How AI Will Restructure the Language Industry

The Breakthrough of LLMs

Three years ago, the industry was still focused on squeezing out marginal gains from technologies and models that had reached maturity. Then, LLMs suddenly emerged, catching everyone completely off guard. While the initial reaction was denial, it soon became clear that LLMs were not merely the next iteration of MT. Today, it is difficult to name a single company in the localization space that does not present itself as AI-first or AI-native.

The transition toward LLM-based workflows is still in its early stages, but it has already become clear that the workflows and tools that have defined the industry for the past 30 years — and indeed the entire segmentation paradigm — will increasingly appear obsolete. I believe our current model is turning into a kind of Ptolemaic system, an increasingly convoluted structure that has become a constraint for everyone involved in the process, from localization managers to translators.

Let’s take software localization as an example. A single string can have very different meanings depending on the context in which it appears: as a button label, window title, or error message. This is why companies generally provide screenshots, videos, and other supporting materials to help translators understand how the text will be used in practice. In the current model, preparing even a single string is a cumbersome process of collecting screenshots, attaching them, uploading them into the TMS, and managing versions. Then the MT engine comes in and ignores all that context, leaving the translator to review the output while reopening reference materials to ensure the text makes sense — a paradox in which automation multiplies the work instead of reducing it.

This happens because legacy tools don’t distinguish between content types. For a computer-assisted translation (CAT) tool, a label, newsletter, and academic paper are all the same thing: a collection of segments. Yet this principle has never really been questioned. Is it an axiom? If we think about it, there is no valid reason to use the same system to localize every kind of content. Translators already rely on specialized tools for subtitling and dubbing, and for good reason.

Let’s consider that an LLM can read instructions, interpret screenshots, use images as context, simulate layout constraints, and adapt to a wide variety of inputs. It becomes evident that localization of software — or any other content type — would benefit from tools designed specifically for LLM-driven workflows. And such tools will not be just a list of strings with some additional information attached.

The New Role of Linguists

In this context, the role of linguists will evolve, moving beyond the traditional production-plus-revision model toward a logic of co-creation, where linguists act not as proofreaders but as directors. The conventional expertise will no longer be sufficient — or even relevant — in an environment where the real leverage comes from designing prompts, shaping context, and orchestrating the behavior of AI systems. In other words, a new paradigm demands new competencies.

These professionals will no longer translate word-for-word. Instead, they will configure, train, and oversee translation and localization agents, ensuring their large-scale operation meets the highest standards for quality, style, and operational efficiency. The primary responsibilities of this new role, which we might call “AI Language Supervisor,” will include:

• Prompt and context engineering. Structuring requests and providing the proper context will become the essential paradigm for achieving reliable, high-quality AI outputs.
• Model customization. Leveraging techniques such as retrieval-augmented generation (RAG) and fine-tuning to adapt systems to specific domains and brand voice will make these activities increasingly simple and accessible even to non-technical staff.
• Quality supervision. Monitoring and improving model performance through systemic metrics and advanced techniques like reinforcement learning from human feedback (RLHF) will reduce reliance on continuous manual reviews.
• Linguistic governance and compliance. Ensuring consistency with corporate guidelines, adherence to regulatory requirements, and sensitivity to the cultural specificities of target markets will be needed.

Acknowledging this emerging strategic role requires recognizing an important implication: There will be fewer linguists overall. The operational act of translating will be performed predominantly by machines, and anyone claiming otherwise is either lying or ignoring the scale of the efficiency gains already visible in LLM-based workflows. Even in highly specialized domains such as legal, literary, and medical translation, the ability of LLMs to accelerate drafting and revision will dramatically reduce the amount of human labor required to translate or correct text line by line. What will remain indispensable is not the manual execution of translation, but the expertise needed to steer, supervise, and govern AI systems at scale.

