Partly, and unevenly, in a pattern that is more specific than the general worry suggests. Automation is not taking entry-level work because it is entry-level. It is taking work that is well-specified, repetitive and verifiable, and a disproportionate share of that happens to sit at the start of careers. Where a first job involves physical presence, judgement under uncertainty or dealing with people, it is largely untouched.
The employment data separates these cleanly. Tellers are forecast to decline 13 percent through 2035 and bookkeeping clerks 6 percent, while medical assistants grow 13 percent and HVAC technicians 11 percent. All four are entry points.
Key points
- The property that matters is task shape, not seniority. Specified, repetitive and checkable work automates; ambiguous, physical and interpersonal work does not.
- Declining entry points cluster in clerical work: tellers at minus 13 percent, bookkeeping clerks at minus 6 percent, information clerks at minus 2 percent.
- Growing entry points cluster in hands-on and care work: medical assistants at 13 percent, HVAC at 11 percent, paramedics at 6 percent.
- The wider labour market is still growing, by about 5.9 million positions to 2035.
- The real problem is training pipelines, not total jobs. Junior work was how people learned.
Which entry-level work is actually exposed
Sorting by task type rather than by industry produces a much clearer picture than any list of at-risk jobs.
Data entry and transcription. The purest case of all three properties at once: structured input, defined output, and a result anyone can verify. Largely automated already.
Routine clerical processing. Filing, form handling, standard queries with documented answers. Heavily automated, which is why the clerical categories decline across the board.
Basic content production. Writing short copy to a brief, resizing assets, producing standard reports. Compressed sharply, and this was a common creative entry point.
Simple coding tasks. Well-specified functions, small bug fixes, adding tests for existing behaviour. This is what models handle best, and it happens to be exactly how developers used to learn what a codebase was made of.
First-line support. Password resets, access requests, documented troubleshooting. Absorbed by self-service, which is why computer support specialists are forecast to fall 3 percent.
Now the other side.
Anything physical. Trades, healthcare delivery, logistics handling, food service, construction. Unstructured environments defeat robotics comprehensively.
Anything with a person in distress. Care work, emergency response, complaint handling, teaching. The interpersonal part is the job rather than an accompaniment to it.
Anything requiring judgement about incomplete information. Diagnosis, triage, investigation, anything where the description of the problem is unreliable.
Anything with legal accountability. Where a named person is required to sign, the role persists regardless of what a system is technically capable of producing.
The real problem: how people learn
This is the part that matters more than any count of jobs, and it is genuinely unresolved.
Historically, someone entered a profession by doing its simplest work. A junior developer wrote small well-defined functions. A junior analyst pulled reports. A junior editor checked spelling. A junior lawyer reviewed documents. None of that was the valuable part of the profession, and all of it was how the valuable part got learned. You saw a hundred small pieces and gradually understood how they fitted together.
That tier is precisely what automates well, and its removal creates a problem nobody has solved. The senior people doing the judgement work today learned it by spending years on tasks that no longer need doing. If those tasks are gone, the question of how the next generation acquires the same judgement is open, and hand-waving about it does not help anyone deciding what to study.
Several partial answers are emerging. Entry through a domain rather than through a function, so someone arrives knowing an industry and learns the craft second. Entry through roles that touch the work without being the work, such as support, operations or customer-facing positions. And deliberate apprenticeship-style arrangements, which are expensive and rare but genuinely address the problem.
None of these is a full substitute for a large tier of junior work, and organisations that removed it are mostly not yet feeling the consequence, because their senior people are still the ones who trained the old way. The cost of that decision lands about five years after it is made, which is exactly the interval that makes it easy to keep making.
Why organisations removed junior work faster than they meant to
The mechanism here is worth understanding, because it explains why this happened quickly and why it may partially reverse.
No employer decided to stop training people. What happened is that each individual decision made sense on its own. A manager with a well-defined task could either brief a junior, wait, review the work and give feedback, or produce it directly in ten minutes. The second option is faster on that task, and nothing in the moment prices the training that did not happen.
