In a conversation with Eddie Arrieta on the Localization Today podcast, I shared an idea that has since become the heartbeat of Gridly’s strategic discussions: the localization triathlon. 

Just as a triathlete masterfully balances swimming, cycling, and running, modern localization requires a cohesive mastery of three disciplines: artificial intelligence (AI) for speed and scale, automation for efficiency and flow, and humans for creativity and cultural nuance. 

The core of this framework is the rejection of silos; we cannot treat these as three separate events. Instead, the goal is to move from being manual executors to becoming “AI-friendly linguistic orchestrators” who design connected workflows where content flows seamlessly from AI generation through automated quality gates to final human refinement. 

While the triathlon provided a winning spirit for the industry, it also set the stage for a much more uncomfortable conversation about what happens when one of those disciplines — the infrastructure beneath it all — begins to fail under the weight of history.

Over the past few months, conversations with game localization leaders have repeatedly reinforced this same idea. Whether discussing AI adoption, live game operations, or vendor collaboration, one challenge continues to surface: the growing complexity of managing localization at scale. Among those conversations was an online panel discussion with three game localization experts — Denis Ivanov, Tucker Mills, and Mike Kim — who carry over 45 years of combined localization battle scars. 

The atmosphere was what Denis described as a “dark forest vibe,” where many studios are wandering in the shadows, terrified that AI might swallow their creative lore or trigger player backlash. The skepticism in the room was palpable, yet Mike Kim dropped a bombshell that shifted the energy: His company no longer employs internal translators, choosing instead to automate the first draft of every single string via AI and utilizing linguists strictly for high-level contextual refinement. 

Despite their different approaches, these experts arrived at a singular, sobering consensus: The game localization industry is drowning in technical debt. We realized that the hidden debt of outdated infrastructure and workflow bottlenecks is the true ghost in the machine that no amount of fancy prompting can exorcise. It was hilarious to watch three seasoned experts agree that our greatest hurdle isn’t the “scary” AI, but the “rusty” spreadsheets we refuse to let go of.

When we pulled back the curtain on these technical challenges, the “debt” was more than just a metaphor; it was a series of literal structural failures. 

One of the most glaring issues is the persistence of hard-coded strings, where developers still embed text directly into the game code rather than using localizable files — a practice that makes AI integration effectively impossible. We also discussed the “Excel hell” of legacy file management, where teams are still manually processing hundreds of spreadsheets instead of using streamlined APIs or connectors. 

Furthermore, AI tools often “sit beside the pipeline” rather than inside it, creating extra steps and complexity instead of removing them. This debt is compounded by a lack of metadata and classification; without knowing which character is speaking or the specific intent of a hero’s skill, large language model (LLM) is a blind messenger. Without source consistency and “templatized” English, AI output requires a “ton of rework,” essentially negating any efficiency gains the technology promised.

However, the debt goes even deeper into the organizational and legal marrow of our companies. We have historically neglected engineering resources for localization, often forcing departments to wait a year or more for internal tools. There is also a significant legal and IP debt; without robust glossary management, AI can inadvertently generate translations that reference trademarked terms from other franchises, creating massive liability. 

This realization resonates deeply with the localization triathlon. Long before AI was the headline of every magazine, we were talking about automation, yet many game studios struggled even then because they weren’t ready for true scale. Now that AI has arrived, it has made our existing debt visible and undeniable. 

I see a growing AI fatigue beginning to appear in many organizations, but it is often misunderstood: AI fatigue can happen when expectations created by impressive AI demonstrations collide with fragile real-world systems. When teams attempt to layer AI on top of fragmented game localization pipelines, inconsistent source text, and manual file management, the technology becomes another source of complexity rather than relief. 

AI fatigue is not caused by AI itself but by the repeated friction between AI expectations and operational reality. For years, the game localization industry has quietly operated under the assumption that the existing system would somehow continue to hold: trust that another spreadsheet, another workaround, another manual patch would carry us through the next release. In debt we trust! 

AI did not create this debt. It simply illuminated it. The models are not rebelling against our workflows; they are faithfully reflecting them. And that reflection can be uncomfortable.

But perhaps this is also the opportunity. If the localization triathlon taught us anything, it is that speed alone never wins the race. Sustainable scale comes from a balance between AI, automation, and human expertise. 

AI may be the loudest leg of the triathlon today, but it is also the one exposing where the other two have been neglected.

In that sense, the real promise of AI is not that it replaces game localization professionals. It is that it finally forces the industry to rebuild the systems we have been trusting for far too long.