When the Work Changes Before the Job Ends

At the end of the demonstration, everyone is invited to be excited.

The new system has taken a stack of customer requests and produced neat answers. It has found the order numbers, checked the delivery dates and drafted replies. A task that used to take much of the morning has taken a few minutes. The screen looks clean. There are no unanswered messages in the demonstration.

Diane looks at one of the replies. The customer has asked to change an order, but the goods are already on a truck. The system has used the date in the office record. The driver left early. Diane knows this because she spoke to him while making tea.

This is a small fact. It changes the answer.

She can see the value of the new tool. She has no deep attachment to copying order numbers between windows. She would gladly stop doing that. But she can also see that someone must decide which answers can be sent, which need a call and which rest on information the system never received. She wants to know who will have time to do that work after the demonstration is over.

Then comes a second question. If the tool saves two hours, what will happen to those hours?

The manager talks about capacity. Diane thinks about her pay. Her colleague thinks about the vacancy the company has not filled. They have watched the same screen and are now imagining different futures.

Much of the public argument about AI happens at this distance from the actual work. A system performs an impressive task, and the discussion jumps straight to the end of employment. Or a system makes an obvious mistake, and someone declares that ordinary workers have nothing to fear. Neither leap tells Diane what her next six months will be like.

A job can change before it disappears. It can become easier in one respect and harder in another. It can lose routine tasks while gaining oversight, customer complaints or responsibility for errors. It can become more interesting. It can become a race to do more with fewer people. Your job title may survive while the bargain behind it changes.

That is where this book begins. Close enough to the work to see what is happening, and far enough from the day’s panic to ask what kind of life the change should serve.

What the demonstration leaves out

When we say a machine can do a job, we often mean that it can do a visible part of it.

A teaching job includes explaining a lesson. It also includes noticing a student who has stopped trying, deciding when a class needs a different approach and dealing with the institution around the classroom. A building contractor prepares estimates, buys materials, judges what an old wall will bear and works around the weather. An office worker may spend part of the day writing and another part discovering what the writing needs to say.

This does not make the entire job safe. Tools can take over enough tasks to change the number of people needed. Employers can divide roles differently. Customers can accept a cheaper service that is less personal. A task need not be perfectly automated for someone’s hours to fall.

Still, the steps matter. What a system can produce is one question. Whether it can do so reliably, at the required cost and with the available information is another. What an organization then chooses to do with that ability is a decision made under its own rules and pressures. The effect on a particular worker comes after those decisions, sometimes quickly and sometimes much later.

Text equivalent: From a capable tool to a changed life

Part of figure	Value or meaning
Title	From a capable tool to a changed life
Status	Original decision aid, not a prediction
Question	Task capability
Question	What can the tool do in a test?
Question	Reliable use at work
Question	Can it work with real information, checks and costs?
Question	Employer choices
Question	How will the organization use that ability?
Branch meaning	Possible choices, alone or combined
Possible employer response	Change the tasks
Possible employer response	Change paid hours
Possible employer response	Change staff numbers
Question	Worker and household outcomes
Question	What changes in income, time, control and security?
Wider conditions	Public support and care shape the options too.
How to read the figure	Arrows show what to ask next, not inevitable events. The three employer responses are examples that can combine. Public support and care also shape what happens to a household. No effect size or probability is shown.

Figure 1. An original guide to questions, not a forecast.

The International Labour Organization’s 2025 analysis of generative AI makes this distinction important. Its index estimates occupational exposure from tasks and technological capabilities. Exposure identifies where work could be affected. It is not a count of jobs already lost, and it does not set a date for future layoffs. The report treats changes within jobs as a central possibility.¹

For a person trying to plan, the difference is practical. A list calling your occupation exposed is a reason to look closely at your work. It is not a letter of dismissal. Nor is an unexposed label a promise that your employer, local economy or household will remain unchanged.

The useful question is smaller and more demanding: which part of my work is changing, and what evidence do I have?

Perhaps requests that once came to you now go through a tool. Perhaps the same number of workers are being asked to handle more customers. Perhaps junior tasks are no longer assigned, so new workers have fewer chances to learn. Perhaps an employer has bought software but staff still spend much of the day fixing its output. Each observation points to a different problem.

Be wary of a broad prediction that asks you to ignore those differences. It may sell a course or gather an audience. It will not necessarily help you choose.

Change can be real before pay changes

One reason this period is difficult to read is that different measures can move at different speeds.

Research in Denmark by Anders Humlum and Emilie Vestergaard links surveys about AI chatbot use with administrative labor records. Their March 2026 paper describes substantial changes in tasks and work practices, while finding no detectable average effect on earnings or recorded hours over the period they studied. That combination matters. Work can be reorganized without an immediate movement in the payroll figures.²

It also means we should resist using the study as a guarantee. The researchers observed particular occupations, in a particular country, over a defined early period. Denmark’s institutions and workplaces do not stand in for every labor market. An average can leave important differences between people. The absence of a detectable effect so far does not settle what a later stage will bring.

Other emerging research, including work using United States payroll records, has examined warning signs for younger workers in AI-exposed occupations. Those patterns deserve attention, especially where entry routes may be narrowing. They do not, by themselves, prove that AI caused every observed change. Hiring also responds to interest rates, business demand and earlier overexpansion. A responsible reading keeps the warning and the uncertainty together.³

This can feel unsatisfying. People want a clear answer: will there be work for me?

