The Real Reason AI Is Cutting Some Entry-Level Tech Jobs, Not Others

A Stanford study found AI is cutting entry-level hiring in some jobs and not others, and the split comes down to one mechanism.

You've applied to sixty entry-level software roles this cycle. Half went nowhere, a handful sent auto-rejections that mentioned "evolving hiring needs," and the one recruiter who did call you asked, almost in passing, "how comfortable are you reviewing AI-generated code versus writing it yourself?" That question is not small talk. It is the exact line Stanford researchers just drew through the entry-level tech market, and it explains why some of your peers are getting offers in roles that look identical to yours on paper while you are stuck refreshing your applicant tracker.

The short answer: AI is not cutting entry-level tech jobs evenly. It is cutting the ones where AI automates the core task, and leaving flat or growing the ones where AI augments it. A Stanford SIEPR study called "Canaries in the Coal Mine," reported 2026-08-25, put numbers behind this pattern for the first time, and it changes how you should be reading job postings this cycle.

What the Stanford study actually found

Stanford's Institute for Economic Policy Research (SIEPR) studied employment patterns in occupations with high exposure to AI tools, tracking outcomes by age group from late 2022 through mid-2025. Two findings stand out:

1. Employment for workers aged 22-25 in the most AI-exposed occupations fell about 6% over that period.
2. Workers 35 and older in the same occupations gained 6-9% employment over the same window.

The study also reported that entry-level hiring specifically, not just overall employment, fell 16% for the 22-25 age group in these occupations. The researchers themselves flag this hiring figure as only barely statistically significant, so treat it as suggestive rather than definitive. Even with that caveat, the direction is consistent across both measures: younger workers in AI-exposed jobs are losing ground while older workers in the identical occupations are not.

That's the "canaries" framing. Entry-level workers, who tend to do more of the routine, easily-automated slice of a given occupation's tasks, are the first to feel a shift that more senior workers, doing higher-judgment work in the same job title, mostly avoid.

The real split: automate versus augment

The mechanism Stanford's researchers point to is more useful than the headline numbers, because it gives you something to actually act on. Their core finding is that what AI does to a specific task predicts the hiring effect, not the occupation's label as a whole.

• When AI automates a task, meaning it can complete the task end-to-end with little human input, employment in roles built around that task tends to decline.
• When AI augments a task, meaning it supports a person's judgment, speeds up their problem-solving, or helps them check their own or someone else's work, employment in roles built around that task tends to hold flat or rise.

Software engineering is the clearest example, because it contains both kinds of tasks inside the same job title.

Notice that this is not "junior developer jobs are disappearing." It's that the specific slice of junior developer work that looks like pure code generation is the exposed part, while the slice that looks like review, debugging, and communication is not. Two candidates with the same job title, "Software Engineer I," can be sitting on opposite sides of this line depending on what their actual day-to-day looks like.

A quick way to test which side your target role falls on

Read the job description and count which verbs dominate:

• Automate-leaning verbs: "build," "write," "implement," "generate," with little mention of review or ownership beyond the task itself
• Augment-leaning verbs: "debug," "review," "own," "collaborate," "diagnose," "communicate with stakeholders"

No posting will say "this role is AI-exposed." You're reading the shape of the actual work between the lines, and asking about it directly in interviews, as the recruiter did in the opening example, is a completely fair question to ask back.

Why 22-25 year olds are the canaries

It's not that companies are targeting young workers. It's that the tasks concentrated in a typical entry-level workload, at any company, skew toward the automatable end: writing first drafts, doing repetitive implementation work, and building the kind of pattern-matching fluency that used to take years of junior-level reps. AI now does a meaningful share of that pattern-matching work directly. Senior engineers in the same companies spend more of their time on system design, cross-team tradeoffs, and judgment calls that current AI tools support rather than replace, which is consistent with the 35+ group gaining employment in the same occupations.

This also lines up with something international students on OPT already feel: the entry-level rung of the ladder is the one getting redefined in real time, right as your job-search clock is running. If you're on post-completion OPT, you have 90 cumulative days of unemployment to work with, and up to 150 cumulative days total if you use the 24-month STEM OPT extension — that clock does not pause for a tighter market, so the practical urgency of finding augment-leaning roles quickly is higher for you than for a US citizen peer who can wait out a slow season. Confirm your specific day count with your DSO before making any close-to-the-line decision.

