The Narrowing Ramp: How AI Is Affecting the Traditional Career Ladder

Career Landscape Reports: 2 of 5 in a series

For most of the last century, the way you became good at a knowledge-work job was by doing the boring parts of it first. Junior lawyers reviewed documents. Junior consultants built slide decks and ran numbers. Junior analysts pulled data and wrote the first draft of the memo someone senior would rewrite. None of that work was glamorous, and companies didn’t pay for it simply because the output itself was valuable.

They paid for it in part because it was how a 23-year-old turned into a 33-year-old who could be trusted with judgment calls.

That traditional career ladder is beginning to break, partly because the routine work that traditionally served as an apprenticeship is also the work AI tools are increasingly able to perform.

The first piece in this series looked at the Jevons Paradox and the broader rebound effect: making a task cheaper can expand demand for the work built around it even when technology reduces the labor required for individual tasks. Software illustrates the distinction well. Employment prospects for software developers remain much stronger than for the narrower occupation of computer programmer, even as AI makes many coding tasks faster and cheaper.

But that is an aggregate story about occupations and markets. This piece is about what happens on the way there: what happens to a particular rung of the career ladder when the work historically assigned to that rung changes or disappears before organizations develop another way for people to acquire experience.

What the Numbers Actually Show

The Burning Glass Institute, a workforce research organization that analyzes job-posting data at scale, has documented a substantial decline in opportunities requiring relatively little prior experience in several AI-exposed fields. Between 2018 and 2024, the share of postings requiring three years of experience or less reportedly fell from 43% to 28% in software development, from 35% to 22% in data analysis, and from 41% to 26% in consulting.

The important point is not simply that some entry-level opportunities have declined. Hiring appears to be shifting toward people who already possess experience at the same time that some of the work through which people traditionally acquired that experience is being automated.

PwC’s 2026 Global AI Jobs Barometer provides especially interesting evidence of this shift. Its analysis of 2.4 million U.S. entry-level jobs found that junior roles highly exposed to AI are now seven times more likely than less-exposed junior roles to require capabilities traditionally associated with more senior workers, including leadership, strategic decision-making, creativity, and judgment. PwC describes this as the “seniorising” of entry-level work.

There is an important wrinkle: these seniorised entry-level jobs are not disappearing. PwC found that postings for them grew 35% between 2019 and 2025, while other entry-level roles declined by 10%. Across highly AI-exposed occupations as a whole, however, entry-level hiring has largely flattened.

That suggests something more complicated than “AI is eliminating entry-level jobs.” Some of the jobs are still there. What is changing is the capability employers expect people to bring with them when they arrive.

The broader market for recent graduates shows the same mixture of difficulty and adaptation.

ZipRecruiter’s 2026 Graduate Survey found that entry-level jobs make up a smaller share of available postings and attract more competition than they did several years ago. Yet 77.2% of the graduates it surveyed reported finding work within three months of graduation, up from 63.3% the previous year.

That sounds encouraging, and it is. But the same survey found that graduates were submitting more applications, receiving fewer offers, and accepting a wider range of work to gain a foothold.

Federal Reserve data cited by ZipRecruiter also complicates the bleakest version of the story. As of December 2025, young college graduates had an unemployment rate of 5.6%—well above the 3.1% rate for college-educated workers overall, but below the 7.8% unemployment rate for people their age without the same educational advantage.

This indicates the college degree has not suddenly become worthless, but the transition from education into professional work has become harder and less predictable.

Underemployment remains a serious problem as well. Research by the Burning Glass Institute and Strada found that 52% of recent four-year college graduates were underemployed one year after graduation, meaning they were working in positions that did not require a bachelor’s degree or make meaningful use of college-level skills.

Put these findings together and a clearer picture emerges. Young people are still getting jobs. AI-exposed fields are still hiring. Some new kinds of junior positions are even growing, but the ramp leading into expertise is getting steeper.

Why “We’ll Just Hire Experienced People” Doesn’t Actually Work

The logic behind reducing junior hiring is straightforward in the short term.

