By Rohana Rezel

Schools don’t create senior software engineers. Scars do.

Every senior engineer you’ve ever relied on was once mass-applying to entry-level jobs, writing bad code, and learning the hard way why you don’t store passwords in plaintext. The grizzled veterans are grizzled because they have spent decades fighting in the trenches against inherited tech debt, poorly written requirements, unrealistic deadlines, and tool vendors who overpromise and underdeliver (you know who you are).

That  messy inefficient pipeline that forges a toughened senior out of a delicate CS graduate is collapsing. And nobody in a position to fix it seems to be thinking past the next quarter.

The numbers are not subtle

A Stanford Digital Economy Lab study analyzed payroll data from ADP, America’s largest payroll provider, tracking millions of workers across tens of thousands of firms from 2021 through mid-2025. Employment for software developers aged 22–25 declined nearly 20% from its peak in late 2022. Developers over 26? Stable or growing. The two cohorts moved in lockstep until ChatGPT launched. Then they diverged sharply^[1]”Canaries in the Coal Mine? Six Facts about the Recent Decline in Employment for Young Workers,” by Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen, Stanford Digital Economy Lab (August 2025). Working paper available at https://digitaleconomy.stanford.edu/wp-content/uploads/2025/08/Canaries_BrynjolfssonChandarChen.pdf.

A Harvard study corroborates this at a larger scale. Using résumé data from Revelio Labs covering 62 million workers across 285,000 U.S. firms (2015–2025), researchers found that junior employment in firms actively adopting generative AI declined by 7.7% relative to non-adopters within six quarters of adoption. Senior employment in those same firms continued to rise. The researchers identified this as “seniority-biased technological change” — a term that should make anyone thinking about the long-term health of this industry deeply uncomfortable^[2]”Generative AI as Seniority-Biased Technological Change: Evidence from U.S. Résumé and Job Posting Data,” by Seyed Mahdi Hosseini Maasoum and Guy Lichtinger, Harvard University (August 2025). Available at SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5425555.

The decline is driven by slower hiring, not layoffs. Companies aren’t firing juniors. They’re quietly not posting the positions. The jobs don’t show up on LinkedIn. There’s no severance package to generate headlines. The pipeline just narrows, invisibly.

U.S. Bureau of Labor Statistics data shows overall programmer employment fell 27.5% between 2023 and 2025^[3]U.S. Bureau of Labor Statistics, Occupational Employment and Wage Statistics, as reported in “AI Shifts Expectations for Entry Level Jobs,” by Emily Waltz, IEEE Spectrum, Top Tech 2026 Special Report (December 2025). https://spectrum.ieee.org/ai-effect-entry-level-jobs. Handshake, a major internship recruitment platform, reported a 30% decline in tech-specific internship postings since 2023^[4]”AI vs Gen Z: How AI Has Changed the Career Pathway for Junior Developers,” by Eira May, Stack Overflow Blog (December 2025). https://stackoverflow.blog/2025/12/26/ai-vs-gen-z/. CS graduates now face a 6.1% unemployment rate — nearly double the national average. Entry-level hiring at the 15 largest tech firms fell 25% from 2023 to 2024^[5]HackerRank 2025 Developer Skills Report, as cited in “AI Made Writing Code Easier. It Made Engineering Harder,” by Ivan Turkovic (February 2026). https://www.ivanturkovic.com/2026/02/25/ai-made-writing-code-easier-engineering-harder/. In the UK, tech graduate roles fell 46% in 2024^[6]”The Crisis of Entry-Level Labor in the Age of AI (2024–2026),” Rezi.ai Research Report (January 2026). https://www.rezi.ai/posts/entry-level-jobs-and-ai-2026-report.

These are not minor corrections. This is a structural shift.

The logic that got us here

The reasoning is straightforward, and on a spreadsheet, it looks compelling. A junior developer costs $80–120K annually and needs 6–12 months of mentorship before contributing meaningfully. GitHub Copilot costs $10/month. Claude Code and Cursor cost $20/month. A senior engineer augmented by these tools can handle the boilerplate, CRUD operations, debugging, and test-writing that used to justify an entire entry-level headcount.

