The Developer Paradox: How AI Created a Shortage of the Humans It Was Supposed to Replace

IBM announced in February 2026 that it would triple its entry-level hiring, specifically for software developers and, in the company's own words, for all the roles people are being told AI can do. The company's chief human resources officer, Nickle LaMoreaux, was explicit about what those roles would look like: rewritten positions built around human judgment, customer interaction, and oversight of AI systems.

IBM had already eliminated thousands of administrative and back-office roles as AI automated routine tasks, particularly in HR and back-office operations. LaMoreaux's stated logic was forward-looking: the companies that doubled down on entry-level hiring in this environment would be the most successful three to five years from now.

The tech labour market surrounding that decision tells a story the announcement alone cannot carry.

Between 2022 and May 2026, more than 900,000 technology workers globally lost their jobs. The explanation that circulated was tidy and confident: AI had arrived, and the humans who built software were the first casualty. The data, it turned out, was telling a different story.

Software developer job postings on Indeed were 11% higher in early 2026 than in the same period a year earlier, growing faster than overall job postings. The US Bureau of Labor Statistics projected software developer employment growth at 15% through 2034. The United States faces a projected shortage of 1.2 million developers by 2027. These are numbers from an industry that cannot hire human technical talent fast enough to meet what AI deployment actually requires.

The paradox at the centre of the current moment is this: the technology most confidently cited as the agent of human displacement has produced, through the specific mechanics of how it actually works, a structural shortage of the humans it was supposed to displace.

AI created a shortage of the humans it was supposed to replace.

The firing wave was real, but its cause was misread

The scale of the tech layoff wave between 2022 and early 2026 is not in dispute. Layoffs.fyi and TrueUp.io, the two primary trackers of global tech workforce reductions, documented 164,969 jobs cut in 2022, rising sharply to 262,735 across 1,186 companies in 2023, the peak year, representing a 59% increase on 2022. The years that followed brought 152,922 in 2024, then 245,953 in 2025. By May 2026, the year-to-date figure had already reached 95,878 across 249 companies. The cumulative total across those four years exceeds one million.

The period immediately before those layoffs explains them better than AI does. Between 2019 and 2022, some of the largest technology companies nearly doubled their employee headcount. The pandemic produced genuine acceleration in digital adoption across almost every sector, and companies that had been growing steadily found themselves needing far more engineering, product, and operations capacity than their existing teams could provide. The hiring that followed was calibrated to a growth environment that proved temporary. When interest rates rose and revenue expectations contracted, the correction was proportional to the excess.

Challenger, Gray and Christmas, which tracks US job cut announcements and their stated reasons, found that AI was cited as the direct reason for approximately 4.5% of all US tech layoffs in 2025. In absolute terms, that represents around 55,000 roles attributed to AI-driven displacement. The remaining 95.5% of layoffs in that year had stated causes rooted in economic rebalancing, over-hiring, reduced venture funding, and the normalization of headcount that had been set during an anomalous three-year growth period. Companies citing efficiency and AI were frequently companies that had hired at an unsustainable pace between 2020 and 2022 and were reducing to levels they could sustain without those growth assumptions.

The AI narrative served a purpose in that environment. Framing layoffs as forward-looking efficiency decisions about technology was more strategically coherent than acknowledging a hiring miscalculation made under pandemic-era conditions. The language was commercially useful. The data behind it was thinner than the headlines implied, and the distinction matters because the entire premise of the displacement conclusion rests on the assumption that those layoffs were driven by AI capability. They were not. They were driven by economic reality, with AI providing rhetorical cover for decisions already made on financial grounds.

The counter-signal that started before the dust settled

The most striking feature of the 2022 to 2023 labor market was what happened simultaneously to different types of job postings within the same sector. As overall information technology job listings fell by 31% during that period, job postings specifically referencing AI skills increased by 42%. Two opposite trajectories, running in parallel, inside the same industry, during the same months when layoff announcements were generating daily coverage.

