When AI Does the Work, What Happens to Us?

AI agents raise questions about jobs, responsibility and who benefits from greater productivity.

Imagine a business that grows without hiring another person.

Customer enquiries are answered. Invoices are checked. Reports are written. Software is updated. Much of that work is carried out by AI agents, with people handling exceptions and deciding what should happen next.

For some businesses, parts of this picture may become practical. How far it goes remains uncertain.

But it raises a question worth asking now: what happens to our working lives if businesses need fewer human hours to produce the same results?

Will agents replace jobs — or change them?

A job is usually a collection of tasks. Some are repetitive. Others depend on experience, relationships, physical presence or judgement.

An agent might take over one part of a job without replacing the person doing it. That could mean less administration and more time for useful work. It could also mean that an employer expects fewer people to handle a larger workload.

The ILO’s 2025 research estimates that one in four workers worldwide is in an occupation with some exposure to generative AI. It judges transformation more likely than complete replacement. That is an assessment of potential exposure, not a prediction that a quarter of jobs will disappear. Read the ILO research.

The outcome will depend partly on what organisations choose to do with the capacity they gain.

Will they improve services? Reduce working hours? Increase profits? Cut jobs?

Those are management choices with consequences for people.

Who is responsible when an agent gets it wrong?

Suppose an agent updates the wrong customer record.

It may have followed a plausible chain of reasoning. Every technical step may have worked. Yet the result is still wrong.

Who should have prevented it? Who notices? Who repairs the damage?

Calling software a “digital worker” does not answer those questions. Responsibility still needs to sit with identifiable people and organisations.

Before giving an agent authority, we should know what it may access, what it may change, when it must ask for approval and how its work will be checked.

We should also know how to stop it.

If managing agents requires constant supervision, we need to count that human effort when judging whether the technology is helping.

What happens to the people starting their careers?

There is another question I find difficult to dismiss.

Many people learn their profession by doing relatively routine work. They prepare the first draft, check the figures, answer straightforward questions and gradually develop judgement.

What happens if agents take over much of that work?

Perhaps AI will become an excellent tutor, helping people learn faster. Perhaps organisations will create better routes into skilled work.

But that will require deliberate effort. We should not assume that removing junior tasks automatically creates experienced people.

A business might improve this year’s efficiency while weakening the way it develops next year’s expertise.

If wages shrink, who pays the tax?

Now imagine that an economy produces more, but a smaller share of its income reaches households through wages.

There could still be plenty of wealth. The question is where it goes — and how people share in it.

If the gains become higher profits or investment income, governments may need to reconsider how they raise revenue and support public services.

That does not automatically mean taxing each robot or agent. IMF economists have argued against a specific AI tax, pointing to practical difficulties and possible effects on innovation. They discuss stronger taxation of capital income and better support for people affected by the transition. Read the IMF discussion.

The wider question is simple: if paid employment becomes a less reliable way to distribute income, what other arrangements will we need?

And if businesses can produce more, how will households have the purchasing power to benefit?

What would a better working life look like?

There is an optimistic possibility here.

AI could reduce tedious work, help small organisations do more, and give people more time for care, creativity and learning. Greater productivity could support shorter hours and better services.

But those benefits will not distribute themselves.

We will need to ask who has access to the technology, who owns it, who bears the risks and who gets a say in how it is used.

That is part of what draws me to experimental work on systems such as Tsunora and Agent Control: exploring how delegated digital work could have clear boundaries, accountable ownership and evidence of completion.

Technical systems can help us manage the work. They cannot decide what kind of society we want.

We do not need to believe that all jobs will disappear to start this conversation. We only need to recognise that changing how work gets done can also change how people earn, learn and find a place in society.

If AI allows us to produce more with less human effort, what should we choose to do with the time and wealth we gain?

Also published on Medium ↗.

This article reflects the evidence and development status at its original publication date.