Representing the language service provider (LSP) perspective, Hofkens explained that LanguageWire first focused on developing a user-friendly AI tool to assist customers who were facing issues with ineffective prompting and inconsistent output quality.
The tool leverages existing customer data to improve AI output while addressing data security concerns. Hofkens emphasized the need to integrate domain-specific knowledge, terminology, and tone of voice into LLMs, and LanguageWire addressed it by developing a system that uses customer data to augment AI prompts, ensuring more accurate and relevant output.
Hofkens remarked that “these LLMs out of the box don’t know much about your domain, your products, so how could they really write content about it? They don’t know your terminology, the way you have translated or the way you have written before, the kind of tone of voice … You might have all this data, but how do you connect it up to the LLM so the LLM can do something useful with it?”
On the buyer side, Pineger admitted initial skepticism with regard to generalized LLMs, particularly for safety-critical content like industrial machinery manuals. In his view, early experiments with AI for content generation proved insufficiently reliable, requiring extensive human oversight.
Pineger found LLMs valuable for research and business-related tasks, but remained cautious about their use in generating customer-facing technical documentation. He cited challenges in integrating AI-generated content into existing workflows and the potential for inaccuracies.
The conversation shifted to the evolving role of LSPs in the age of AI. Pineger expressed his preference for LSPs to focus on their core expertise — translation and localization — rather than attempting to replicate AI capabilities in-house.
“Along the way, we get a lot of data from our customers, be it the translation memories, term bases, the files that they send us, and so on. That’s a big data asset,” added Hofkens. “We can use that to tune our AI systems … to tune the LLM. Using retrieval augmented generation we can do exactly that.”
Hofkens noted that although many companies struggle to scale their in-house AI solutions effectively, LSPs are adapting by integrating AI into their services and leveraging customer data to enhance AI output. He sees LSPs evolving into data platforms, leveraging customer data to enhance AI services and provide valuable consultancy on AI adoption.
“Along the way, we get a lot of data from our customers, be it the translation memories, term bases, the files that they send us, and so on. That’s a big data asset,” added Hofkens. “We can use that to tune our AI systems … to tune the LLM. Using retrieval augmented generation we can do exactly that.” — Roeland Hofkens, Chief Product and Technology Officer at LanguageWire
Both Pineger and Hofkens anticipate a continued focus on improving the accuracy and reliability of AI-generated content. Hofkens highlighted the “agentic approach,” where multiple specialized AI models collaborate to refine content, as a promising avenue for development.
The panelists also agreed that human expertise will remain essential, even as AI takes on more tasks. They envision a future where humans focus on higher-level tasks like strategy, validation, and ensuring content aligns with business goals.
Continued innovation in AI, particularly in areas like automated quality assessment and the agentic approach will further shape the content landscape. Pineger thinks that people will still be working at a higher level as more and more work is performed for them. “But they’re still there to make sure that it’s appropriate … it’s [about] addressing the right challenges … further specialization. I think we’re just becoming more productive over time,” he added.
