Playbooks
AI's Impact on Entry-Level Developer Hiring: What the Data Actually Shows
Employment of workers aged 22-25 in AI-exposed occupations sits about 19% below where it would be had it kept pace with less-exposed peers, according to Stanford's Digital Economy Lab — and that gap has widened steadily since researchers first documented it in August 2025. Experienced workers in the same occupations show no comparable decline. Critically, the researchers are explicit that this is descriptive evidence, not proof of causation — "canaries in the coal mine," in their own words, not a causal estimate. Separately, US Federal Reserve Bank of New York data shows recent computer science and computer engineering graduates facing meaningfully higher unemployment rates than the general new-graduate population.
Last updated Aug 21 · 18 min read
The core finding, and what it actually claims
The single most rigorous, most-cited data point in this space comes from Stanford's Digital Economy Lab, in research titled "Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence." The headline finding: employment of workers aged 22-25 in AI-exposed occupations is roughly 19% below where it would be had it kept pace with the employment trend of their less-exposed peers (Stanford Digital Economy Lab). That divergence has widened steadily since the researchers first documented it in August 2025, meaning this isn't a one-time blip — it's a trend that has continued to grow.
The critical caveat, stated directly by the authors themselves rather than added by later commentary: these are "early, descriptive indicators — canaries in the coal mine — rather than causal estimates" (Stanford Digital Economy Lab). That distinction matters enormously for how this data should be used. The research establishes a strong, real, and growing correlation between AI exposure and reduced young-worker employment — it does not, by the authors' own framing, establish definitive proof that AI is the singular cause, as opposed to other contributing factors (a broader tech hiring slowdown, interest-rate-driven cutbacks, post-pandemic over-hiring corrections) that could be operating alongside it.
The mechanism: fewer hires, not more layoffs
One of the more specific and useful findings is about *how* the decline shows up. It operates primarily through reduced hiring of young workers, rather than through increased separations or layoffs (Stanford Digital Economy Lab).
This distinction changes how a new graduate or junior developer should think about their own risk profile. If the effect were driven by layoffs, an existing junior employee already at a company would be the one most at risk. Because it's driven by hiring instead, the people most affected are those trying to get into the field in the first place — the effect is concentrated at the front door, not inside companies that have already hired someone.
Substitution roles vs. complementary roles
Not every AI-exposed role behaves the same way. The Stanford research draws a specific distinction between two categories (Stanford Digital Economy Lab):
– "Substitution" roles — where AI directly replaces tasks a human previously did. Employment in these roles shows the decline described above.
– "Complementary" roles — where AI assists a worker rather than replacing their tasks outright. Employment in these roles stays flat or even increases, especially for experienced workers.
Applied to software development specifically, this maps onto a distinction the industry itself is starting to make explicit: a junior role that consisted mostly of routine, bounded tasks (writing boilerplate, fixing straightforward bugs, writing basic test scripts) looks structurally more like a "substitution" role than one where a developer is doing judgment-heavy design, architecture, or ambiguous problem-solving work that AI tools don't yet perform reliably on their own.
What separate labor-market data shows for CS/CS-engineering grads
Independent of the Stanford AI-exposure research, US Federal Reserve Bank of New York data on recent graduate unemployment shows a specific and notable pattern: computer engineering graduates face a 7.5% unemployment rate and computer science graduates face 6.1%, compared to an overall US unemployment rate of 4.3% — both substantially higher than fields like nursing (1.4%) or elementary education (1.8%) (CIO).
This is a separate dataset from Stanford's AI-exposure research, using a different methodology (general graduate outcomes tracking rather than occupation-level AI-exposure modeling), but it points in the same direction: something specific is happening to new graduates in tech-adjacent fields that isn't happening to new graduates broadly. Treat these as two independent, mutually reinforcing signals rather than the same finding reported twice.
The industry narrative: why companies say hiring is changing
Industry reporting captures how hiring managers and engineers themselves are talking about this shift, which is useful context even though it's practitioner opinion rather than rigorous labor economics:
– One senior software engineer, quoted in industry reporting, framed the calculus bluntly: "Why hire a junior for $90K when GitHub Copilot costs $10?" (CIO).
