The Labor-Market Problem Is Much Bigger Than “AI Is Replacing People”

AI is changing work. But the deeper problem is that the systems we use to connect people with work were already struggling to understand capability.

The public conversation about AI and employment has become dominated by one question:

Which jobs will AI replace?

It is an understandable question. It may also be the wrong place to begin.

AI is undoubtedly changing work. Some tasks can now be performed faster, differently or with far less human effort. Some roles will contract. New ones will appear. Others will change so substantially that the same job title may describe very different work five years from now.

But treating the future of work primarily as a contest between humans and AI hides a much larger problem.

The labor market was already struggling to connect people with opportunity.

AI did not create that problem.

It is exposing it.

Jobs are becoming a poor unit of measurement

A job title sounds reassuringly concrete.

Product manager. Accountant. Designer. Analyst. Customer service representative.

But a job is not one thing. It is a collection of responsibilities, tasks, decisions, relationships, constraints and capabilities.

MIT researchers studying AI and work make an important distinction: AI does not encounter occupations as indivisible units. It encounters particular skills and tasks within them. Some can be automated, some augmented, and others remain heavily dependent on human judgment. The work inside an occupation can therefore change long before its title disappears.

The International Labour Organization reached a similar conclusion in its global analysis of generative AI exposure. About one in four workers worldwide is employed in an occupation with some degree of exposure to generative AI, but the ILO concludes that transformation of jobs is more likely than wholesale replacement, because most occupations still contain tasks requiring human input.

That matters because much of our employment infrastructure still starts with the occupation rather than the work.

We advertise a job title.

We search for people who have held similar job titles.

We count their years of experience.

We look at where they worked.

We examine their education.

Then we try to infer whether they will be able to perform the work in front of us.

As work changes faster, that chain of inference becomes increasingly fragile.

We have confused history with capability

A CV is largely a record of previous access to opportunity.

Where did you work?

What title did someone give you?

How long were you there?

Which institution admitted you?

Which qualifications did you obtain?

Those things can contain useful information. But they are not the same thing as capability.

Someone may have ten years of experience while repeatedly doing a narrow range of work.

Another person may have gained relevant capability through freelance work, caregiving, voluntary work, self-directed learning, entrepreneurship, a different occupation, community leadership or simply solving difficult problems outside a conventional career path.

Yet hiring systems frequently treat the first history as legible and the second as risky.

This creates a circular problem:

Previous opportunity becomes the evidence required to obtain future opportunity.

That affects people entering the labor market for the first time, but it also affects career changers, returners and anyone whose experience does not follow the expected sequence of employers, titles and credentials.

AI did not create this circularity.

But a labor market changing at the level of tasks and capabilities makes it much harder to justify.

Even “skills-first” hiring has struggled to escape the old system

There is growing recognition of this problem.

In June 2026, the OECD published A Skills-First Labour Market, arguing that rapidly changing work is widening the gap between the skills employers require and what traditional qualifications alone can signal. The OECD describes skills-first systems as a way to reduce information asymmetry and improve access to employment. It also reports substantial skills mismatch across several major economies.

But changing the language of a job advertisement is not the same thing as changing the hiring decision.

Research from Harvard Business School and the Burning Glass Institute examined employers that removed degree requirements from roles. After those requirements disappeared, the share of workers without degrees actually hired into those positions increased by only about 3.5 percentage points.

The policy changed considerably more than the behavior did.

That tells us something important.

If an employer removes one proxy but leaves the rest of the decision system untouched, hiring does not automatically become capability-based.

We need a way to see capability.

The first rung of the ladder is also disappearing

The difficulty becomes particularly visible at the beginning of a career.

A 2026 study examining 243 million hires and 407 million job postings across the United States, United Kingdom, Canada and Australia found a substantial decline in junior hiring.

At first glance, this appeared to support the idea that generative AI was replacing inexperienced workers.

