In January 2026, Ian Tracey — a former Pinterest engineer now working in applied AI — published a widely shared post on X titled The K-Shaped Future of Software Engineering. His thesis: the tech industry is diverging, with some engineers becoming more valuable than ever while others watch their skills depreciate in real time. As Tracey put it, “AI is leverage, and leverage amplifies whatever you already are.”
He also made a point that gets lost in most AI jobs impact coverage: when productivity spikes, healthy organizations don't immediately cut their workforces. They chase adjacent markets, pay down lingering operational debt, clear multi-year backlogs, and raise the bar for what ‘good enough’ means.
There's a name for this. Jevons Paradox says that as a commodity gets cheaper, previously unprofitable uses become profitable, so total demand goes up. As intelligence gets cheaper, the set of worthwhile things to do with it gets broader. If the paradox holds for AI, we should see it in the workforce statistics as rising demand, and we do. But the same statistics also show something else, and it's worth looking at both carefully.
Bureau of Labor Statistics: same industry, opposite trajectories
The BLS projects that employment of software developers, QA analysts, and testers will grow 15% from 2024 to 2034 — five times the 3% average for all occupations — while narrow “computer programmer” roles decline 6% over the same decade. Zoom out and computer and mathematical occupations overall are projected to grow 10.1%, more than triple the economy-wide rate.
Read those numbers together. The market isn't shrinking the software workforce; it's repricing what it pays for. Code generation stopped being the bottleneck, so demand shifted from people who know and type code syntax to people who design systems — and the total volume of software under management keeps growing. That's Jevons Paradox showing up in a federal dataset.
Federal Reserve CFO surveys: reinvestment with some fine print
In the Fed's joint CFO survey with Duke University, surveying roughly 750 financial executives, productivity and innovation outranked cost-cutting as the reason firms invested in AI, and researchers found little evidence of near-term, AI-driven employment declines. That's the expansionary story from the people who control headcount. Companies are mostly spending their AI gains on doing more, not on doing the same with fewer people.
The fine print: In the same body of research, 44% of CFOs plan some AI-related cuts, modest in the aggregate, but not nothing. And a separate New York Fed survey found that firms are quietly scaling back hiring, a development Fed officials flagged as hitting newer entrants to the job market hardest.
That said, 2026 has produced AI layoff headlines. AI was the most-cited reason for US job cuts in the first half of the year, but many of those cuts are being walked back. Roughly a third of managers who cut a role for AI have since rehired it, and over half of the executives who made those cuts now call them a mistake. The firings are loud, frequently reversed, and partly a cover story. The smaller change that’s sticking is the hiring that never happens.
World Economic Forum: the exposure is global
The WEF's 2026 briefings on entry-level work put the scale in global terms, with more than one in three young workers worldwide are in occupations with medium-to-high exposure to AI-driven task change and in some regions, three in four. The workers themselves see it coming, with 28% of entry-level employees believe half or fewer of their current skills will still be relevant in three years.
The WEF's warning to employers is the same one we'd give, that cutting junior talent buys short-term efficiency at the cost of succession plans, knowledge transfer, and the professional development pipeline itself.
Revelio Labs: the split, quantified
Revelio's workforce data shows US entry-level job postings down roughly 35% since early 2023, with junior software and data roles down as much as 67%. But their most important finding is the AI impact split, with a 10-point increase in an occupation's AI exposure correlates with an 11% drop in entry-level demand and a 7% rise in demand for experienced roles in the same occupation. This is the exact K-shaped future that Ian Tracey wrote about in January.
Meanwhile, entry-level job mobility sits about 40% below 2019 levels, which is a far steeper pullback than mid or senior-level workers have seen, this while the average new hire now brings two more years of experience than in 2019. This isn't a hiring freeze that thaws with the next rate cut. It's the career ladder being rebuilt without its bottom rungs.