What this means for your job search

1. Audit your target list by task shape, not job title. Two "Junior Backend Engineer" postings at two companies can sit on opposite sides of the automate/augment line. Read the actual bullet points.
2. Ask about it directly in interviews. "What does a typical week look like in terms of writing new code versus reviewing, debugging, or extending existing systems?" is a legitimate, well-informed question that also signals you understand the market.
3. Build visible proof of judgment, not just output. A portfolio of things you built with AI assistance is now table stakes. A portfolio that shows you catching a bug an AI tool introduced, or explaining why you rejected a generated approach, demonstrates the augment-side skill that the data says is holding up. For structured system-design practice that trains exactly this kind of judgment, see this system design interview prep guide for international candidates.
4. Don't assume "AI-proof" means "no AI in the stack." Almost every entry-level tech role touches AI tools now. The distinction is whether you're the one operating the tool with judgment or the one being substituted by it.
5. Recalibrate compensation expectations honestly. Roles concentrated in augment-heavy work, like review-focused or systems-integration positions, are not necessarily lower-paid than pure build roles. For a broader look at how new-grad tech pay actually breaks down by role type, see this new-grad tech compensation breakdown.
6. Widen your search past headline "software engineer" titles. Roles like QA/SDET, developer relations, technical support engineering, and implementation engineering skew augment-heavy by nature and are worth weighting higher than the raw job title might suggest. Our fuller entry-level tech playbook for new grads navigating AI-era hiring walks through this in more depth.

Common mistakes

• Treating "entry-level tech is dying" as the takeaway. The study documents a redistribution by task type and age, not a collapse of the field. Overreacting by abandoning software engineering entirely throws away the augment-side roles that are holding steady or growing.
• Assuming every rejection is AI-related. Most rejections still come from ordinary competition, mismatched experience, or an unclear resume. Don't let one study become an excuse that stops you from fixing fixable problems in your application.
• Leading interviews with "I'm worried about AI replacing me." It reads as anxious rather than informed. Ask specific, task-level questions instead, as shown above.
• Ignoring the statistical caveat on the hiring number. The 16% entry-level hiring decline is explicitly flagged by the study's own authors as only barely statistically significant. Cite the employment numbers (the 6% decline, the 6-9% gain for 35+) with more confidence than the hiring figure.
• Running your OPT clock on autopilot while you wait for the "right" role. If your search is taking longer than expected in an automate-heavy niche, that's a signal to pivot your target list toward augment-leaning roles sooner rather than later, given the fixed unemployment-day limits you're working within.

Frequently asked questions

What did the Stanford canaries in the coal mine study actually find? Stanford SIEPR's study, reported 2026-08-25, found employment for workers aged 22-25 in the most AI-exposed occupations fell about 6% between late 2022 and mid-2025, while workers 35 and older in those same occupations gained 6-9%. Entry-level hiring specifically fell 16% for the 22-25 group, though the authors flag that figure as only barely statistically significant.

Does AI replace junior developers? Not uniformly. The study's core finding is that AI automating a task, such as writing code end-to-end with little human involvement, correlates with hiring declines, while AI augmenting a task, such as supporting problem-solving or checking someone else's work, correlates with flat or rising employment.

How do I know if my target entry-level tech job is exposed to AI automation? Look at what the job description spends most of its words on. Descriptions dominated by producing first-draft code or routine scripts lean toward automate; descriptions emphasizing debugging, review, system design conversation, or verifying AI output lean toward augment.

Is this Stanford data reason to give up on a software engineering career? No. The study documents a redistribution within AI-exposed occupations by age and task type, not the disappearance of entry-level tech work. Weight your search toward augmentation-heavy roles rather than abandoning the field.

Where can I read the original Stanford SIEPR study? The policy brief, "Canaries in the Coal Mine," is published on the Stanford Institute for Economic Policy Research site. Check any secondhand summary, including this one, against that primary source for the full methodology and caveats.

If you're trying to figure out which roles on your target list actually lean augment versus automate, or you want a second set of eyes on how your resume signals judgment rather than just output, reach out to F1Jobs and we'll help you sort through it.