Career & Jobs
What Happens to Entry-Level Jobs When AI Does the First Draft
By Jim Vernon, Editor, AI Intelligence International · Published 16 January 2026 · Reviewed against our editorial standards · About the author
The traditional entry-level job was an apprenticeship disguised as production. Juniors produced the low-stakes output — first drafts, research summaries, data cleanup — and learned the craft by having it corrected. That output is exactly what a model now produces for pennies.
The consequence is not that junior jobs vanish. It is that the training mechanism embedded inside them breaks, and organisations have to rebuild it deliberately or lose their pipeline.
Key takeaways
- The apprenticeship problem: Consider a law firm.
- What junior roles look like now: The emerging shape is fewer juniors, hired later, with a broader remit.
- How to compete for one of the remaining seats: Build artefacts that show the review skill.
- What hiring managers should change: Assume every take-home is model-assisted, and design for it.
The apprenticeship problem
Consider a law firm. Juniors historically learned by drafting hundreds of routine documents and reading the partner's edits. Take that volume away and the junior arrives at year three without the pattern library that made year-three lawyers useful.
The same structure appears in agencies, accounting practices, journalism, and software. Production was the curriculum. When production is automated, the curriculum has to be replaced with something intentional: structured review work, supervised judgement calls, deliberate exposure to messy cases.
Firms that notice this early build review-first junior roles. Firms that do not will discover the gap in about four years, when the mid-level bench is empty.
What junior roles look like now
The emerging shape is fewer juniors, hired later, with a broader remit. Instead of one task family, a junior is expected to run a small end-to-end slice: gather the input, direct the tool, check the output against a standard, and escalate the ambiguous cases.
That is a harder job than the old one. It requires the judgement that used to be developed over two years, on day thirty. It is also more interesting, and it pays better where firms have adjusted their bands.
For candidates, the practical implication is that demonstrating judgement matters more than demonstrating diligence. Nobody is impressed by volume any more, because volume is free.
How to compete for one of the remaining seats
Build artefacts that show the review skill. Take a piece of machine output in your target field, mark it up with what is wrong and why, and publish the annotated version. It demonstrates domain knowledge, standards, and the ability to work with the tools in one document.
Do the same with a small end-to-end project: define a question, gather real data, produce an answer, and write a short honest section about what you are not confident in. That last section is what separates a candidate who understands the work from one who has produced a nice-looking deck.
Applications themselves are now cheap to generate, which means volume applications have stopped working. Five tailored applications with a relevant artefact beat two hundred generated ones by a wide margin.
What hiring managers should change
Assume every take-home is model-assisted, and design for it. The interesting question is no longer 'can you produce this' but 'can you critique this'. Hand the candidate a plausible but flawed deliverable and ask what they would change before it goes to a client.
Replace the volume apprenticeship with a review apprenticeship: pair juniors with reviewers, make the standard explicit in writing, and rotate them through the messy cases rather than shielding them.
And be honest in the job description about how much of the role is direction and review. Candidates who expect to produce and find themselves editing all day leave within a year.
The wage picture
Two forces pull in opposite directions. Fewer junior seats push starting salaries down through competition. Higher expectations at entry push them up, because the role now requires judgement that used to be a year-two capability.
In practice the market is splitting. Firms that redesigned the role pay more for fewer, stronger juniors. Firms that simply cut headcount and kept the old job description pay the same and get worse candidates each cycle.
For a candidate, this means employer selection matters more than it used to. Ask in the interview how juniors learn there. A firm without an answer is a firm where you will stagnate.
Tools that help at this stage
Run your CV through the resume score to see whether it reads as production-focused or judgement-focused; most junior CVs are lists of tasks, which now reads as a list of things a model does.
Use the interview answer builder to convert your projects into decision stories with a structure an interviewer can follow, and the job exposure score to sanity-check which entry paths are compressing fastest before you commit two years to one.
Frequently asked questions
Are internships still worth doing?
Yes, and they matter more than before, because supervised exposure to real cases is the scarce resource. Choose the internship with the most review contact, not the most prestigious logo.
Will graduate schemes disappear?
Not broadly, but they are shrinking and getting more selective in exposed professions. Sectors with regulated sign-off and physical work are largely unaffected.
Should I hide that I used AI on an application task?
No. Declare it and show your review. Hiding it fails the honesty test; using it well demonstrates exactly the skill the role now needs.
Is a master's degree a good hedge?
Only if it grants a licence, a lab, or a network you cannot otherwise reach. A general taught master's rarely changes an exposure profile.