– An AI company CEO/CTO predicted that hiring freezes for new developers would become common, with future junior-level roles shifting toward overseeing AI-generated code rather than writing it from scratch.
– A head of AI at a consulting firm pointed specifically at the kind of work juniors traditionally did — fixing bugs, writing test scripts, cranking out boilerplate — as exactly the category of task current AI tools now perform.
Separately, a Resume.org survey of 1,000 US business leaders found 6 in 10 companies said they were likely to lay off employees in 2026, and 4 in 10 planned to specifically replace workers with AI by then (CIO). This is self-reported executive sentiment, not verified outcomes — a genuinely useful signal of intent and mood in the industry, but not the same evidentiary weight as the Stanford employment data above.
Practical examples
Real, sourced example — the boilerplate/bug-fixing task shift. Industry commentary specifically identifies the tasks junior developers traditionally cut their teeth on — fixing straightforward bugs, writing test scripts, generating boilerplate code — as the exact category of work current AI coding tools handle capably (CIO). This is a real, named mechanism, not a hypothetical: if the entry-level job used to consist substantially of that category of task, and that category of task is now largely automatable, the structural argument for why entry-level hiring specifically (not senior hiring) would soften follows directly.
Real, sourced example — the cost-comparison framing. A senior engineer's "$90K junior vs. $10 Copilot" comparison, while informal and anecdotal rather than a rigorous cost-benefit study, captures a real economic logic that shows up in the Stanford data's "substitution roles" category: when a tool can perform a bounded task at a small fraction of a junior salary's cost, employers facing budget pressure have a specific incentive to not fill that junior headcount, rather than an incentive to actively lay off someone already doing the job.
Illustrative example (hypothetical, clearly labeled) — two new graduates, two outcomes. One new CS graduate applies broadly for "junior developer" roles focused on routine feature work and struggles to get interviews, consistent with the substitution-role pattern above. Another new graduate, with the same degree, targets roles explicitly framed around AI-tool-assisted development — reviewing and directing AI-generated code, focusing on system design and integration — and finds more traction, consistent with the complementary-role pattern the Stanford research describes staying flat or growing. This is a composite illustration of the substitution/complementary distinction, not a documented real case.
Data and evidence
– Employment of 22-25-year-olds in AI-exposed occupations is roughly 19% below where it would be had it kept pace with less-exposed peers, and this gap has widened steadily since August 2025 (Stanford Digital Economy Lab).
– The decline operates through reduced hiring, not increased layoffs; experienced workers in the same occupations show no comparable decline (Stanford Digital Economy Lab).
– "Substitution" roles show the decline; "complementary" roles show flat or rising employment, especially for experienced workers (Stanford Digital Economy Lab).
– The authors explicitly frame these as "early, descriptive indicators... rather than causal estimates" — this is the single most important caveat in this entire topic area (Stanford Digital Economy Lab).
– Recent-graduate unemployment (US Federal Reserve Bank of New York, via CIO): computer engineering 7.5%, computer science 6.1%, versus an overall US rate of 4.3% and comparison fields like nursing (1.4%) and elementary education (1.8%) (CIO).
– A Resume.org survey found 6 in 10 of 1,000 surveyed US business leaders said they were likely to lay off employees in 2026, and 4 in 10 planned AI-driven worker replacement by then (CIO) — self-reported sentiment, not a verified outcome.
– Various industry blog posts circulate more dramatic figures — entry-level tech hiring "down 60%," new-grad hiring at major firms "down 50-65% since 2019," a specific claim of "67%" decline in one 2026-dated blog title — evidence not sufficiently verified: these come from secondary blog/aggregator sources (Medium posts, SEO-oriented "2026 report" sites) rather than the primary, methodologically transparent research this article otherwise relies on, and this research did not trace them back to an underlying dataset that could be independently verified. Treat the Stanford figure (19%, with explicit non-causal framing) as the credible anchor, and treat larger round-number claims circulating elsewhere with real skepticism until you can trace them to a primary source.
– On whether the effect is uniform across company size, geography, or specific programming specialization: evidence not sufficiently verified — the Stanford research is framed at the occupation-exposure level, not broken out by these dimensions in the material reviewed.