Marc Benioff announced Salesforce would hire no new software engineers in 2025, explicitly citing AI-driven productivity gains. Google’s Sundar Pichai told the BBC that AI-powered productivity means the company can do more with the same workforce^[7]”Junior Developers in the Age of AI: Future of Entry-Level Software Engineers (2026 Guide),” CodeConductor (January 2026). https://codeconductor.ai/blog/future-of-junior-developers-ai/. A 2025 LeadDev survey found 54% of engineering leaders plan to hire fewer juniors because AI copilots enable seniors to handle more^[8]LeadDev 2025 Engineering Leadership Survey, as cited in “Junior Developer Extinction: 67% Hiring Collapse Explained,” ByteIota (January 2026). https://byteiota.com/junior-developer-extinction-67-hiring-collapse-explained/.

Here’s the part that should alarm you: the self-reported productivity gains that justify these decisions don’t match measured reality. Companies self-report a roughly 25% productivity increase from AI adoption. Actual measured results tell a different story. Google’s DORA 2024 report found roughly a 2% overall productivity increase for every 25% increase in AI adoption^[9]”2024 Accelerate State of DevOps Report,” by the DORA (DevOps Research and Assessment) team, Google Cloud (2024). Key findings cited in “A 2026 Guide on How to Use AI for Developer Productivity,” Axify (March 2026). https://axify.io/blog/use-ai-for-developer-productivity. The gap between executive expectation and engineering reality is roughly 12x.

AI completes boilerplate 55% faster. But boilerplate was never the bottleneck.

What juniors actually did (that nobody talks about)

The conversation about junior developers usually frames them as consumers of resources — expensive, slow, and error-prone. This framing misses what junior developers produce that doesn’t show up in sprint velocity.

They stress-test your documentation. When a junior can’t set up their dev environment from your README, that’s a signal your onboarding is broken. It’s a signal you won’t get from AI.

They ask the questions that expose hidden assumptions. “Why are we doing it this way?” is the most valuable question in engineering, and it comes disproportionately from people who haven’t yet internalized the codebase’s historical accidents as normal.

They build the institutional memory that keeps organizations functional. The senior engineers who keep your systems running developed their understanding of those systems by working on them when the stakes were low. That understanding cannot be prompted into existence.

Most critically, they are the pipeline for your future senior engineers. This is not metaphorical. It is arithmetic. A significant reduction in junior hiring in 2024–2026 means a proportional reduction in candidates for senior roles in 2031–2036. The industry is eating its seed corn.

The Klarna warning

In 2023, Klarna stopped hiring altogether. By 2024, the company had slashed customer service and marketing departments, partnered with OpenAI, and publicly declared that AI could perform all human jobs at the company. CEO Sebastian Siemiatkowski celebrated $10 million in savings. The employee count dropped from 5,500 to 3,400^[10]”As Klarna Flips from AI-First to Hiring People Again, a New Landmark Survey Reveals Most AI Projects Fail to Deliver,” by Irina Ivanova, Fortune (May 9, 2025). https://fortune.com/2025/05/09/klarna-ai-humans-return-on-investment/.

By mid-2025, Klarna was scrambling to rehire. Customer satisfaction had dropped. Service quality was inconsistent. The company started pulling software engineers and marketers from their specialized roles to answer customer service calls^[11]”Company Replaces Customer Support With AI, Then Panics and Forces Engineers to Work the Phones as the AI Fails,” Futurism (September 4, 2025). https://futurism.com/klarna-ai-automation-engineers. Siemiatkowski admitted: “We focused too much on efficiency and cost. The result was lower quality, and that’s not sustainable”^[12]”Klarna Walks Back AI Overhaul: Rehires Staff After Customer Service Backlash,” LaSoft Blog (May 21, 2025). https://lasoft.org/blog/klarna-walks-back-ai-overhaul-rehires-staff-after-customer-service-backlash/.

Research from Orgvue and Forrester found that 55% of companies that executed AI-driven layoffs now regret their decisions^[13]Orgvue and Forrester Research AI workforce surveys, as reported in “The Klarna AI Experiment: Why Replacing Humans with AI Backfired,” Linkifico (November 2025). https://www.linkifico.com/post/the-klarna-ai-experiment-why-replacing-humans-with-ai-backfired. An IBM survey of 2,000 CEOs found that just one in four AI projects delivers on its promised ROI. Only 16% are scaled across the enterprise^[14]IBM Institute for Business Value CEO Study (2025), as reported in Fortune (May 9, 2025). https://fortune.com/2025/05/09/klarna-ai-humans-return-on-investment/.

Klarna is a customer service story, not a software engineering story.  But the pattern of aggressive replacement, followed by quality degradation, followed by expensive and chaotic re-hiring applies directly to engineering organizations that assume AI can substitute for human development pipelines.