Stanford and Harvard: seniority-biased technological change
Two academic studies close the loop. A Stanford analysis of ADP payroll records found a 16% relative decline in employment among early-career workers in the most AI-exposed occupations since late 2022, concentrated in software development and customer service, which are the two fields that historically absorbed the most entry-level white-collar workers. And Harvard economists analyzing résumé and posting data covering 66 million workers across more than 280,000 US firms found that at companies adopting generative AI, junior hiring drops sharply while senior employment keeps growing. They gave the pattern a name worth remembering: ‘seniority-biased technological change.’
One caveat is that part of the entry-level crash coincided with 2023's aggressive rate hikes, so AI isn't the only culprit. That said, the firm-level studies control for those macro trends and the seniority split shows up anyway, and keeps widening.
Two stories, one letter
So what’s the takeaway? Is it Jevons expansion or a hiring collapse? It’s both in a K-shaped future.
Tracey's post described a divergence in demand among software engineers. The labor data shows that the same dynamic is permeating the org chart. The up-stroke is aggregate demand for software and knowledge work rising, senior professionals more productive and more valuable than ever. The down-stroke is the entry point contracting in nearly every dataset that measures it. Jevons Paradox is real. It just isn't distributed evenly, with the new demand being met by experienced people with AI stacks, not by new hires learning the trade.
What 2027 looks like
Through 2027, expect this to shift from an emerging pattern to standard operating procedure. As enterprise licensing for advanced reasoning models matures and integrations become part of default office infrastructure, organizations under pressure to maximize profitability will keep minimizing entry-level hiring because a senior professional equipped with an AI stack can temporarily mimic the output of an entire junior team.
By the end of 2027, the observable evidence will be a visibly thinning pipeline: a third consecutive year of shrunken analyst classes, campus programs paused or quietly discontinued, junior postings that never recovered to their 2022 baseline, and senior-plus-AI as the default staffing model in corporate law, investment banking, and software engineering.
What won't be visible by 2027 is the bill. The juniors not hired in 2025 through 2027 were scheduled to become the mid-level professionals of 2030 and beyond. That shortfall is being locked in right now, but it won't show up on an org chart until the end of the decade — which is precisely why quarterly-focused planning doesn't see it, and why somebody outside the quarterly cycle needs to point it out.
The end of apprenticeship as we knew it
The immediate casualty of this optimization is the fundamental process of apprenticeship. Historically, a professional's career followed a predictable trajectory: beginners paid their dues executing the low-stakes, time-consuming drudge work of research, data compilation, and document formatting. Those foundational tasks built the pattern recognition, industry knowledge, and cognitive stamina required to become journeymen, and eventually masters.
By automating the bottom of the ladder, companies are inadvertently removing those rungs and building workforces with no scheduled replacements for their middle.
Because the execution of basic tasks can now be offloaded to an algorithmic layer, the very definition of entry-level knowledge work has to change. The junior professional of the near future can no longer be a raw producer of draft assets, code snippets, or basic financial models. That layer of work is shrinking fast, and there's little reason to expect it to rebound.
What replaces the bottom rung
Entry-level work will have to evolve into a discipline of forensic auditing, structural verification, and boundary testing.
Junior professionals will increasingly begin their careers at what used to be the midpoint of a project. Their daily responsibility shifts from generating a standard contract, spreadsheet, or codebase to aggressively auditing machine-generated output for hallucinations, logical fallacies, edge-case failures, and hidden compliance liabilities. They become system operators who know how to orchestrate, prompt, validate - and sometimes break - the automated systems beneath them.
To survive this shift, early-career workers will have to cultivate an advanced level of critical skepticism and systems fluency. They'll bypass the mechanics of typing the prose or building the formula, and go straight to the higher-order judgment tasks required to make sure the machine's output doesn't drive the enterprise in the wrong direction.
The bottom rung isn't coming back. The organizations that thrive will be the ones that deliberately build a new one.