Comparisons
Junior developers vs. senior developers, AI impact. The Stanford data's core distinction — young workers in AI-exposed roles down ~19% relative to peers, experienced workers showing no comparable decline — is itself the clearest documented comparison available. It's specifically an age/experience-level effect within the same occupations, not a claim that senior roles are immune to any AI impact at all.
Substitution roles vs. complementary roles, employment effect. This is the sharpest, most actionable distinction in the research: substitution-role employment declined; complementary-role employment stayed flat or rose. For a new graduate, this reframes the useful question from "will AI affect tech hiring" (yes, broadly) to "is the specific role I'm targeting structured around tasks AI substitutes for, or tasks where I'd be directing/complementing AI."
Correlational AI-exposure data vs. a causal claim. The Stanford researchers' own framing — descriptive, non-causal — is the most important comparison in the entire topic. Multiple other plausible contributing factors (interest-rate-driven tech hiring pullbacks generally, post-pandemic over-hiring corrections, a broader tech-sector slowdown independent of AI) could be operating alongside AI exposure, and the available research doesn't isolate AI's contribution from those other factors with certainty.
Real-world use cases
– A computer science student choosing between specializations or a bootcamp track — the substitution/complementary distinction is directly useful here: routine CRUD-app, boilerplate-heavy tracks look more exposed than tracks emphasizing systems design, ambiguous problem-solving, or AI-tool-direction skills.
– A hiring manager deciding how to structure junior roles — industry commentary suggests the roles surviving best are ones explicitly redefined around reviewing/directing AI output and handling the judgment-heavy work AI doesn't reliably do, rather than roles defined the same way they were five years ago.
– A career changer evaluating whether to enter software development now — the Fed data on elevated CS/CS-engineering graduate unemployment (6.1%/7.5% vs. 4.3% overall) is a concrete, relevant data point for that specific decision, though it describes a cohort effect, not a guarantee about any individual's outcome.
– A journalist or analyst citing this trend — the distinction between the rigorously caveated Stanford figure and the much larger uncaveated numbers circulating in secondary blog posts is directly relevant to reporting this accurately rather than amplifying an unverified round number.
Common mistakes
– Treating the Stanford 19% figure as proof that AI is definitively "causing" the decline. The researchers themselves explicitly reject that framing — it's descriptive evidence of a strong, widening correlation, not a causal estimate.
– Citing round, dramatic numbers ("60% decline," "67% crisis") from SEO-oriented blog aggregators without checking their underlying source. This research could not verify the methodology behind several such figures circulating online — they should be treated with real skepticism until traced to a primary source.
– Assuming the effect is uniform across all developer roles. The substitution-vs-complementary distinction specifically shows it isn't — role structure matters as much as job title.
– Confusing the two separate data points (Stanford's AI-exposure research and the Fed's graduate-unemployment data) as if they were the same study. They use different methodologies and measure different things, even though they point in a broadly similar direction.
– Treating executive survey sentiment (like the Resume.org "6 in 10 plan layoffs" figure) as equivalent to verified employment outcomes. It's self-reported intent, which is useful context but not the same evidentiary category as measured employment data.
– Assuming this means junior developer roles are disappearing entirely. The research shows a relative decline in hiring pace within AI-exposed occupations specifically, and shows complementary roles staying flat or growing — not a uniform, total elimination of entry-level tech hiring.
Best practices
– Anchor any claim about "AI killing junior developer jobs" to the actual, caveated Stanford data (a ~19% relative gap, explicitly non-causal) rather than larger uncaveated numbers from secondary sources.
– If you're a student or new graduate, evaluate a specific role or track by whether it looks like a substitution role (routine, bounded, AI-automatable tasks) or a complementary role (judgment-heavy, ambiguous, AI-tool-directing work) rather than by job title alone.
– If you're a hiring manager, consider explicitly redefining junior roles around reviewing and directing AI-generated output and handling escalations AI can't resolve, rather than assuming the traditional junior-developer task list still makes sense unchanged.
– Track this data over time rather than treating any single snapshot as final — the Stanford researchers themselves note the gap has been widening since August 2025, meaning the picture is actively evolving.
– Distinguish clearly, in your own thinking and in anything you write about this topic, between correlational evidence and causal proof — this is the single most misused distinction in how this topic gets discussed publicly.
