The First Career Rung Is Changing
Generative AI is changing the first rung of professional work. This series examines the seniorification of entry-level jobs: what happens when AI systems absorb tasks once assigned to beginners while employers still expect those beginners to exercise judgment, lead small pieces of work, and navigate ambiguity. The practical question is not only whether AI eliminates jobs. It is who now performs the work through which people become capable of doing more consequential work later.
That distinction matters because a job has always served two purposes at once. It produces something an organization needs, and it gives a less experienced person repeated contact with real problems, imperfect attempts, feedback, and gradually increasing responsibility. When AI makes the first purpose cheaper, the second can disappear unnoticed. An organization may see faster output and no immediate loss of productivity while losing part of the process that creates its next generation of experienced practitioners.
The first rung is more than a list of tasks
“Entry-level” sounds like a description of the work: simple assignments, limited responsibility, and close supervision. It is also a description of a position in time. A beginner is expected to start with bounded tasks and, through practice, become someone who can recognize errors, make trade-offs, and take responsibility for outcomes. The tasks are not all valuable because they are difficult. Some are valuable because they place the learner inside the real flow of work.
Generative AI changes that arrangement by making some of those tasks inexpensive to perform. Code generation, initial debugging, documentation, drafting, and routine analysis can often be produced quickly by a senior worker using an AI tool. The immediate decision is attractive: if a senior can complete the task with AI faster than they can explain it, delegate it, review it, and repair the result, why assign it to a beginner? But that calculation treats the task only as output. It does not count the experience the beginner would have gained by attempting it.
To understand the resulting labor market, it helps to separate AI exposure from automation. An occupation or sector is AI-exposed when the capabilities involved in its work are relevant to what AI systems can do. That does not tell us whether the occupation will shrink. It tells us that its tasks may be reorganized. Some roles may retain more expert work for people after AI handles the basics. Others may have their complex work automated while people remain with less expert tasks.
This is the distinction behind the two-track picture in PwC’s 2026 Global AI Jobs Barometer. A professionalised role is one in which AI takes on more basic work and people retain, or acquire, more expert responsibility. A democratised role moves in the opposite direction: AI takes on more complex work, leaving a larger share of less expert activity to people. Neither label is a moral verdict. A professionalised job may offer more autonomy and pay, but it may also impose a higher threshold on newcomers. A democratised job may become easier to enter or perform, but the route to expert judgment can narrow.
The central tension can therefore be stated simply:
The issue is not whether every beginner should be forced to perform work that a machine can already do. Some routine work is genuinely wasteful, and removing it can free people to learn faster. The issue is whether the organization replaces the developmental function of that work with something equally real: a bounded problem, a chance to make decisions, an opportunity to see consequences, and feedback that turns an attempt into competence.
Growth and access can move in opposite directions
The broad labor-market evidence makes the problem harder to reduce to either optimism or alarm. PwC analyzed more than one billion job advertisements across six continents, alongside firm and occupational data. Its results show strong signs of AI-enabled growth. In 2025, companies in the most AI-exposed quartile recorded 33.5% growth in turnover per employee from a 2018 baseline, compared with 24.0% among the least-exposed companies—roughly a 40% difference. The most-exposed group also showed faster headcount growth, about 52% compared with 36%, and faster wage growth, about 24% compared with 17%.
These figures are consistent with AI being used to amplify people and create capacity, not only to remove labor. AI-specialist job postings rose 68.9% from 2024 to 2025, while total postings rose 8.6%. The report also estimates an average advertised wage premium of 61.9% for postings requiring at least one AI-related skill. Employers are clearly placing a high value on some forms of AI capability.
But none of those findings establishes that AI caused the gains, or that the gains will be distributed evenly. The firm comparison is based on larger formal companies with the financial and headcount data needed for the analysis; firms that exited were excluded. The exposure measure is built from sector and occupational relationships, not a direct observation of how each company uses AI. The wage premium is a one-year comparison of advertised salaries and does not control for education, experience, location, or other factors. These are important qualifications, not technical footnotes. They determine what a worker or manager can reasonably infer from the numbers.