That divergence accelerated from there. Year-over-year growth in US job postings mentioning AI reached 114.8% in 2023, according to GlobalData analysis published by Autodesk. In 2024, that figure rose further to 120.6%. Through April 2025, growth was still running at 56.1%. Generative AI as a specific required skill appeared in 16,000 US job postings in 2023. By 2024, that figure had reached more than 66,000, a nearly fourfold increase in twelve months, documented by Stanford's Human-Centered AI Institute and Lightcast.

By the end of 2025, the divergence between overall hiring and AI-specific hiring had become a structural feature of the US labor market. Indeed's Hiring Lab reported that total job postings were only 6% above pre-pandemic baseline levels. Job postings mentioning AI were more than 130% above the same baseline. AI-related roles had reached 4.2% of all US job postings by late 2025, up from close to zero in 2019, with the acceleration point arriving sharply after the public release of ChatGPT in late 2022.

The hiring market did not grow. What changed was its composition. Companies were doing less overall hiring because their financial constraints required it, and markedly more AI-oriented hiring because deploying AI at scale requires specific human expertise that the existing workforce could not supply in sufficient quantity. The workers who had developed that expertise before the demand appeared captured the moment. The workers who had not were not primarily replaced by AI systems. They were replaced by other people who had already made the relevant transition, a distinction with real consequences for how the workforce transition should be understood and responded to.

What companies are paying for, and what that price signals

The AI Engineer role was not a defined function at most organizations five years ago. By 2025, it was the fastest-growing job title on LinkedIn, with year-over-year growth of 143.2%, according to Autodesk's AI Jobs Report. In 2026, AI Engineer remains the fastest-growing title on the platform, sustained across three consecutive years and now appearing as the fastest-growing role specifically among younger workers entering the labor market for the first time, according to LinkedIn and Fortune reporting from early 2026.

The compensation signal attached to AI skills is equally clear. Workers with demonstrable AI competencies earn an average of 56% more than peers in comparable roles who lack them, according to 2025 industry research. A wage premium of that magnitude, appearing within a three-year window, signals genuine market scarcity rather than speculative demand. Wages this far above comparables hold only when supply falls short of what the market needs. Companies are paying that premium because the supply of workers who understand both the technical architecture and the practical limits of AI systems has not kept pace with deployment.

The supply of humans who understand AI has not kept pace with AI itself.

The demand extends well beyond the technology sector, and this is the element that most workforce narratives miss. By 2024, more than half of all AI-skill job postings in the United States came from sectors outside information technology, according to Lightcast and CBS News research.

Healthcare, financial services, manufacturing, logistics, and professional services were all generating significant demand for workers capable of operating AI tools, evaluating AI outputs, and applying human judgment where AI performance is unreliable, contested, or unverifiable.

By January 2026, 45% of all data and analytics job postings were mentioning AI specifically, the highest share of any occupational category tracked by Indeed's Hiring Lab. Software development and general IT roles followed closely behind. One in four entry-level postings in consulting and finance now requires demonstrable AI fluency, up from fewer than one in twenty just two years earlier, according to LinkedIn's Economic Graph review for 2026.

Workers with AI skills are not concentrated in AI companies. They are distributed across every sector of the economy that has started deploying AI systems and discovered, at the point of deployment, that those systems require ongoing human involvement to produce outputs worth trusting.

Why non-tech sectors are competing for the same talent

The spread of AI-skill demand into traditional industries carries a specific implication for workers currently weighing whether to invest in building those skills. The competition for AI-literate talent is not contained within technology companies. It is now distributed across legal services, regulated financial industries, healthcare operations, and supply chain management, each of which brings its own hiring budgets and its own urgency.

Fintech companies sit at a particular point of concentration within this broader dynamic. The sector, projected by Fortune Business Insights to grow at a compound annual rate of 20.5% through 2030, is deploying AI across fraud detection, credit decisioning, compliance monitoring, and customer service at a pace that requires technical talent on both sides: the engineers who build the systems and the domain-knowledgeable professionals who validate that the outputs are trustworthy in contexts where errors carry regulatory or financial consequences. The workers who can bridge those two functions, who understand both the AI tooling and the domain it is operating in, are the most scarce and therefore the most expensive.