If an AI system plus one experienced employee can perform work that previously required an experienced employee and a junior one, eliminating or reducing the junior position can look like an obvious efficiency gain.

For an individual company making this year’s staffing decisions, it may be entirely rational. The problem appears when everyone does it.

Experienced workers do not simply appear in the labor market. They are produced by years of doing progressively more difficult work, receiving feedback, observing experienced colleagues, making mistakes with limited consequences, and gradually developing judgment.

The Burning Glass Institute calls the current disruption an “Expertise Upheaval.” AI is arriving alongside leaner post-pandemic staffing and an oversupply of graduates competing for a smaller number of conventional entry points.

The people companies hope to hire five or ten years from now are supposed to be gaining experience somewhere today.

If fewer organizations are willing to provide that experience, the labor market eventually faces a collective-action problem: every company wants to hire experienced people, while fewer companies want to bear the cost of producing them.

That is the structural problem hidden inside the short-term productivity gain, and it is not simply a question of headcount. It is a question of how expertise gets created.

A junior analyst who once spent hours assembling data learned something in the process about where the numbers came from, which anomalies mattered, which assumptions were fragile, and what senior people asked when they reviewed the result. A junior programmer debugging code encountered failure modes that eventually became intuition. A junior consultant revising a presentation for the fourth time began to understand why one argument persuaded a client and another did not.

Some of that work was inefficient. It was also educational. When AI removes the inefficient part, organizations have to decide whether they are also removing the learning that was embedded inside it.

Entry-Level Work Is Being “Seniorised”

This is where the PwC findings become particularly important.

The problem is not necessarily that organizations will have no use for early-career employees. It may be that they increasingly expect those employees to perform work that once came later in a career.

If AI produces the first draft, the human job becomes evaluating the draft.

If AI performs the preliminary analysis, the human job becomes deciding whether the analysis makes sense.

If AI generates the code, the human job increasingly involves architecture, integration, verification, security, and deciding what should have been built in the first place.

Those are valuable responsibilities. They are also responsibilities that traditionally depended on experience.

This creates a paradox for early-career workers: employers increasingly value judgment precisely when some of the tasks through which judgment was developed are disappearing. You cannot solve that problem simply by teaching someone to use an AI tool. AI fluency is becoming necessary. It is not the same thing as expertise.

The Response So Far Is Mostly Not a Response

Given the scale of the change, you might expect organizations to be aggressively redesigning jobs and developmental pathways around AI. Most are still early in that process.

Deloitte’s 2026 State of AI in the Enterprise research found that 84% of surveyed organizations had not redesigned jobs around AI capabilities, even though 36% expected at least 10% of their jobs to be fully automated within a year and 82% expected that level of automation within three years.

Fifty-three percent were focusing on educating employees to increase AI fluency. That matters, but it addresses only one side of the problem.

Teaching employees to use AI more effectively does not answer the harder question: if AI now performs some of the work through which novices historically became experts, what replaces that developmental process?

Some organizations are beginning to experiment with an answer. Instead of asking junior employees to create every first draft manually, they can ask them to evaluate AI-generated work, identify errors, compare alternatives, justify recommendations, and explain when the machine’s answer should not be trusted.

That shifts development away from pure production and toward judgment and verification;this only works if organizations deliberately teach those skills. Otherwise, there is a real danger of asking inexperienced people to exercise expert judgment before they have had enough opportunities to develop it.

Mentorship, apprenticeship, sponsorship, simulations, supervised decision-making, rotational assignments, and explicit feedback may therefore become more important rather than less important in AI-intensive workplaces.

The old system often produced expertise almost accidentally, as a by-product of doing large amounts of junior work. The new system may have to produce it deliberately.

What This Means If You’re Early in Your Career

In many AI-exposed knowledge-work fields, relying on the traditional path to seniority is becoming a worse bet than it was five years ago.

That does not mean there is no path. It means you may have to become more deliberate about acquiring experiences that used to arrive automatically with the job.

Seek work where you are not simply producing output with AI but evaluating it. Look for opportunities to explain why an answer is right or wrong, make decisions under uncertainty, defend recommendations, deal with exceptions, and see how experienced people think through problems.