Key takeaways
– The most credible data point (Stanford's Digital Economy Lab) shows employment of 22-25-year-olds in AI-exposed occupations roughly 19% below where it would be otherwise — a real, widening gap since August 2025.
– The researchers explicitly frame this as descriptive/correlational evidence, not proof of causation — a distinction routinely lost in less careful coverage of this topic.
– The effect operates through reduced hiring, not increased layoffs, and is concentrated in "substitution" roles rather than "complementary" ones, where employment stays flat or grows.
– Separate US Federal Reserve data shows recent CS and computer engineering graduates facing notably higher unemployment (6.1% and 7.5%) than the general graduate population (4.3%).
– Larger, more dramatic percentage figures circulating in secondary blog sources could not be independently verified in this research and should be treated with real skepticism.
– The most actionable framework for a new graduate or hiring manager isn't "is AI bad for tech jobs" broadly — it's whether a specific role is structured around substitution or complementary AI use.
Relevant tools.scult.in resources
If you're a developer or new graduate updating your resume and job-search materials around AI-assisted development skills, the Career & Job Search prompt library and GitHub Copilot prompt library have practical starting points for framing AI-tool fluency effectively rather than as an afterthought.
Frequently asked questions
Is AI reducing entry-level programming jobs?
The best available rigorous evidence (Stanford's Digital Economy Lab) shows a real, widening ~19% relative employment gap for young workers in AI-exposed occupations, though the researchers themselves call this descriptive, not causal, evidence (Stanford).
How many fewer young workers are being hired because of AI?
Stanford's research puts AI-exposed 22-25-year-olds' employment about 19% below where it would be had it tracked less-exposed peers — again, a correlational finding, not a precisely attributed causal count (Stanford).
Will AI replace junior software engineers entirely?
No evidence reviewed here supports total replacement — the research shows a relative hiring slowdown concentrated in substitution-type roles, alongside flat-or-growing employment in complementary roles.
Is this AI job-loss data causal or correlational?
Explicitly correlational/descriptive, by the researchers' own stated framing — "canaries in the coal mine... rather than causal estimates" (Stanford).
Why is it harder to get an entry-level developer job right now?
A combination of factors likely contributes — AI automating some traditionally junior tasks, a broader tech hiring slowdown, and post-pandemic hiring corrections — though the research doesn't cleanly isolate AI's specific share of the cause.
Are companies actually hiring fewer junior developers because of AI, or is this overstated online?
Some of the more dramatic percentage claims online could not be verified back to a primary source in this research — the credible, caveated figure is Stanford's ~19% relative gap, not the larger uncaveated numbers circulating elsewhere.
What does "AI-exposed occupation" mean?
An occupation where a substantial share of tasks can plausibly be performed or assisted by current AI tools, as classified in the underlying research methodology.
Does this affect all tech jobs, or just software developers?
The Stanford research is framed at the occupation-exposure level generally, not specifically isolated to software development — though software development is commonly cited as a prominent AI-exposed occupation.
Is this a US-specific finding or global?
The Stanford research and the Fed graduate-unemployment data cited here are both US-focused; this research did not verify equivalent data for other countries.
What's the single most important caveat to know about this topic?
That the best available data (Stanford's) is explicitly described by its own authors as descriptive and non-causal — a real, important, and widening pattern, but not proof that AI alone is "the" cause.
Has AI actually reduced hiring of young workers in AI-exposed occupations?
Yes — employment for 22-25-year-olds in these occupations sits about 19% below where it would be had it kept pace with less-exposed peers (Stanford).
Is AI causing layoffs of junior employees, or something else?
The decline shows up primarily through reduced hiring, not increased layoffs — existing junior employees aren't being separated at a higher rate; new hiring into the role has softened (Stanford).
Do experienced workers in the same occupations show the same decline?
No — experienced workers show no comparable employment decline (Stanford).
Are all AI-exposed jobs affected the same way?
No — the research distinguishes "substitution" roles (declining employment) from "complementary" roles (flat or rising employment, especially for experienced workers) (Stanford).
Has this trend been getting better or worse over time?
Worse, in the sense that the gap has widened steadily since researchers first documented it in August 2025 (Stanford).
