Designing Work Humans Still Want to Do

The article below is a guest contribution by Daniel Sebesta, an experienced linguist and successful auto-entrepreneur. His written essay is based on his presentation "Creative Control in AI Translation: Designing Work Humans Still Want to Do," delivered as part of the AI ThoughtCon 2026 on March 31.

From production to judgment

AI has made translation faster. Drafts appear instantly, volumes scale easily, and the blank page is no longer a bottleneck. But focusing only on speed and cost misses a more important shift. What is changing is not just how fast translation happens, but what the work actually consists of. The center of gravity is moving away from typing and toward judgment.

Judgement is not a new skill. Linguists have always evaluated and refined draft texts. But in AI-mediated workflows, this activity becomes more concentrated and more cognitively demanding. The easier it becomes to generate language, the harder it becomes to assess it well.

Yet, in many workflows, the human role is still described as “post-editing.” The label suggests a simple sequence: the machine produces, the human corrects. In practice, however, the work is far more complex.

Linguists are making continuous decisions: Is this accurate? Is it consistent? Does it fit the intent and audience? These are not mechanical adjustments. They are acts of judgment under uncertainty.

And that matters. Once the work is framed as “editing,” it becomes easier to design workflows that prioritize speed over decision quality: fragmented tasks, limited context, and pricing models that assume uniform effort. But if the core activity is judgment, then the conditions under which that judgment happens become critical.

The easier it becomes to generate language, the harder it becomes to assess it well.

Judgment as the new creative core

This also changes how we should think about creativity. There is a persistent assumption that AI removes it. But creativity has never been limited to producing text from scratch.

In AI-supported workflows, it shifts. AI generates possibilities. The human role is to select, shape, and refine. That is a different form of creativity, not a lesser one. Judgment is creative. So is designing how the work gets done: choosing tools, structuring workflows, defining inputs.

At the same time, this shift will not appeal to everyone. Not every linguist wants to design workflows or build a tech stack. That is fine. But it makes workflow design more important, because different setups lead to very different experiences of the work.

The problem of context collapse

One of the most significant issues in current workflows is context collapse. Translation is often handled as isolated segments, increasingly filtered by automated quality scoring. On paper, this looks efficient. In practice, it creates a structural problem.

Good judgments and decisions depend on context: the surrounding text, the document as a whole, and the communicative purpose. Terminology, tone, and meaning unfold across segments. When that broader context is missing, linguists are asked to make local decisions without seeing the system those decisions must fit into. You cannot make good decisions without understanding that broader context: the text, the workflow, and the intended outcome. Even experienced professionals can only approximate coherence under these conditions.

The result is reduced quality and increased cognitive strain, because the linguist is constantly filling in missing information.

The hidden cognitive load

This connects to another misconception: that AI makes the work easier. It reduces certain types of effort. Typing decreases. Repetition can be automated. But effort does not disappear; it only shifts. Less production, more evaluation. And evaluation requires fragmented but sustained attention. Small decisions accumulate, leading to decision fatigue, especially under time pressure and with limited context.

From the outside, productivity increases. From the inside, the work can feel more demanding because the cognitive load has been reallocated into a less visible form.

At this point, it helps to separate what AI can do from how it is used. Technically, AI systems can adapt, incorporate feedback, and reduce the need for repeated corrections. So when the same issue is fixed repeatedly, that is not a technical limitation but a design choice.

Many workflows are still static: outputs are generated upfront, feedback loops are weak, and systems do not adapt. From a management perspective, this is understandable. Static workflows are easier to standardize and scale. But they also limit linguists' ability to influence the process and to do their best work. This is the distinction between capability and control.

Workflows that enable expertise

AI expands what is possible. Workflow design determines what is realized. For those designing multilingual content workflows, this has practical implications. If you involve human experts, the goal should be to enable them to add value, not constrain them to correction.

Some basic signals matter:

• Context access: Can linguists see beyond isolated segments when needed?

• Adaptation: Do corrections improve future output, or repeat endlessly?

• Upstream input: Can linguists influence the assets, prompts, or instructions used?

• Incentives: Is the work optimized only for speed, or also for decision quality?

If these are missing, human contribution becomes reactive by design. If they are present, the same technology supports more meaningful work.

This also shapes where linguists sit in the process. Some operate as executors within predefined structures. Others participate with some influence. A smaller group helps design tools and workflows. These positions affect not just efficiency, but the experience of the work itself.

Designing for better work, not just faster work

So the question is not whether AI will continue to change translation. It will. The more relevant question is what kind of work we are designing around it.

If human expertise is part of the system, then the system should allow that expertise to function through context, feedback, and room for judgment. Otherwise, we risk workflows that are efficient in appearance but fragile in quality and unsatisfying to perform.

AI can remove friction and expand what is possible. But whether it leads to better work or just faster work depends on how much control remains with the people doing it.