2026-07-14 · 9 min read · Raw Drafts / Work and AI
I Want to Work Somewhere the Work Is Already Necessary
AI-generated work is creating a particular exhaustion: having to explain, every day, why human judgment, expertise, maintenance, and responsibility still matter.
The strangest thing about AI-generated work is that it does not only make people afraid of losing their jobs. It makes them tired of having to explain why their jobs were real in the first place.
A software engineer now has to explain why software is more than code. A designer has to explain why design is more than producing screens. A writer has to explain why writing is more than arranging plausible sentences. A researcher has to explain why finding information is not the same as understanding a problem. The machine produces the visible part of the work, and suddenly the human is responsible for defending everything that was previously understood to surround it.
This is what makes the current moment so exhausting. The work has not disappeared. In many cases, there is more of it. Someone still has to decide what should be built, judge whether the result is correct, understand the context, catch the failure, maintain the system and accept responsibility when it breaks. But because those parts are less visible, they have become harder to sell.
The worker is left in an absurd position. They are doing the job and making the case for the existence of the job at the same time.
There is a familiar term for the fear that a role might disappear: job insecurity. But that is not quite enough to describe what is happening. Many people are still employed. Their more immediate problem is that the meaning of their profession has become unstable. They are no longer only asking whether their position will survive. They are asking whether the knowledge they spent years developing still has a recognized place in the world.
Researchers studying generative AI in software engineering have described this as occupational identity threat. The phrase sounds academic, but the experience is ordinary. A profession gives a person more than a salary. It gives them a way to know what they are good at, what other people rely on them for and how their effort connects to something larger. When a new technology turns the visible output of that profession into a cheap demonstration, it can make the entire field look like a trick that has finally been exposed.
The code appears in seconds. The image appears in seconds. The report appears in seconds. The presentation appears in seconds. From the outside, this can look like proof that the work was always mostly production.
It was not.
The generated result often hides more work than it removes. It still has to be checked, integrated, repaired, adapted and maintained. It still depends on someone noticing what is missing. But the person doing that work now performs it under suspicion. Their expertise has become a claim that must be continuously demonstrated.
This is why so much of the current conversation about AI at work feels dishonest. It is framed as a story about assistance and productivity, while many workers experience it as a permanent public audition. They are expected to use the tools, show that they are using them, find impressive use cases and appear excited about the transformation. Caution sounds defensive. Doubt sounds outdated. Saying that something requires care can make you seem like the obstacle.
The result is not only more automation. It is more performance.
The employee must perform adaptability. The manager must perform transformation. The company must perform technological inevitability. Everyone has to produce evidence that they are moving fast enough, even when nobody has clearly defined where they are going.
A person can be enthusiastic about AI and still find this unbearable. The problem is not necessarily the tool. It is the requirement to rebuild one’s professional identity around every new version of the tool.
That is not the same as saying the tools are useless. A six-month randomized field experiment involving 6,000 knowledge workers found that access to generative AI reduced time spent on email and moderately sped up document completion. It did not significantly reduce time spent in meetings, because meetings required coordination across the organization rather than a faster individual workflow.
That difference is the whole problem in miniature. AI can make a task faster without changing the system that created the task. The saved time may simply become capacity for more work.
The pace increases while staffing remains unchanged. Workers may be left with fewer routine tasks and a higher concentration of difficult, ambiguous and emotionally demanding ones. Research on AI and autonomy at work describes both possibilities: automation can give workers more freedom by removing repetition, or reduce autonomy through monitoring, intensified pace and the devaluation of experiential knowledge. The technology does not decide which version appears. The organization does.
There is also no guarantee that companies will automate the parts of work people dislike. The easiest tasks to automate may be the parts that provide variety, satisfaction or a visible sense of completion. What remains may be supervision, correction, escalation and responsibility for failures produced by systems the worker did not choose.
This helps explain why some knowledge workers have started fantasizing about leaving for more tangible fields.
They talk about healthcare, public service, construction, hospitality, teaching, farming, transport, manufacturing or opening a small physical business. These fantasies are often naive. A hospital can destroy a person with bureaucracy. A school can be underfunded and cruel. A bakery can consume every waking hour. Physical work does not come with automatic dignity.