What's the mechanism — is it about compensation, or employment?
The adjustment is occurring through employment (hiring volume), not base compensation — wages for those who are hired aren't the primary channel of this effect (Stanford).
What kinds of tasks specifically are cited as being automated away from junior roles?
Fixing straightforward bugs, writing test scripts, and generating boilerplate code, per industry commentary (CIO).
Do computer science graduates specifically face higher unemployment than other new graduates?
Yes, per US Federal Reserve Bank of New York data cited in industry reporting — 6.1% for CS grads and 7.5% for computer engineering grads, versus 4.3% overall (CIO).
Is there executive-level intent to keep reducing junior hiring, or is this just a data artifact?
Survey data suggests real intent — 4 in 10 of 1,000 surveyed business leaders said they planned AI-driven worker replacement by 2026 — though this is self-reported sentiment, not a verified outcome (CIO).
Does this research prove AI is directly causing the entry-level hiring decline?
No — the authors are explicit that these are early, descriptive indicators rather than causal estimates (Stanford).
How do I get a developer job despite this trend?
Target roles and skills framed around directing/reviewing AI-generated work and handling ambiguous, judgment-heavy problems, which the research associates with the "complementary" category showing flat-or-growing employment.
How do I stand out as a junior developer using AI tools?
Demonstrate that you can use AI tools to produce more, verify their output critically, and handle the parts of a task AI still gets wrong — rather than only demonstrating you can write code AI could largely generate on its own.
How do I evaluate whether a specific junior role is "substitution" or "complementary" before applying?
Look closely at the actual day-to-day task description — routine, bounded, repetitive tasks lean substitution; ambiguous, judgment-heavy, cross-functional tasks lean complementary.
How do I position my resume/portfolio given this hiring environment?
Emphasize projects where you used AI tools as part of your workflow and still had to exercise independent judgment, debugging, or design decisions AI didn't make for you.
How should new CS graduates respond to elevated unemployment rates in their field?
Treat the Fed data (6.1%/7.5% vs. 4.3% overall) as a real signal to diversify job search strategy — broader tech-adjacent roles, complementary-role framing, and networking rather than only applying to traditionally-titled "junior developer" postings.
How do I find out if a specific company is one of the ones increasing junior hiring rather than cutting it?
Research recent reporting on that specific company's hiring patterns directly — the broader trend doesn't apply uniformly across every employer.
How do I talk about this trend accurately if I'm writing about it (blog, social post, article)?
Cite the caveated Stanford figure and its explicit non-causal framing rather than repeating larger, unverified round numbers from secondary sources.
How do I know if my current junior role is at risk given this trend?
The research suggests the risk is concentrated more at the hiring stage than at the separation stage for those already employed — though this doesn't guarantee any individual role's safety, since company-specific decisions vary.
How do I decide between specializing early versus staying generalist as a new developer?
No direct evidence in the sources reviewed answers this specifically — evidence not sufficiently verified; the general substitution/complementary framework is the most directly applicable lens available.
How do I responsibly cite this research in my own writing or reporting?
Lead with the Stanford figure and its own stated caveat, and explicitly flag any larger secondary-source numbers as unverified rather than presenting them with equal confidence.
How was "AI exposure" actually measured in the Stanford research?
Not described in granular methodological detail in the material reviewed for this article — evidence not sufficiently verified beyond the general substitution/complementary occupational classification described above; consult the original Stanford publication directly for full methodology.
Could factors other than AI explain some or all of the observed 19% gap?
Plausibly yes — a broader tech-sector hiring slowdown, interest-rate effects on tech investment, and post-pandemic hiring corrections are all independently documented phenomena that could contribute alongside AI exposure; the research's own non-causal framing leaves this open.
Is the 19% gap specific to software development, or an average across many AI-exposed occupations?
The figure as reported describes AI-exposed occupations broadly for the 22-25 age cohort — the material reviewed did not provide a software-development-specific breakdown separately.
Has the gap been measured at multiple points in time, or is 19% a single snapshot?
The researchers note the gap "has widened steadily since August 2025," indicating this is tracked over time, not a single point estimate (Stanford).
What would it take to move this from correlational to causal evidence?