The 70% year-over-year increase in US roles requiring AI literacy reported by LinkedIn's Economic Graph in January 2026 is not a technology-sector statistic. It reflects a workforce recalibration happening simultaneously across multiple industries, each of which arrived at AI deployment at different speeds but is converging on the same conclusion: AI literacy has moved from a competitive advantage for individual workers to a baseline requirement in a growing share of professional roles. The WEF projected that AI and machine learning specialist roles would grow by 40% by 2027, in its Future of Jobs 2025 report. That growth is distributed across sectors in proportions that were not anticipated three years ago.

The practical consequence is a talent market where technology companies are no longer the only bidder for workers who understand AI systems. That competition raises the price of those workers across every sector competing for them, and it means the 56% wage premium visible in technology is likely understating the true value of AI competency in sectors where relevant expertise is even scarcer than in tech.

The IBM signal: what it tells us about junior talent specifically

The expectation in most workforce analyses was that AI would hit junior workers the hardest. A junior engineer exists partly to handle the tasks a senior engineer finds repetitive or below the productive threshold of their expertise. If AI absorbs those tasks, the junior role should logically contract.

Stanford's Digital Economy Lab found evidence consistent with that concern. Entry-level hiring at the fifteen largest technology firms fell by 25% between 2023 and 2024. Young workers aged 22 to 25 in AI-exposed jobs saw employment decline by 13% between 2022 and 2025. Those numbers are real and represent genuine displacement concentrated at the early-career end of the workforce, during the period when the tools were arriving faster than early-career workers could develop fluency with them.

IBM's decision to triple entry-level hiring lands against that Stanford data with a specific complication: the inference that naturally follows from those numbers is not the one IBM's move supports.

After ChatGPT's release, a conviction spread quickly through the industry that AI could absorb junior-level work entirely, and the Stanford data confirmed companies were acting on that belief. IBM had been among them, pausing and slowing hiring for roles it believed AI would eventually replace.

The February 2026 announcement is a direct reversal of that calculation. The company discovered, at the scale of actual deployment, that AI still needs human help to function reliably in complex organizational environments, and that the entry-level roles it had assumed were redundant still required people to make them work. That discovery made junior hiring economically rational again, for a company that had spent the preceding period betting it would not be.

Software development as a category: the evidence against displacement

The case for AI displacing software developers broadly has faced consistent pressure from employment data through 2026. Software developer job postings on Indeed were up 11% year over year by early 2026, a faster growth rate than overall job postings on the same platform. The Federal Reserve Economic Data showed software development postings up 15% since mid-2025, with AI and machine learning roles within that category growing at 85% year over year. The Bureau of Labor Statistics projected software developer employment growth at 15% through 2034, its most recent update following several years of monitoring the AI impact on the profession.

The mechanism behind those numbers is consistent and worth stating plainly. Every AI system operating inside an organization requires engineers who understand its failure modes, monitor its outputs for drift or degradation, correct its errors before those errors compound, and rebuild it when its operating environment changes in ways the original training did not account for.

The work of building, evaluating, maintaining, and improving AI is engineering work. Every company that deploys AI at meaningful scale creates a demand for technical talent required to keep that deployment functional, and that demand grows with the scale of deployment.

The LinkedIn and GitHub joint study examining what happened to engineering hiring after GitHub Copilot adoption found a small but measurable increase in engineering headcount at companies that adopted the AI coding tool. The tool did not reduce demand for developers. It expanded what developers could build, which expanded what companies wanted to build, which expanded the number of developers those companies needed. This pattern is consistent with the history of software productivity tooling: better tools produce more software, not less demand for the people who write it.

The projected US developer shortage of 1.2 million by 2027, documented by Boundev drawing on Bureau of Labor Statistics data, is structural in origin. It reflects AI investment requiring engineering support, digital adoption still spreading through sectors that were late to it, and senior technical talent retiring at a rate that graduate pipelines cannot match.

The companies currently framing their AI ambitions in terms of workforce reduction are, in the same reporting period, competing for engineering talent in the same labor market where that shortage is becoming visible.