Multiply that by every manager and every task over two years and a training pipeline disappears without anyone deciding to remove it. This is a familiar pattern in organisations: a cost that is diffuse, delayed and borne by someone else gets optimised away, and the consequence arrives later as a shortage nobody can trace to a decision.
The reason it may partially reverse is that the shortage does eventually arrive. Organisations that stopped hiring juniors in 2024 will be short of mid-level people in 2029, and the cost of buying them on the open market is considerably higher than the cost of having trained them. Several sectors have been through exactly this cycle before with apprenticeships, and the pattern is that firms cut training in a downturn, discover the gap five years later, and rebuild at greater expense.
That is cold comfort to someone looking for a first job today. But it argues against treating the current situation as permanent, and it suggests that the organisations rebuilding a training route first will be the better ones to join.
What to know before deciding
Several practical points follow for anyone choosing a first career or advising someone who is.
Total jobs are not falling. The Bureau of Labor Statistics projects employment rising from 170.3 million to 176.2 million between 2025 and 2035, an increase of about 5.9 million. The composition is shifting, not the total.
Growth is concentrated in healthcare and trades. Both are hands-on, both are demographic rather than technological in their drivers, and both have entry routes that do not require a degree.
Exposure is not a prediction. The Bureau publishes AI exposure categories for 831 occupations and states plainly that exposure “does not imply job loss, productivity gains, automation probability, or wage effects.”
Certification is doing more work than it used to. Where a qualification is required to hold a role, that requirement is written into rules and does not soften as tools improve. Fields with formal entry standards have held up notably better than those without.
Location and licensing matter more than national figures. Many of the growing entry points are locally regulated, and the practical availability where you live differs from the national picture.
Entry routes that look robust
- Skilled trades with apprenticeship routes. Plumbing, electrical, HVAC. Growing, well paid, and physically resistant to automation.
- Direct healthcare roles. Medical assisting, paramedicine, therapy support. Strong demographic demand and hands-on delivery.
- Roles with licensing attached. Where the law names a qualified person, the position persists.
- Domain-plus-function combinations. Learning an industry alongside a skill, rather than the skill alone.
- Anything where the customer is present. Physical presence with a person remains stubbornly hard to remove, and it is undervalued relative to how reliably it holds.
- Roles that generate a training record. Positions in regulated fields where hours are logged and signed off tend to survive, because the qualification structure requires someone to do the work.
Decision framework
Five questions for anyone choosing a first job now.
- What is the task shape? Specified and repetitive is exposed. Ambiguous, physical or interpersonal is not.
- Is there a licence or certification? Formal entry requirements are the strongest structural protection available.
- How will you learn the senior version? If the junior tier is thin, ask directly what the path looks like rather than assuming one exists.
- Does the field face demographic or technological pressure? Healthcare and trades are pulled by demography, which does not reverse.
- Are you learning a domain or only a tool? Tool knowledge depreciates. Understanding an industry does not.
Learning to work alongside these systems, and to catch where their confident output is wrong, is itself a durable skill that transfers between fields. If you want a structured route in, explore Coursiv AI lessons and check current plan details on the official site.
Your next step
One more thing worth doing before you commit anywhere. Ask in an interview how the organisation trains people into the next level, and listen for whether the answer is specific. A firm that can describe its route, name someone who took it recently, and say what the first two years actually contain has thought about the problem. A firm that talks about learning on the job without being able to say what that means has probably not, and the difference matters more now than it did when the junior tier taught people by default.
Take the job you are considering and write down what a typical hour actually involves, in tasks rather than in the title. Then mark each task as specified and repetitive, or ambiguous and physical or interpersonal.
That ratio predicts exposure far better than any list of at-risk professions, because automation follows task shape rather than job labels. It also tells you which part of the role to get good at, which is more useful than knowing whether the whole thing is safe.