The senior burnout spiral

Here’s the part that software engineering managers are starting to feel but haven’t fully connected to the junior hiring freeze.

A study by researchers at UC Berkeley’s Haas School of Business, published in Harvard Business Review in February 2026, followed 200 employees at a U.S. tech company for eight months. Workers didn’t use AI to finish earlier. They used it to do more. AI accelerated certain tasks, which raised expectations. Higher expectations increased reliance on AI. More reliance widened scope. Wider scope expanded workload. Eighty-three percent of workers in the study said AI increased their workload. Burnout was reported by 62% of associates^[15]”AI Doesn’t Reduce Work—It Intensifies It,” by Aruna Ranganathan and Xingqi Maggie Ye, UC Berkeley Haas School of Business, published in Harvard Business Review (February 2026). https://hbr.org/2026/02/ai-doesnt-reduce-work-it-intensifies-it.

A report from Multitudes, a New Zealand-based engineering analytics firm, assessing over 500 developers found that engineers merged 27% more pull requests but experienced a 20% rise in out-of-hours commits. They’re producing more, and they’re producing it at night^[16]Multitudes Software Engineering Report (2025), as reported in “Why Developers Using AI Are Working Longer Hours,” by Sophie Bushwick, Scientific American (March 2026). https://www.scientificamerican.com/article/why-developers-using-ai-are-working-longer-hours/.

Without juniors to delegate to, senior engineers absorb the work that used to be distributed. They review more AI-generated code. They handle more oncall incidents because there’s no one to gradually take over. They make more architectural decisions per day because there’s nobody asking basic questions that force them to slow down and think.

And here’s the irony. Companies cut juniors partly to reduce the mentorship burden on seniors. But the mentorship burden has been replaced by an AI-review burden that’s worse.At least juniors learn and eventually become net contributors. AI-generated code doesn’t level up over time. You’re reviewing the same kinds of mistakes in perpetuity.

The vacancy chain is broken

In a healthy labor market, economists speak about “vacancy chains”: a senior leaves, a mid-level moves up, and a junior is hired. AI disrupts this chain by automating the bottom link. The bottom rung of the ladder has been sawed off^[17]”The Crisis of Entry-Level Labor in the Age of AI (2024–2026),” Rezi.ai Research Report (January 2026). https://www.rezi.ai/posts/entry-level-jobs-and-ai-2026-report.

Graduates from mid-tier institutions are the worst affected. The Harvard study found that graduates from colleges and universities with very high or very low status were less affected^[18]”AI Is Already Taking Junior Jobs — the Challenge to HE,” Future Campus (September 9, 2025), summarizing findings from Hosseini and Lichtinger (2025). https://futurecampus.com.au/2025/09/09/ai-is-already-taking-junior-jobs-the-challenge-to-he/. If you went to Stanford or MIT, your network and brand still open doors. If you went to a community college, you likely weren’t targeting these roles to begin with. The students in the middle, the ones who did everything “right,” studied CS, built projects, applied to internships, are the ones sending out hundreds of applications and hearing nothing.

This isn’t evenly distributed geographically, either. Among 400 classmates at the Indian Institute of Information Technology in Jabalpur, fewer than 25% had secured job offers as of late 2025. Indian IT services companies have reduced entry-level roles by 20–25%. LinkedIn, Indeed, and Eures noted a 35% decline in junior tech positions across major EU countries in 2024^[19]”AI Is Wiping Out Entry-Level Tech Jobs, Leaving Graduates Stranded,” by Moinak Pal, Rest of World (December 19, 2025). https://restofworld.org/2025/engineering-graduates-ai-job-losses/.

The counterargument, and why it’s incomplete

The standard rebuttal: “The jobs aren’t disappearing, they’re changing. Juniors just need to learn AI skills.” There’s truth in this. OpenAI is hiring junior software engineers. AI/ML roles are growing. The BLS projects 328,000 new software developer jobs by 2033^[20]U.S. Bureau of Labor Statistics, Occupational Outlook Handbook, Software Developers projection 2023–2033, as cited in “Developer Hiring Crisis 2026: 40% Worse, Junior Drops 73%,” ByteIota (March 2026). https://byteiota.com/developer-hiring-crisis-2026-40-worse-junior-drops-73/.

But these projections assume a functional pipeline. You can’t hire 328,000 developers in 2033 if you didn’t train them from 2026 to 2030. The skills gap doesn’t resolve itself retroactively.