The same report finds a more uneven pattern at the start of the career ladder. In its early-career series, postings for jobs requiring zero to two years of experience in the highest AI-exposure quartile have flatlined relative to 2012, while lower-exposure groups continued to grow. The comparison uses data from Canada, Singapore, the United Kingdom, and the United States, and about 73% of the sample comes from the United States. PwC explicitly warns that the chart is not a causal test of AI: the weakening of exposed entry-level jobs began before ChatGPT’s public release, and broader economic changes and occupational differences may also explain part of the pattern.
Even with those limits, the combination is consequential. A labor market can show stronger productivity and growing demand for AI capability while offering fewer conventional starting points in the occupations most affected. The growth is real in the measured categories; access to that growth is a separate question.
There is also evidence that the surviving entry-level roles are being asked to carry more traditionally senior capabilities. In U.S. postings requiring zero to two years of experience, the most AI-exposed occupations had an average of 207 newly appearing skills per occupation in 2025, compared with 144 in the least-exposed quartile. PwC describes the most-exposed roles as seven times more likely to demand skills such as strategic decision-making, people management, stakeholder management, mentorship, and resourcefulness. Within the highest-exposure quartile, early-career roles classified as “seniorised” grew 35% from 2019 to 2025, while non-seniorised roles declined 10%.
That is not simply a story of disappearing jobs. It is a story of changing entry requirements. The labor market may be rewarding jobs that keep expert responsibility with people, but the people expected to perform that responsibility still need somewhere to acquire it. A job advertisement can reveal a higher threshold; it cannot, by itself, reveal whether the job supplies the practice and supervision needed to cross that threshold.
The missing question is how judgment gets built
The software-engineering study by Sumin Yu and Taesup Moon offers a close view of that missing question. It is not a representative survey. The researchers interviewed 14 people in South Korea: eight juniors who were students or recent graduates at the threshold of entering the workforce, and six software engineers with at least six years of experience. The sample was purposive, the interviews were cross-sectional, and South Korea’s unusually rapid AI adoption and changing recruitment system make it a revealing case rather than a universal template.
Its value is different from PwC’s. The labor-market report shows patterns across postings and firms. The interview study examines the mechanism participants experience underneath those patterns. Its central finding is called Absorption: generative AI expands senior engineers’ capacity, so work that once reached junior engineers moves directly into senior–AI workflows. Routine implementation, initial debugging, and documentation can be handled by someone who already has the context to judge the result. The beginner no longer receives the task, even if the organization sees no immediate decline in output.
One senior participant described a startup that had let many junior engineers go without a measurable productivity impact. Another said that juniors were increasingly not given the tasks through which earlier cohorts had learned. Yet the study also includes a crucial boundary case: a senior at a large enterprise said junior work and learning had changed little because formal onboarding and training structures continued to allocate work. The authors treat this contrast cautiously. Absorption appears more able to proceed where junior development is weak or implicit; it is not an unavoidable consequence of using AI.
This is where the economic and developmental meanings of entry-level work come apart. A routine task can be easy for a senior to automate and still be useful for a beginner. Writing a small change, testing it, seeing it fail, revising it, and having someone review the result builds more than an isolated technical skill. It builds a sense of what a plausible answer looks like, what can go wrong, when a requirement is ambiguous, and which shortcuts are unsafe. Those capacities later allow a senior to evaluate AI output rather than merely accept it.
The study’s participants describe the risk in terms of ownership and recognition. A learner can produce a good-looking result without understanding the concepts behind it. A stable grade or a completed ticket can therefore hide a weaker mental model. The cost may appear only when the person must work without the tool, verify a subtle error, or make a decision where there is no obvious answer. The very competence employers increasingly ask for—judgment under uncertainty—may be the competence that routine, error-prone work once helped develop.
That creates the defining paradox of seniorification. AI can remove drudgery and make some roles more expert. At the same time, it can remove the lower-stakes practice through which people learned to become expert. The first outcome is visible in productivity, job postings, and wages. The second is delayed, distributed across classrooms and workplaces, and harder to measure.
The question for organizations is therefore not simply, “What can AI do instead of a junior?” It is, “What will now give a junior the experience needed to become the person who can use AI responsibly on harder work?” Answering that requires looking more closely at the two tracks of the labor market, the tasks that are moving upward, and the kinds of failure and feedback that produce judgment. It also requires accepting that a higher-skill job description is not the same thing as a functioning career pathway.
If the first rung is disappearing, the important measure is not only whether today’s work gets done faster. It is whether tomorrow’s capable practitioners are still being formed.