At the beginning of 2025, I wrote a post trying to anticipate what the year would bring for technology and interpreting. As usual with predictions, I got some things right and some things wrong.
I was right about the increase in interest in AI interpreting. That trend was unmistakable and has only accelerated. I was also right about technical progress in end-to-end interpreting systems1. What I got wrong was something more subtle, but even more important: deployment of end-to-end systems. I predicted that we were not there yet, but we have seen first systems adopting this paradigm, and I had to correct myself. This suggests that the current acceleration of improvements can surprise even experts in the field.
Once again, it is time to look ahead to the new year: 2026. The shortcomings of my past predictions serve as a useful reminder of how difficult forecasting truly is, particularly when we focus on very short horizons like “the year ahead”. Real technological trends unfold over longer periods, often unevenly, and rarely in a straight line. Still, even with this admission of fallibility, there are directions that are already visible today, even to non-experts. They will not fully unfold in 2026, but they clearly point somewhere. I think it is worth to have a look.
My predictions base on thoughts and analyses I wrote down in the past few months. I will therefore refer to them here in forms of links to past articles; the patient reader might want to follow them, they often contains links or references to further literature and examples to substantiate my thoughts. Here are, in my view, the most relevant ones.
Machine interpreting quality will continue to improve
The most evident trend for 2026 is straightforward: machine interpreting quality will improve further. This will be especially clear in turn-taking interpreting (everything that is not simultaneous), where latency constraints are less extreme. But even simultaneous interpreting systems are bound to improve, incrementally but consistently, for example in terms of latency, and with first hints to multimodality in sight.
The main driver is no longer a mystery: large language models used as translation engines. Their ability to reason, contextualise, explain choices, and adapt output based on complex instructions is changing the game. From a programming perspective, the possibility to steer translation behaviour through sophisticated prompting strategies — rules, constraints, style guides, domain instructions — opens a space that older MT engines simply did not have.
There has been a certain reticence toward this approach, however. Many companies were skeptical about LLMs “translating properly” (hallucinations, latency, costs to name just a few). This cautions approach was well placed, but skepticism is dissolving. Major written translation systems are making the shift (think of the new models of DeepL and Google Translate). In the field of interpreting, I have seen working prototypes that are, quite frankly, incredibly good, especially in dialog interpreting. Think of a low-latency voice chatbot, like the ones already available today for mass usage, but explicitly instructed to translate, with detailed rules about register, omissions, reformulations, terminology, and interaction management. These systems are no longer speculative. Companies will start adopting them.
This does not mean the end of other approaches (for example cascading pipelines with ‘classical’ neural machine translation). On-device systems, small dedicated models, and client-hosted infrastructures will remain essential and an asset, especially where data privacy, determinism, and auditable behaviour matter (and this will matter more and more). The market will not converge on a single solution, for now. It will fragment, by design.
CAI tools will remain stable (and marginal)
Computer-assisted interpreting (CAI) tools will continue to exist, but their position will remain unchanged: useful, niche, and comparatively small.
In 2025 we saw new CAI products enter the market, mostly focused on speech recognition and simultaneous support, some focussing on training. This is good news. In 2026, I do not expect major novelty. The reason is not technological, it is economic. The market is simply too small to justify large investments, and the total addressable market is more likely to contract than to grow. CAI will remain the “smaller sister” of AI applications in interpreting: appreciated by professionals, but structurally limited in scale and ambition.
However, there may be a surprise here. Some market segments are now ripe to harvest the power of AI within interpreting tools. We might see a shift away from the elitist niche of conference interpreting toward broader deployment in public service interpreting, healthcare interpreting being the most obvious example. In my view, there is substantial untapped potential there: it is the largest interpreting community, yet it has benefited the least from technological innovation. Carefully designed AI support could bring tangible benefits both to interpreters and to the institutions and individuals who rely on their services.
From a technical perspective, 2026 might become the year in which CAI tools adopt an offline-first approach. This is particularly crucial in the area of real-time support. One of the main obstacles to adoption so far has been the need to send audio to cloud services in order to perform tasks such as real-time transcription. For many potential users, this has been a major bottleneck: data protection concerns have prevented large-scale adoption. The technology is now mature enough to deploy these systems locally on the interpreter’s device, albeit with some constraints.