Generally, this would require methods like natural experiments, difference-in-differences designs isolating AI adoption timing from other confounds, or randomized/quasi-randomized variation in AI tool adoption across otherwise similar firms — the current research doesn't claim to have done this, by its own framing.
Junior developers vs. senior developers — who's more affected by AI?
Junior/young workers show the documented decline; senior/experienced workers in the same occupations show no comparable decline (Stanford).
AI substitution roles vs. complementary roles — which is the safer bet for a new graduate?
Complementary roles, per the research, show flat or growing employment, while substitution roles show the decline — a materially different risk profile.
Stanford's employment data vs. the Fed's graduate-unemployment data — are these the same finding?
No — different methodologies and different specific measures, though both point toward real difficulty for new tech-adjacent graduates specifically.
This research's caveated ~19% figure vs. the larger, uncaveated "60-67% decline" figures circulating online — which should I trust?
The caveated Stanford figure, specifically because its methodology and limitations are transparently stated — the larger figures could not be traced to a verifiable primary source in this research.
CS/computer engineering graduate unemployment vs. other fields (nursing, education)?
Notably higher — 6.1%/7.5% for CS/computer engineering versus 1.4%/1.8% for nursing/elementary education, against a 4.3% overall rate (CIO).
I'm a new CS graduate struggling to get interviews — is this "just me," or a real trend?
The data suggests it's a real, documented cohort-level trend, not solely an individual performance issue — though individual factors (portfolio quality, network, specific role fit) still matter within that broader trend.
My company says it's not hiring juniors "because of AI" — is that a legitimate reason, or an excuse?
Could be either — the broader trend documented here makes it a plausible real factor, but companies also cite "AI" as convenient cover for budget-driven hiring freezes unrelated to actual AI capability; without company-specific detail, it's not possible to say which applies in a given case.
I keep seeing wildly different statistics about this topic online — which ones are real?
Prioritize sources with transparent methodology and explicit caveats (like the Stanford research) over round, dramatic numbers from SEO-oriented aggregator blogs that don't cite a traceable primary source.
How do I know if a "junior developer" job posting is really a substitution-type or complementary-type role before I apply?
Read the actual task description closely rather than the title — routine/bounded work signals substitution risk; ambiguous, cross-functional, judgment-heavy work signals complementary framing.
I was told AI will make junior developers obsolete within a few years — is that supported by the data?
Not by the data reviewed here — the research shows a relative hiring slowdown in a specific occupational category, explicitly framed as non-causal and early-stage, not a prediction of total obsolescence.
Should I still pursue a computer science degree given this data?
The data shows real, elevated near-term friction for new CS/CS-engineering graduates specifically — a relevant factor to weigh, but not, on its own, evidence that a CS degree lacks long-term value; long-run occupational outcomes weren't part of the research reviewed here.
Should a bootcamp or self-taught path still be viable given this trend?
No specific verified data on bootcamp-graduate or self-taught-developer outcomes appeared in the sources reviewed — evidence not sufficiently verified; the general substitution/complementary framework applies regardless of educational path.
Should hiring managers keep junior developer budget lines, or reallocate toward AI tooling?
The research on complementary roles suggests value remains in roles that pair junior talent with AI-tool direction and oversight — a full reallocation away from junior hiring isn't clearly supported as the only viable path.
Is it worth specializing in "AI-oversight" or "AI-assisted development" skills specifically as a new developer?
The substitution/complementary distinction in the research suggests this kind of positioning is directionally sound, though no source reviewed here quantifies the specific career-outcome benefit of doing so.
Should career counselors and universities be updating guidance based on this data?
The elevated CS/CS-engineering graduate unemployment figures and the widening Stanford employment gap both suggest this is a legitimate, current factor worth incorporating into career guidance — while still communicating the explicit non-causal caveat rather than presenting it as settled fact.
Sources
- https://digitaleconomy.stanford.edu/publications/canaries-in-the-coal-mine/
- https://www.cio.com/article/4062024/demand-for-junior-developers-softens-as-ai-takes-over.html
- https://tryuncle.com/learn/ai-at-work/will-ai-replace-junior-developers-in-2026
- https://jobsbyculture.com/blog/junior-developer-crisis-2026