Still, the fantasy points toward something that digital work increasingly lacks.
A patient needs care. A train has to run. A building has to remain safe. A meal has to be prepared. A water system has to function. A public record has to be preserved. These forms of work can be badly managed, but their underlying necessity is usually not reopened for debate every time a new model is released.
That distinction matters.
Some workplaces are built around products whose existence must be continually justified. Others are built around obligations that remain even when the tools change.
A hospital can adopt AI, but the obligation to care for patients does not disappear. A transit authority can automate parts of its operation, but people still need to move through the city. A university can use language models, but someone still has to teach, evaluate, research and preserve knowledge. A public agency can change its software, but the service remains necessary.
In these places, technology may transform the method without erasing the reason.
That does not make the work safe. It does not guarantee respect, money or stability. But it changes the emotional position of the worker. They are less likely to begin every morning by proving that the category of work still deserves to exist.
This may be what many people mean when they say they want to escape tech. They do not necessarily want to stop using computers. They want to leave an environment where every useful skill is treated as a temporary inefficiency waiting to be automated.
They want to work somewhere with a durable obligation.
A durable obligation is something an institution must continue doing through changes in fashion, funding and technology. It has to care for people, maintain infrastructure, preserve knowledge, provide access, keep records, protect safety or make a public system function. The techniques can change. The responsibility remains.
Work built around a durable obligation has a different relationship to speed. Faster is useful, but speed is not the final argument. A generated answer is not enough when the payment must reconcile, the diagnosis must be defensible, the bridge must stand or the information must be correct. Reality eventually audits the output.
This makes judgment easier to value.
Long-term ownership matters too. Many digital workplaces are organized around launch. A system is produced, announced and handed over. The people who made it may never see what happens after six months. In institutions that have to operate something for years, maintenance cannot remain invisible forever. Documentation, reliability, accessibility and accumulated knowledge become part of the work rather than embarrassing evidence that the first version was incomplete.
The same is true of expertise. Some organizations treat experience as something to preserve. Others treat it as friction.
The difference becomes obvious in how they introduce AI. Does the organization ask workers where the tools are useful? Does it allow them to identify risks and limits? Does it distinguish between assistance and replacement? Does it care about the knowledge that may disappear when a task is automated? Or does it simply announce that everyone must become AI-first and wait for the workforce to reorganize itself around the slogan?
The lack of worker influence is not incidental. A 2026 Jobs for the Future survey found that 56 percent of workers said their employers had not consulted them about how AI tools were used in their work. Workers who reported having a lot of influence over workplace technology were more than twice as likely to report high job satisfaction as those who had no influence.
The OECD similarly reports that training and worker consultation are associated with better outcomes, and that consultation can improve both acceptance of new tools and job quality. People are more willing to adapt when they are allowed to participate in the adaptation. The technology feels less threatening when it is placed inside a stable purpose. It feels more threatening when the purpose of the worker is placed inside the technology.
There may be no truly AI-proof career. That is probably the wrong thing to look for. The more useful question is whether the work belongs to something that will still matter after the current wave of excitement has passed.
Does the organization exist because it has found a temporary market, or because it carries a responsibility? Does it value people for producing visible output, or for keeping a real system working? Does competence accumulate there, or must it be reinvented every quarter?
The desire to stop proving that your work matters is not necessarily a desire for comfort. It may be a desire to spend more time being useful and less time performing usefulness.
Some people will thrive in fields where the definition of the job changes constantly. They will enjoy the speed, the reinvention and the lack of fixed boundaries. Others will look for slower institutions, physical consequences, long-term systems and obligations that cannot be dismissed by a demo.
They are not running away from the future.
They are looking for somewhere the work is already necessary.
AI authorship note
This essay grew from Kris Haamer’s attempt to understand why AI-generated code can create a desire to leave for work whose necessity does not need to be defended every day. AI supported research, framing, drafting, editing, and implementation. Kris is responsible for the original question, the concept of a durable obligation, the editorial direction, and the decision to publish it.