The “just learn AI” advice also suffers from a composition problem. If every junior developer becomes an “AI engineer,” who builds and maintains the systems that AI engineers need to exist? Somebody still has to understand DNS, memory management, database indexing, and the hundred other fundamentals that AI tools abstract away but don’t eliminate.

There’s a real risk, one the Harvard researchers flag explicitly, that AI is eroding the bottom rungs of career ladders by automating intellectually mundane tasks that junior employees typically handle. Those tasks were the training ground. Debugging someone else’s code teaches you how systems fail. Writing boilerplate teaches you how systems are structured. Reviewing PRs teaches you how other people think about problems^[21]Hosseini and Lichtinger (2025): “In many such jobs, workers begin at the bottom of the career ladder performing intellectually mundane tasks, i.e., routine yet cognitively demanding activities such as debugging code or reviewing legal documents, which are likely to be especially exposed to recent advances in AI.” Available at SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5425555.

What would a plan look like?

Nobody in a position of influence is proposing a coherent solution. Here’s what one might look like.

Structured apprenticeship programs with AI integration. Instead of eliminating junior roles, redesign them. A junior developer in 2026 shouldn’t be writing CRUD endpoints by hand. Instead, they should be reviewing AI-generated CRUD endpoints and learning to evaluate code they didn’t write. This is, arguably, a harder skill than writing the code from scratch. It requires understanding without the scaffolding of having built it yourself.

Economic incentives for training. The market alone won’t fix this because individual companies bear the cost of training but the industry captures the benefit. If you invest 12 months in a junior developer and they leave for a competitor, you’ve subsidized someone else’s hiring pipeline. This is a collective action problem. Tax incentives for accredited training programs, industry-funded apprenticeship pools, or consortium-based training models could realign the incentives.

Honest metrics. Stop measuring AI productivity by lines of code generated or tasks completed. Measure what matters: time-to-resolution for production incidents, defect rates over 6-month windows, and onboarding time for new team members. These metrics capture the downstream cost of eliminating the training pipeline in ways that sprint velocity never will.

Senior engineers need to speak up. If you’re a senior or staff engineer, you understand what’s happening better than the executives making headcount decisions. The mid-level engineers you rely on today were juniors five years ago. If your company hires zero juniors this year, and next year, and the year after, who exactly is going to replace you when you burn out?

The math problem nobody wants to do

Here is the calculation that keeps me up at night.

The average tenure of a software developer at a large tech company is 2–3 years. The average time to grow a junior into a reliably independent mid-level engineer is 2–4 years. The average time from mid-level to senior is another 3–5 years. These numbers vary, but the order of magnitude is consistent: it takes roughly 5–9 years to produce a senior engineer from a new graduate.

If the industry meaningfully reduces junior hiring for 3 consecutive years (2024–2026, which is happening), the effects won’t be felt immediately. They’ll be felt in 2029–2033, when the pipeline of mid-level engineers thins, and in 2032–2036, when the pipeline of senior engineers thins.

By then, the companies that cut junior hiring will be competing for a smaller pool of experienced engineers and complaining about a “talent shortage”. The CEOs will blame everyone but themselves for the problem they created through three years of quarterly optimization.

We’ve seen this before, albeit in different contexts. Hospitals that cut residency programs in the 1990s faced physician shortages in the 2000s. Airlines that reduced pilot training pipelines after 9/11 spent the 2010s dealing with pilot shortages. In each case, the people who made the cuts weren’t around to deal with the consequences.

The Stanford researchers put it clearly: the overall economy-wide impact of AI on employment is still small, but with a large amount of uncertainty about young workers in particular^[22]”AI and Labor Markets: What We Know and Don’t Know,” by Erik Brynjolfsson, Stanford Digital Economy Lab (January 2026). https://digitaleconomy.stanford.edu/news/ai-and-labor-markets-what-we-know-and-dont-know/. The Harvard researchers are more direct: “If AI disproportionately targets entry-level tasks, the bottom rungs of these ladders may be eroding”^[23]Hosseini and Lichtinger (2025), p. 2. Available at SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5425555.

The question isn’t whether the junior developer pipeline is broken. It is. The data speaks loudly and clearly to this. The question is whether anyone with the authority to fix it will do so before the consequences become irreversible.

Right now, the answer is no. Nobody has a plan.

Rohana Rezel is a technologist, researcher, and community leader based in Vancouver, BC