a feeling I had at GALA Valencia in 2024 (borrowing the title of a great song, go look it up!): We’re arriving somewhere, but not here. The language services industry is in flux (still), and the destination remains uncertain. Three key learnings (plus a bonus one) emerge.

1. The keynote speaker was addressing our souls when he delivered, “We control nothing, we influence everything.” This industry has long been obsessed with convention, legacy processes, and structure, and now AI is challenging our impulses for control. Someone (or something?) had to. This conference (and others, too) has done a good job showing that people are taking on the responsibility to navigate the sea of change.

2. The destination doesn’t matter nearly as much (we’re all just grasping at straws, aren’t we?) as the questions you ask and the solutions you find for yourself, your team, or your organization. I think it’s high time we (as in the collective industry) start asking sharper questions to nudge us out of our stupor and toward actively shaping a mutually beneficial future. Less navel-gazing, more questions, more experiments, and more of influencing the little things!

 This LocWorld wasn’t as good as it gets, nor is the present day the peak of localization as we know it. Our best work is yet to come. Good luck to us to figure out what that work will look like.

1. If the destination doesn’t matter, it’s the humans you meet along the way that do. Humans inspire you (can you really say AI has done the same for you?), challenge your ideas, and help you ask sharper questions. It’s from this co-mingling of ideas and viewpoints that small initiatives are born that can then move mountains.

2. Bonus learning: This LocWorld wasn’t as good as it gets, nor is the present day the peak of localization as we know it. Our best work is yet to come. Good luck to us to figure out what that work will look like.

Yota Georgakopoulou’s takeaways

LocWorld this year was about butterflies, chaos and AI.

"We control nothing, but we influence everything.”

As UCL professor Brian Klaas analysed chaos theory for us, I kept thinking of the chaos we find ourselves in as the entire industry is trying to figure out how to redefine itself with AI. I loved the image of seeing ourselves as butterflies, flapping our wings with purpose, hoping to cause some positive change in the course of future events. This is exactly how I see our think tank 😉

His advice is what stayed with me - focus on “resilience over optimisation”. Don’t just rush into AI for AI’s sake. Test and learn. Start small, plan for the risks. Monitor and adjust course. Deliver translation solutions. Peloton provided a good example with their AI subtitle workflow of how to do just that.

Resilience to me also means that the human in the loop remains the most important part of the localization equation. Yes, we can deliver automatic translations, but we still need to know how to translate without machines, otherwise how can we tell when the machines are wrong? And how can we fix things when they break? Spain without electricity for a day earlier this year was a good reminder that infrastructure fails and resilience matters.

As the most important part of the equation, we need to look after this human - the translator (or cultural consultant, or civilisation engineer, or whatever you want to call translators). There wasn’t much talk about displaced workers and loss of jobs at the event, and this is a conversation we also need to have.

I loved the ‘four visions’ panel. After all, we do go to conferences to try and take a sneak peek into the future.

We should definitely take note of Jaap van der Meer’s vision, or rather call to action. “Aufhebung”. Our industry is changing - it is getting uberized. For a while now, might I add. Two years ago at this very event Jaap also predicted that LSPs would disappear in the future and become a core part of larger organisations, much like IT departments. I can see that happening to an extent already, I just hope it’s not the IT departments that take over localization but that we partner with them, bringing our language intelligence to their technical expertise and have them help us in our mission to provide universal accessibility.

Human translation will be viewed as a luxury good in the future. And why not labelled as such too?

I fully endorse the ‘pop localization’ idea by Olga Stokowiec as a perfect description of what is happening in our market: AI-powered solutions being the great averager and accounting for the largest translation volume, how mindless use of AI can go really-really bad at the one end of the spectrum and how human translation will be viewed as a luxury good in the future. And why not labelled as such too? What I like to call in the entertainment space ‘translation with a WOW factor’ - cheers to that!

Marta Nieto Cayuela’s takeaways

This year, it seemed like the Translation Quality flag was flying high at LocWorld. Many sessions included it in the programme. I was as thrilled as I was overwhelmed and had to carefully choose what to attend and what to miss.

Is quality the new clickbait? It was for me. Quality should not be oversold, or underexplained. Most panels, with a few exceptions, revolved around the new definitions of quality and the need to align it with business goals and user expectations. But the amount of new information was limited. Was it due to secrecy? Competitive advantage and confidentiality still play a big role in our industry, and while the Quality flag is raised, it often feels like it is not waving that high. Only a few sessions, like those from Spotify, Skyscanner, or Microsoft, offered actual facts, data, and how-tos. It is this kind of transparency and generosity that moves the field forward.

One of the biggest gaps in this year’s conversation, as Myriam Garcia Vizuete from Kobalt Languages noted in the panel “Survival Strategies for Small and Medium LSPs in the Age of AI,” was the absence of discussions on AI costs, rates, ethics and data confidentiality. Four of the most pressing topics to the industry.

LocWorld53 offered the best pulse check possible we could ask for: a moment to face the second half of 2025 more aware, more intentional and more conscious of how even the smallest actions or most random decisions can shape the future of the industry, ourselves, and the world. For me, that attention to the noise is what quality is about. And while we are putting it at the forefront, it still remains a secret to some.

AI showed up across the programme and in every corner. However, the spotlight was mostly on innovation, potential and workflows. What felt missing were the tougher conversations —again facts, data and how-tos. What does this cost, and who is paying? How are rates shifting? What happens with user data, especially when it is sensitive? And how are we defining ethical boundaries in practice, not just in theory?

As we continue to integrate (more) AI into localization, as we continue to put order in the chaos as the storytelling animals we are, we need to talk about the real cost; not just in budgets, but in quality, trust, and long-term impact. Until then, our conversations will remain incomplete, and experiments will not turn into operations.

At times, despite the pace, it seems like we are not moving. Like progress is stalling. And that the flags we raise feel more like placeholders than banners of progress. But don’t get me wrong, we are; we just don’t have yet the right perspective to measure it from where we stand today.