What happens to your job when artificial intelligence becomes capable of doing much of the work itself? Clara Shih, a former AI executive at Salesforce and Meta, believes the answer may be less about individual occupations disappearing and more about how workers are divided in an increasingly automated economy.

Her prediction is stark. She says the workforce could increasingly split into three broad groups: people who build AI, people who use AI and people who are managed by AI.

The third category already exists, she argues, but is now spreading into white-collar work.

Shih made the comments in a recent conversation with CBS News’ 60 Minutes while discussing the impact of rapidly advancing AI on jobs, particularly entry-level employment.

She stressed that nobody has a “crystal ball” for exactly how the transformation will unfold. But after spending years building AI systems at major technology companies, she said the changes she witnessed firsthand have made her increasingly concerned about what happens to workers at the beginning of their careers.

The three groups Shih sees emerging

Shih describes the future workforce in three broad categories.

The first group consists of people who build AI models and systems. These workers develop the underlying technology, train models, build AI infrastructure and create the products that other businesses and consumers use.

According to Shih, this group is likely to remain highly valuable and could command exceptionally high salaries because relatively few people possess the expertise needed to build increasingly sophisticated AI systems.

The second group consists of people who use AI to perform their jobs.

This could include lawyers using AI to review documents, marketers generating and analyzing campaigns, software developers working with coding assistants, financial professionals using AI for research and workers across many other industries incorporating AI into everyday tasks.

This sounds like the safest position to occupy.

Shih, however, believes this middle category may be smaller than people imagine and could shrink as AI systems become more capable.

The third group consists of people whose work is effectively managed by AI.

That does not necessarily mean the person is replaced by a machine.

Instead, an algorithm or AI system can assign tasks, determine the pace of work, monitor performance and evaluate results.

Shih points to gig-economy workers as an early example. Uber drivers and DoorDash delivery workers may technically be employed by or contract with human-run companies, but algorithms can determine where work comes from, how it is prioritized and how performance is measured.

She believes that model is increasingly moving into office jobs.

AI is beginning to manage white-collar workers

The idea that AI will “manage” workers is more consequential than the familiar prediction that chatbots will replace repetitive tasks.

Customer-support representatives, for example, may increasingly receive cases according to automated systems that decide which employee handles which interaction.

Sales development representatives can similarly be assigned leads, follow-up schedules and performance targets by software.

An AI system may also analyze how quickly a worker completes tasks, how customers respond and whether performance meets a predefined target.

The result is a workplace in which the employee remains human but much of the management layer is automated.

Shih argues that this distinction is frequently overlooked in discussions about AI and jobs.

People tend to imagine only two possibilities: humans doing work or AI doing work.

Her third category introduces a different model in which humans continue to perform the work but increasingly operate under algorithmic supervision.

Why Shih is particularly worried about entry-level workers

The most immediate impact, she says, is falling on people trying to enter the workforce.

Entry-level positions often involve exactly the tasks that AI can perform quickly: drafting routine communications, summarizing information, conducting basic research, organizing data, preparing standard reports and handling repetitive customer interactions.

For decades, these jobs also served another purpose.

They were training grounds.

A recent college graduate might begin by doing relatively simple work before gradually taking on more complicated responsibilities and eventually becoming a manager or specialist.

If AI absorbs the beginner tasks, workers may lose the opportunity to acquire that experience.

That creates what Shih describes as a disappearing “bottom rung” of the career ladder.

A person cannot easily move into a mid-career job without first gaining the skills that entry-level positions traditionally provide.

Shih says she saw the change inside Big Tech

Her warning carries additional weight because it is based partly on what she says she witnessed while working in AI.

Shih spent years at Salesforce before moving to Meta, where she led the company’s Business AI group. She later left Meta and founded the New Work Foundation, a nonprofit focused on helping young workers navigate an AI-driven labor market.

During her time building AI products and businesses, Shih says companies discovered that some new AI systems were “so much more capable right out of the gate” than many entry-level candidates they could otherwise hire.

That realization changed hiring behavior.

Rather than hiring large numbers of junior workers to perform tasks that could increasingly be automated, companies could deploy AI systems immediately.

The economic incentive is obvious.

An AI system does not need a salary, benefits, paid leave or years of onboarding.

But the wider social consequence is less obvious.

When companies stop hiring beginners, they may also reduce the number of people who eventually develop into experienced professionals.

The data behind the Gen Z job squeeze

Shih also points to labor-market data showing that young workers are being hit particularly hard.

Her New Work Foundation says roughly one in two recent US graduates are unemployed or underemployed in jobs that do not require their degree.

That div combines unemployment with underemployment, so it does not mean that half of all recent graduates are literally without jobs.

Independent research paints a similar, though less dramatic, picture.

Federal Reserve Bank of New York data cited by Stanford researchers show that unemployment among recent college graduates remains higher than the unemployment rate for the overall workforce, while a large share of recent graduates are employed in jobs that do not require a college degree.

The New York Fed’s data are also useful because they show that the problem predates some of the most recent AI advances.

That means AI should not be treated as the sole explanation for the weak entry-level market.

A combination of slower hiring, economic conditions, changing employer expectations and automation can all influence opportunities for young workers.

The entry-level hiring slowdown is real

Recent labor-market analysis has found that America’s youngest workers are facing an unusually difficult environment.