Interpreting programmes will continue to decline
Another trend that will continue, slowly, but inexorably, is the decline in student enrolments in full-fledged interpreting curricula (one- or two-year MA programs). I anticipated this years ago, which at the time earned me a fair amount of resentment from academic colleagues.
From the students’ perspective, the logic is hard to contest. Enrolling in a long, demanding program with a high probability of professional obsolescence is not an attractive proposition2. This was already problematic when most graduates did not end up working as interpreters in a pre-AI era. Today, it is even harder to justify.
The paradox is that the market (hence society) will still need highly skilled interpreters, even in an age of highly capable machines. Not many, but those who are needed will need to be exceptionally good. The solution is not to defend the status quo. It is to rethink training. In my opinion, the era of overspecialization, such as a two-year master’s degree exclusively in translation or interpreting, is coming to an end. What we need instead are more holistic curricula, while professional training should move away from monolithic, long university programs toward short, specialized, professional, ad-hoc courses designed for specific profiles and real market needs. Some universities are already moving in this direction; many are not.
AI fatigue and anti-AI sentiment will intensify
AI is also generating a growing sense of fatigue. People are exhausted by the endless oscillation between hype and counter-hype. It is emotionally and cognitively draining. There is a real chance, one I personally hope for, that AI simply becomes normal. That the narrative bubble bursts, much like the dot-com bubble did. It is also possible that parts of the AI industrial bubble burst. That would not be a tragedy; it would be a correction.
At the same time, an anti-AI movement is clearly mounting, including in translation and interpreting. It is everywhere. In academia, it often appears in a peculiar form: papers that declare themselves “not anti-AI” and then proceed to list, in detail, why AI is harmful3. Among professionals, LinkedIn has become home to some truly bizarre expressions of outright hostility toward anything AI, especially when it threatens established interests.
These reactions are predictable, and partially understandable. It will escalate further. I have written about this dynamic in the past. The problem is that it will not lead to solutions. Frustration as resistance alone never does.
A personal note: why I remain optimistic
I look at 2026 with a great deal of optimism. I am convinced that the future of interpreting will be AI-first. When a technology becomes good enough, it does not remain an option, it becomes the default. The question is not whether this will happen, but how responsibly, transparently, and intelligently we use it. This opens up a wealth of opportunities, for many stakeholders (first and foremost for users). Ultimately, AI interpreting is aiming at the same goals and objectives I was trained to achieve as a student of interpreting (making people understand each others across language barriers), only by radically different means. There are opportunities also for interpreters, at least for the ones working in what i call the top market (or for all those working with languages for which AI is simply not good enough, and these are many). Furthermore, it is not only a matter of automation: AI-augmented interpreting is giving professionals tools that allow them to shine even more. There is space for many things.
I discovered that I am a builder, first and foremost. While I love to pause and write and theorize about how I perceive the changing interpreting ecosystem because of technological innovation, I still need to get my hands dirty with technology. Speaking while knowing, at least partially, the technology I am talking about. In 2026, I will continue investing time and energy in this direction: in automation, with a large project I am currently working on (more on that soon), and in augmentation, with important new developments for InterpretBank. I will also continue teaching conference interpreting. For my students, there is only one viable path if they want to thrive as interpreters: to be at the very top. That is not an easy feat.
2026 will not bring final answers, but it will make the direction I am predicting here increasingly clear. The upheavals ahead are profound, at times even dramatic, and they extend far beyond the small niche of translation and interpreting. Pretending that nothing is unfolding before our eyes, and continuing as if business were as usual, would be the wrong response. And with every year am so surprised to see stakeholders inside the field not recognizing this. It is the time to rethink everything. This is also the message I try to give to my readers and students.
- In a nutshell, these are systems that translate directly audio from one language to another language without relying on the intermediate stop of transcription. ↩︎
- Economic indicators are very tough for the translation and interpreting market, see this CSA forecast. ↩︎
- A beautiful, but not unique example, is the position paper “AI for Translation and Interpreting: A Roadmap for Users and Policy Makers”, signed by many academics, which is available here: https://www.iti.org.uk/resource/ai-for-translation-and-interpreting-roadmap.html ↩︎