Rethinking Outsourcing

LLMs are poised to reshape the very architecture of corporate localization — its workflows, its governance, and the division of responsibilities between internal teams and external vendors. Under the current model, companies rely heavily on LSPs to execute and scale repetitive, routine tasks. Outsourcing keeps internal teams small, but comes at a high cost. In the current model, only a small fraction of the value of a translation reaches the linguist who actually performs the work; a translator is fortunate if they receive even 30% of the value paid by the client. The rest is absorbed by layers of project management, overhead, and administrative processes that add coordination but not linguistic expertise — or by the hours spent sitting silently with the camera off during alignment calls.

LLMs enable the creation of systems of AI agents that can perform the following tasks with excellent efficiency and precision:

• Content and context preparation. An AI agent can analyze the source content (text type, complexity, required languages, and relevant glossaries), gather reference materials (such as samples of previously translated texts and complementary information like marketing strategy guidelines), extract terminology, and assemble all contextual elements needed for accurate localization.
• Quality assurance (QA) and risk detection. An AI agent can run automated QA checks (of terminology, formatting, placeholders, and consistency) and flag high-risk segments that may require human intervention due to ambiguity, cultural nuance, or compliance concerns.
• Governance, reporting, and compliance. An AI agent can verify alignment with brand voice and regulatory requirements; ensure cross-market consistency; generate dashboards on cost, quality, and turnaround time; and provide insights that support strategic decision-making.

In this scenario, cost and process optimization do not come from cutting translators, but from rethinking the end-to-end workflow — letting AI handle repetitive, low-value tasks while enhancing human contribution in the critical stages of orchestration, supervision, and validation.

This shift also raises an important question: Will outsourcing to a traditional LSP still make sense when the nature of the work itself is changing? As localization shifts from executing operational tasks to governing AI-driven systems, the content of outsourcing will change accordingly. What companies delegate will no longer be routine production work, but strategic capabilities such as designing prompts, shaping context, supervising AI behavior, and aligning multilingual content with business objectives.

For companies with a global footprint, these are not peripheral activities — they are core functions. Continuing to outsource them risks externalizing the very capacity to govern the localization process, a capability that will become increasingly central in organizations operating across multiple markets. Naturally, this approach requires organizational transformation; corporate localization teams must develop generative AI (GenAI) skills to continue playing their role in steering localization strategy, governing processes, and ensuring that AI-driven systems align with business objectives.

Orchestrating Humans and AI

With the rise of LLM-based systems, the role of localization manager will no longer be just about coordinating vendors and files, but about orchestrating AI agents that autonomously handle much of the operational workload. It’s a transformation similar to what has already happened in other business functions: in marketing, where campaigns are no longer built manually one by one but are generated, tested, and optimized in real time by automated systems; or in sales, where customer relationship management systems (CRMs) have evolved from digital calendars into intelligent platforms that record calls, prepare follow-ups, and highlight the most promising opportunities.

The traditional responsibilities of the localization manager will not disappear, but they will be reshaped through the use of AI. Strategy definition and resource allocation, for instance, will no longer rely solely on static planning but on real-time insights generated by AI agents. With the growing adoption of vibe coding and natural-language interfaces, corporate data is becoming increasingly accessible to non-technical teams; employees can generate dashboards, explore datasets, or run complex analyses simply by describing what they need. Applied to localization, this means that managers can directly integrate marketing, product, or sales signals into the translation process, linking decisions not only to operational metrics such as volumes and turnaround times, but also to business outcomes — including engagement, conversions, and retention.

Alongside these responsibilities, the new role of “AI Localization Manager” will require skills that reflect a shift from operational control to orchestration and leadership:

• Designing orchestration frameworks: Defining how AI agents, human reviewers, and external partners interact across the workflow, as well as ensuring the right balance between automation and human oversight.
• Coordinating AI Language Supervisors: Channeling linguistic expertise into system behavior, integrating linguist feedback into agent design, and making sure quality standards scale across markets.
• Ensuring team enablement: Providing team members with the proper training and resources to work effectively with GenAI, enabling them to craft effective prompts, design adaptive workflows, and collaborate with AI systems.
• Driving organizational learning: Using insights from AI performance and business outcomes to continuously refine localization strategies, linking language operations more closely to engagement, conversion, and customer experience.