A March 2026 analysis in Fortune described the entry-level job market as the weakest in 37 years, while researchers cited New York Fed data showing elevated unemployment among young workforce entrants.

The problem is especially visible in technology.

Software development, information services and other fields that traditionally provided attractive entry-level jobs have become more difficult for new graduates to enter.

AI is part of that story because companies can automate portions of junior-level work.

But the technology may also be amplifying an existing hiring slowdown rather than creating the entire problem from scratch.

Why “learning on the job” could become a problem

One of Shih’s deeper concerns is not simply job losses.

It is the loss of career pathways.

Imagine a young employee starting in an administrative, research or customer-service role.

Traditionally, that worker performs basic tasks, learns how the organization operates, makes mistakes, receives feedback and gradually handles more complicated responsibilities.

AI can perform many of those beginner tasks.

That sounds efficient from the company’s perspective.

But it creates a potential problem for the worker: how do you become experienced when the jobs designed to give you experience disappear?

This could produce a strange labor market in which businesses have plenty of senior professionals but an increasingly thin pipeline of people capable of replacing them.

Does this mean all AI users are safe?

No.

That is another point at the heart of Shih’s argument.

Simply using AI at work does not guarantee that a person belongs to the relatively secure middle category.

A worker might use AI every day while still having an algorithm determine their workload, evaluate their performance and decide which tasks they receive.

In that case, the worker is technically an AI user but functionally belongs to the group being managed by AI.

The distinction is about control, not merely whether someone has opened ChatGPT or another AI tool.

What jobs are most exposed?

Shih does not offer a definitive list of professions that will disappear.

Instead, her framework focuses on tasks and power structures.

Jobs involving large amounts of predictable, repeatable digital work are generally easier to automate than roles requiring physical dexterity, complex interpersonal interaction, high-stakes judgment or accountability.

But even jobs considered “safe” can change.

An accountant may still be required, but could oversee AI-generated financial analysis.

A lawyer may still be needed, but could review AI-produced documents rather than prepare every document from scratch.

A customer-service representative may remain, while AI handles the initial conversation and determines which cases require a human.

The result could be fewer workers doing more work with AI assistance.

That could raise productivity, but it could also reduce the number of entry-level positions available to people trying to enter an industry.

Which group does Shih believe is most desirable?

In her framework, the people who build AI occupy the top tier because they control the technology itself.

The second group, people who use AI, remains important but is shrinking.

The third group, workers managed by AI, has the least control over how work is assigned and evaluated.

Shih’s concern is that many people assume they will automatically belong to the second group.

She thinks that assumption is dangerous.

As AI becomes more capable, some workers who initially use AI as a productivity tool may eventually find that the same systems are determining their tasks and measuring their output.

Shih’s warning about cities and manufacturing

The former AI executive believes the impact could eventually extend far beyond software and office work.

“Without intervention, what happened to manufacturing will happen across every major city in America,” she warned.

The comparison refers to the long-term transformation of manufacturing employment through automation, globalization and industrial restructuring.

Shih’s concern is that AI could create a similar disruption in occupations that have traditionally supported urban middle-class employment.

The difference is speed.

Industrial automation and globalization unfolded over decades.

AI systems can be deployed globally through software almost instantly.

That could compress the transition and make it harder for workers and educational institutions to adapt.

Shih is not arguing that AI should be stopped

Her position is more complicated than simple opposition to automation.

Shih remains optimistic about AI’s potential and has built a nonprofit around the idea that young people should learn to use the technology rather than compete against it blindly.

The New Work Foundation launched in 2026 to help Gen Z workers navigate an AI-shaped economy.

Its tools include career information, mentoring and AI-focused resources intended to help graduates identify opportunities rather than simply react to automation.

The organization’s core argument is that AI itself can be part of the solution.

Workers can use AI to become more productive, learn new skills and compete in a labor market that is changing rapidly.

What should students and workers do?

Shih’s framework points toward a practical lesson: learning how to use AI may become as important as learning the underlying professional skill.

A programmer who can direct AI coding systems effectively may have an advantage over someone who ignores them.

A marketer who knows how to work with AI-generated analysis could potentially accomplish more than one working without those tools.

But workers also need something AI cannot simply supply on demand: judgment.

Understanding a field deeply allows a person to identify when an AI system is wrong, incomplete or making an unreasonable recommendation.

That is especially important for young workers.

If AI performs every beginner task from day one, new employees may have fewer opportunities to develop the underlying judgment needed to supervise AI later.

The biggest risk may be a weaker career ladder

That is ultimately what makes Shih’s prediction more troubling than the usual “AI will take your job” headline.

The biggest disruption may not be millions of occupations disappearing overnight.

It may be the gradual disappearance of the early-career jobs that teach people how to do those occupations.

A society can replace a task with software.

Replacing the human career pathway built around that task is much harder.

Shih’s three-category model is therefore less a prediction that particular professions will survive than a warning about who controls work in the AI economy.

Some people will build the machines.

Some will use them.

And an expanding group may find that the machines are deciding what they do.

The crucial question is whether workers, employers, schools and policymakers can reshape the system quickly enough to ensure that the third category does not become the default destination for everyone who cannot afford to become an AI specialist.