I would like the answer to be clear too. But certainty invented for comfort can lead you into a costly choice. Certainty invented for fear can do the same. If someone tells you that your whole field will vanish, you might abandon a useful skill too soon. If someone says nothing important will change, you might miss the time you had to prepare.

You can act without knowing the final shape of the economy. You can learn which tasks your workplace values, which new tools it is actually using, and whether there is a path from your present role into the work that remains. You can also notice when an employer’s reassuring language does not match its staffing decisions.

Attention is more useful than either panic or sleepwalking.

Start with a day you recognize

Think of a recent working day. An ordinary one, rather than the day of your greatest success or worst mistake.

What did you actually do? Include the work that never appears in a job description. Someone had to explain an unclear request. Someone remembered a promise. Someone calmed an angry customer, moved an appointment, caught a mistake or showed a new worker where to begin. If you work with your hands, include the judgment in the movement. If you work on a screen, include the conversation that made the screen work possible.

Now look at what has changed. You do not need to give every task a risk score. You need to see where the change touches you.

Diane might notice that drafting a routine reply has become easier, but she is spending more time checking whether the record is current. She might also notice that the organization counts sent messages and barely notices prevented mistakes. If management uses that count to set targets, her careful work could make her look slow.

That is a question about measurement and responsibility, not simply skill. Learning to use the tool faster will not solve it on its own. The employer needs to decide how checking will be staffed and how errors will be handled.

A self-employed translator faces another version. A client may ask for a lower price because a machine produced the first draft. Yet the translator may still have to check meaning, tone and errors throughout the document. The price of a first draft has changed; the amount of professional responsibility may not have fallen with it. The next decision concerns scope, rates and which clients will pay for reliable work.

A person in physical work can be affected without using AI at all. Scheduling, dispatch, purchasing and customer booking can change around the job. A new system may reduce wasted travel. It may also create a schedule with no room for the unexpected. The point is to find the change where it happens, rather than wait for a robot to arrive wearing your uniform.

You will not control every response. You may have little room to challenge a target or refuse a new duty. Even so, understanding the problem accurately helps you decide where support is needed. A skill gap calls for learning. An impossible workload calls for changes in the workload. An income gap calls for money and access to assistance. Treating all three as failures of attitude wastes time the person may not have.

Your worth is not waiting for a market forecast

There is a cruelty hidden in some cheerful advice about the future. It tells people to find the part of themselves that a machine cannot copy, then makes their right to a good life depend on staying ahead forever.

Learn. Of course. Be curious about the tools that can help you. But no person should have to win an endless contest against improving software to earn food, shelter or the respect of other people. A society capable of producing more with less labor has more room to protect people. Whether it uses that room well is a question about institutions and choices.

This is also why I do not want to defend every existing job as though it were a sacred human need. Some work gives people pride, friends and a place to grow. Some work wears out their bodies or uses up days they would rather spend elsewhere. Often the same job contains both. You can care about colleagues and still want to stop working nights. You can be proud of a skill and grateful that a machine removes its most tedious part.

We should be able to say that without being accused of laziness or indifference to people losing income.

The income loss is precisely the problem. Release from unwanted work is a very different event when you still have the means to live. If all the savings go to owners while workers lose wages, the technology’s promise has been divided unfairly. We should change the division rather than insist that the unwanted work continue forever.

In Who Gets the AI Dividend?, I argued for shared claims on the gains from AI. Here the concern is what those claims make possible in a person’s life. Time to train without losing a home. A shorter week without a smaller food budget. Room to care for someone. The ability to turn down work that harms you. Ownership and public services belong in this story because they can change the choices available at the kitchen table.

There is personal work to do as well. An income floor cannot decide what you care about. A new course cannot choose your friends. Even a good transition policy leaves a person facing questions about habit, identity and direction. We need to work on those questions without pretending they replace the need for material security.

Three different starting points

You may be reading while your job is still secure. The change is a rumor, a new tool or a question you have not wanted to ask. Your advantage is time, if you can use some of it. Find out how the work is changing. Try a tool on a low-risk task. Ask what training and time your employer will provide. Preserve a little room to consider an alternative before urgency makes every choice look the same.

You may already have lost work or income. Advice about long-term purpose can sound offensive when a bill is due. Begin with the essentials and the people or services that can help keep them covered. There will be time to think about identity, but you do not have to solve your whole future before asking for immediate support.

Or your job may remain while becoming less bearable. You have learned the new system. You have become more productive. The reward has been more work, closer monitoring or a thinner team. Your task is not to convince yourself that this must feel like progress. You need to understand which conditions could change, what support exists and what exit would actually require.

These starting points overlap. A person can have a job, lose income through reduced hours and carry most of the care at home. A worker can welcome a tool and fear what management will do with it. We should leave room for such mixed feelings. They often describe reality more accurately than the demand to be either an optimist or a pessimist.

At the end of Diane’s demonstration, she has not learned whether her job will exist in five years. She has learned enough to ask a better question at the next meeting. Who will be responsible for the replies that need checking, and how much time will they have?

That question reaches the work. It gives people something concrete to decide. It also begins to reveal whether the employer intends to improve a job or simply remove its slack.

A changing world will ask you to adjust. You are allowed to ask something of it in return. You can ask for the tools, time and security that make adjustment possible. You can ask that progress arrive in a form an ordinary household can use. And you can begin planning from the person you are now, without accepting the idea that a change in your market price is a verdict on your life.

Sources and limits