But the researchers found something more complicated.

Generative-AI exposure was strongly correlated with another major change in work: working from home. Once the researchers examined both effects together, much of the apparent relationship between AI exposure and declining junior hiring weakened, while the relationship with remote work remained.

Their conclusion was not that AI has no effect. It was that declaring AI responsible for the collapse in junior hiring is premature.

That distinction matters.

If we diagnose every labor-market problem as automation, we may build solutions for the wrong problem.

Perhaps part of what is disappearing is not simply the junior job.

It is the infrastructure through which inexperienced people used to become experienced.

More hiring activity does not necessarily create more opportunity

AI can also make the existing hiring machinery dramatically more productive without making it more effective.

MIT researchers studying AI-assisted job advertisements found that employers offered generative AI produced postings much faster. They were 19% more likely to post a job.

But the AI-written postings were 15% less likely to result in a hire.

The technology made it cheaper to produce a job advertisement. That encouraged some employers with little immediate intention of hiring to post anyway, while AI-generated descriptions could also become more generic and less informative to candidates.

This reveals a critical distinction:

More hiring activity is not the same thing as better labor-market matching.

A system can generate more vacancies, more applications, more screening, more messages and more data while becoming less effective at connecting a particular person with a particular piece of work.

And generative AI can operate on both sides.

Employers can generate job descriptions faster.

Candidates can generate applications faster.

Screening systems can process them faster.

Recruiters can send communications faster.

Soon we can have machines helping one side generate enormous quantities of material for machines on the other side to filter.

The numbers may increase dramatically.

The amount of useful information may not.

The deeper problem is evidence

Perhaps the most important question for the future of work is therefore not:

What job did this person have before?

Nor:

Can an AI predict whether this person will be successful?

It may be something much more straightforward:

What does this work actually require, and what evidence do we have that this person can do it?

That changes both sides of the employment relationship.

Employers have responsibilities too.

Before asking hundreds of people to invest time applying for an opportunity, an organization should be able to explain what work needs doing, what outcomes matter, which requirements are genuinely necessary, what constraints exist, how much autonomy the person will have, where AI will be used and how decisions will be made.

Candidates should then have ways to demonstrate relevant capability without requiring the correct employer, job title, degree or career chronology to have granted them legitimacy first.

And where capability cannot be established from existing evidence, we can create new evidence.

A realistic work simulation can tell us something a keyword match never will.

AI can help. It should not become another gatekeeper.

There is a temptation to solve the failures of hiring by placing a more sophisticated prediction engine on top of them.

Give AI every CV.

Give it every job description.

Ask it which people are best.

Rank them.

Score them.

Predict who will succeed.

But that risks preserving the assumptions of the existing system while making them faster, less visible and harder to challenge.

There is another possibility.

AI can help people and organizations understand complex information.

It can help translate between different descriptions of work and capability.

It can identify missing evidence.

It can help someone understand why an opportunity may be relevant.

It can help an employer express what work actually needs to be done.

But those are fundamentally different uses of AI from quietly deciding who deserves access to opportunity.

We need better opportunity infrastructure

This is the problem Adonea is exploring.

Not how to make recruiting faster.

Not how to process more applicants.

And not how to build a more powerful ranking algorithm.

We are asking whether the infrastructure connecting people with work can be redesigned around actual work, demonstrated capability, evidence, transparency and human decision-making.

That means defining the work before opening the opportunity.

It means treating evidence as something richer than a CV.

It means allowing uncertainty rather than converting everything into a score.

It means explaining why an opportunity appears relevant rather than claiming that an algorithm has identified the “best” candidate.

And it means recognizing that access to previous opportunity cannot continue to be the primary evidence required for access to future opportunity.

AI makes this work more urgent.

But AI is not the whole problem.

The deeper problem is that we have spent decades building systems that are extremely good at processing applications and surprisingly poor at understanding people.

The future of work will require something better.