The mechanism nobody is counting: quiet attrition

The layoff trackers capture announcements. They do not capture the decision to leave a position unfilled when someone resigns. The distinction is becoming central to an accurate account of how AI is affecting workforce size, because the most consequential mechanism of headcount reduction in 2025 and 2026 is the one that generates no announcement.

Field research by Shawn Kanungo, whose work draws on documented observation across Fortune 500 companies, identified a consistent pattern in how organizations are using AI to reduce workforce size without public layoff events. A team with twelve people in 2023 has seven people in 2026. The workload is approximately unchanged. The five missing roles were never announced as cuts, never appeared in a tracker, and never generated a press release. The people who held those roles left through normal attrition, and leadership decided not to backfill them, because AI tools had absorbed the marginal capacity those positions represented.

Organizations using this approach are technically accurate when they state they are not replacing humans with AI. The attrition is natural, the decision not to backfill is administrative, and the AI deployment that fills the resulting gap is a separate operational choice. The outcome in headcount terms is identical to a structured layoff event, but the mechanism is invisible to every measurement tool used to track AI-driven displacement, because it registers as the absence of hiring rather than the presence of terminations.

The Boston Consulting Group's 2026 analysis estimated that between 10 and 15% of US jobs could be eliminated within five years through processes like this, while simultaneously projecting that the majority of affected roles would be reshaped rather than eliminated entirely. The distinction between elimination and reshaping carries real weight. A reshaped role requires its holder to develop different skills, produce different outputs, and work alongside AI systems in ways that did not exist three years earlier. An eliminated role simply ceases to exist. Both are happening, at different rates and across different types of work, in ways that current aggregate statistics are not yet equipped to separate cleanly.

The reskilling equation the headline projections obscure

The World Economic Forum's jobs projections are the most frequently cited headline in optimistic accounts of AI's workforce impact. AI could eliminate 92 million jobs by 2030 while creating 170 million new ones, producing a net global gain of 78 million positions. The arithmetic is encouraging. The mechanism it requires is more demanding than the summary communicates.

The 78 million net gain assumes that workers displaced from the 92 million eliminated roles can access, afford, and complete the transition into roles among the 170 million created. The WEF's own research complicates that assumption directly: 59% of the global workforce requires significant reskilling before 2030 to remain productive in an AI-integrated labor market. More than half of every working person on the planet needing substantive skill development within four years is not a training program challenge. It is a structural problem requiring institutional responses, industry coordination, and sustained investment at a scale that most labor markets are not currently approaching.

59% of the global workforce requires significant reskilling before 2030 to remain productive.

The 56% wage premium for AI-skilled workers is the market's measurement of how constrained supply currently is. If the pipeline from displaced worker to reskilled worker were operating at the throughput the net job creation numbers require, that premium would be compressing as supply increased to meet demand. It has not compressed. The premium has held and in specific segments grown, which indicates that the transition pathway is not functioning at the pace the aggregate projections assume.

The workers who have captured the gains visible in the WEF projections and the IBM hiring announcement are not a random sample of the workforce. They are workers who had the preparation, institutional access, and timing to develop AI literacy before the demand for it exceeded supply. The workers who did not are not absent from the WEF net gain number. They are contained within it, as the denominator that the 170 million created jobs are intended to absorb. Whether the absorptive capacity is real depends on infrastructure that most labor markets are still in the early stages of building.

McKinsey and IBM: the same year, opposite moves, the same logic

In late 2025, McKinsey cut approximately 200 technology and support roles, explicitly citing AI efficiency as the driver, according to reporting by Metaintro. The roles eliminated were concentrated in back-office functions: research, compliance, reporting, and administrative work that AI tools had made automatable at sufficient scale. Client-facing roles were protected. The action was noted, reported, and moved past.

In February 2026, IBM announced the tripling of its entry-level developer hiring.

Both decisions were rational given the type of work involved. McKinsey automated the work that AI can currently absorb: structured, rule-following, pattern-recognition tasks at scale, in a back-office environment where errors are correctable and the cost of AI failure is manageable. IBM hired for the work AI cannot currently absorb without human oversight: the engineering judgment required to build, deploy, evaluate, and improve AI systems in complex organizational environments where the cost of AI failure is not manageable.