Those are the activities from which judgment develops.

Pay attention to internships, apprenticeships, project work, and other opportunities to accumulate genuine experience before your first conventional full-time role. ZipRecruiter’s 2026 research found that working while in college substantially improved graduates’ chances of landing employment, and internships and apprenticeships remained particularly valuable entry points.

Treat mentorship and sponsorship as something you actively pursue rather than something you assume will happen because you have been hired.

Evaluate an “entry-level” job partly by what it will allow you to learn.

A position that gives you constant exposure to AI-generated output but little access to decision-making, feedback, customers, consequences, or experienced colleagues may be less valuable to your long-term career than its title or salary suggests.

The question is no longer simply:

Can I get this job?

It is also:

Will doing this job make me more capable of doing the next one?

What This Means If You Hire and Train Juniors

If you make hiring decisions, the short-term economics of reducing junior labor may be real. So is what you are trading away.

An organization that stops developing its own early-career people becomes increasingly dependent on a labor market populated by people somebody else trained.

That works for an individual employer only as long as enough other employers continue doing the training; it cannot be a stable strategy for an entire profession.

Organizations therefore need to distinguish between automating work and automating development.

If AI eliminates the repetitive task, ask what the employee was learning by performing that task and design another way to teach it.

If junior analysts no longer have to construct the first analysis, require them to interrogate one.

If junior programmers produce less code manually, expose them earlier to testing, architecture, failure analysis, and code review.

If junior professionals spend less time preparing basic material for senior colleagues, create more deliberate opportunities for them to observe how those senior colleagues reason.

The goal should not be to preserve inefficient work simply because it used to serve as training; it’s to preserve the development of expertise after the inefficient work disappears.

The Uncomfortable Part of the Optimistic Story

The first article in this series argued that greater AI productivity does not automatically mean fewer jobs. Lower costs can expand demand, create new applications, and increase the total amount of economically useful work.

That remains an important counterweight to simple predictions of mass technological unemployment, but aggregate growth does not guarantee that the traditional path into a growing field remains intact.

AI may simultaneously expand the amount of valuable work a profession can perform and narrow the pathway through which inexperienced people become capable of performing its most valuable work. That is the part of the transition we should be paying much more attention to.

The problem is not simply that the bottom rung of the career ladder may be disappearing; it may be that the distance between novice and expert is being compressed faster than our organizations, schools, and professions are redesigning how that distance gets crossed.

For people early in their careers, that means becoming much more intentional about acquiring judgment-producing experiences.

For employers, it means recognizing that expertise is not a resource you can indefinitely buy on the open market without somebody investing in its creation.

And for anyone trying to understand where AI is taking the labor market, it means looking beyond the number of jobs. We also need to ask how the people who will do tomorrow’s high-value work are going to learn how to do it.

I use AI tools to help research sources and draft these posts. I review, edit, and take responsibility for everything published here, including any errors.

---

Sources

This piece drew on the following research and reporting:

• The Burning Glass Institute, “No Country for Young Grads,” 2025. Research on deteriorating early-career employment outcomes and the “Expertise Upheaval” affecting the traditional transition from education into professional work. (The Burning Glass Institute)
• Burning Glass Institute and Strada Institute for the Future of Work, “Talent Disrupted.” Research finding that 52% of four-year college graduates are underemployed one year after graduation. (Inside Higher Ed)
• PwC, 2026 Global AI Jobs Barometer. Analysis of more than one billion job advertisements globally and 2.4 million U.S. entry-level jobs, including evidence that highly AI-exposed junior jobs are increasingly requiring traditionally senior capabilities. (PwC)
• ZipRecruiter, Graduate Survey 2026. Research on the shrinking share of entry-level postings, increased competition among graduates, employment outcomes, and the importance of work experience, internships, and apprenticeships. (ZipRecruiter Economic Research)
• Deloitte, The State of AI in the Enterprise, 2026. Research on the gap between projected automation and organizational redesign, including the finding that 84% of surveyed companies had not redesigned jobs around AI capabilities. (Deloitte)