The two moves, from the same reporting period, describe the fault line that runs through the current data on every axis. AI is absorbing rule-following work faster than it is absorbing contextual judgment work. Workers whose roles consisted primarily of structured pattern-recognition tasks have faced genuine displacement. Workers whose roles require the kind of judgment that AI systems still fail at unpredictably are being competed for. The McKinsey cut and the IBM hire are not contradictions.

They are a precise description of where the technology currently stands: capable enough to remove structured support functions, dependent enough on human engineering judgment that companies building on it are hiring that judgment aggressively.

What the data cannot yet resolve

Every credible projection of AI's employment impact runs through approximately 2028. That boundary marks the horizon of what current deployment data can support, and it is also, approximately, the boundary of the construction phase.

The construction phase is the period in which humans are needed to build, train, evaluate, and maintain AI systems that are not yet capable of doing those things for themselves. During this phase, demand for human technical talent grows because AI deployment grows, and AI deployment at meaningful scale requires human engineering support that AI cannot currently provide. Every projection showing net job creation by 2028 reflects the economics of this phase, and Gartner's assessment that AI's job impact will be neutral through 2026 before turning net positive by 2028 is grounded in observable deployment data consistent with the construction phase dynamic.

What that framing does not model is the phase that follows. Multiple AI research organizations have stated the long-term objective of developing systems capable of building, training, and improving AI systems with minimal human involvement. When that capability arrives at meaningful scale, the construction-phase demand for engineering talent that is currently absorbing much of the disruption contracts. The timing of that transition is genuinely unknown. No analyst is currently able to date it with confidence that the underlying research would support. The projections through 2028 are reasonable given current evidence and provisional given what comes after.

The WEF and LinkedIn's January 2026 joint report framed the current labor market as requiring a new social contract for the digital age, acknowledging that the pace and scale of change will test employment frameworks, social support systems, and the institutions that connect displaced workers to new roles, in ways that job creation projections alone cannot address. What this means in practice for the 52% of global workers who, according to the same report, describe themselves as actively seeking new employment in 2026, is a question the current data raises and does not resolve.

What the full dataset, taken together, actually says

The evidence assembled from 2022 through May 2026 does not support the conclusion that AI is making humans unnecessary. It supports a more specific and more uncomfortable conclusion: AI has made human judgment more expensive to replace than it was before AI existed.

The layoff wave was real, and a portion of it, approximately 4.5% by the most credible direct measure, was AI-driven. The larger portion was an economic correction to over-hiring during an anomalous growth period, and the two causes were conflated in public narratives that served the rhetorical interests of companies making difficult financial decisions.

The construction phase of AI deployment is creating new roles faster than AI is eliminating old ones, by every credible projection through 2028. IBM is tripling junior developer hiring. The US faces a structural developer shortage of 1.2 million by 2027. Workers with AI skills earn 56% more than peers who lack them, and that premium has held because supply has not caught up with demand.

The quiet attrition mechanism, the reskilling gap, the concentration of gains among workers already positioned to claim them, and the genuine uncertainty about what happens when AI systems become self-sustaining are all simultaneously true. The workers who have benefited from this moment and the workers who have not are not separated by chance. They are separated by preparation, access, and timing, which are categories that aggregate projections flatten into net numbers but that individual workers experience as the entire difference between advancement and displacement.

The technology did not eliminate the market for human technical judgment. Through the specific mechanics of how AI systems actually work in 2026, it generated a structural demand for that judgment that neither its advocates nor its critics predicted. IBM's decision to triple junior developer hiring, while other companies are still citing AI efficiency in workforce reduction announcements, is the most precise single description of where AI deployment stands today: dependent on the humans it was meant to replace, competing for the ones who are ready, and creating a shortage it was supposed to prevent.

Editor's note

Every piece published on The Bright Minded goes through careful verification, but mistakes can happen. If you spot an error, have additional information, or want to flag anything, write to rosalia@thebrightminded.com.