The Stanford 2026 AI Index, released this week by researchers at Stanford Human-Centered AI, is the most comprehensive annual snapshot of AI’s real-world economic impact available. It draws on government data, academic research, and industry surveys. It does not have a political agenda. And this year, its findings are harder to argue with than ever before, and harder to feel good about if you are not already in a position of advantage.
Three data points from the index, read together, describe an economic split that is no longer a projection. It is a documented outcome.
AI Productivity Gains Are Real and Measurable
The AI sceptics who argued that large language models produce noise without substance are losing the empirical argument. The Stanford index documents productivity gains that are specific, measured, and significant.
On the SWE-bench Verified coding benchmark, which tests whether AI can resolve real software engineering problems, performance rose from 60% to near 100% in a single year. That is not incremental improvement. That is a step change in capability that compresses what used to take a senior developer hours into something an AI system handles in minutes.
More directly relevant to the workplace: AI is boosting productivity by 26% in software development and 14% in customer service, according to research cited in the index. These are not lab results. They are measured outcomes from deployed systems in real organisations.
The productivity gains are real. That matters because it rules out one of the more comfortable explanations for AI disruption: that companies are overhyping tools that do not actually work in order to justify cost-cutting they would have done anyway. Harvard Business Review noted in January 2026 that many layoffs are being made in anticipation of AI performance rather than because of it. But the Stanford data makes clear that the performance, in some domains, is now arriving.
Entry-Level Software Jobs Are Disappearing First
Employment for software developers aged 22 to 25 has fallen nearly 20% since 2022, according to economists at Stanford whose work is cited in the index. That is not a rounding error. That is a structural shift in who can enter the technology workforce.
The pattern is consistent with how automation has historically entered labour markets: it removes the bottom rungs of the ladder first. Junior developers writing boilerplate code, entry-level analysts producing standard reports, junior customer service agents handling routine queries. These are the roles where AI’s current capabilities are most directly substitutable. They are also the roles through which most people build the experience required to reach senior positions.
The consequence, which the index does not spell out but which follows logically, is that the pipeline into senior technology roles is narrowing. The people who currently hold senior positions gained them by spending years in junior roles that are now being automated. The generation entering the workforce now does not have the same route available. As we documented in our earlier analysis of Gen Z’s entry into the workforce, the disruption is not evenly distributed across career stages. It lands heaviest at the beginning.
One-third of employers surveyed in the index expect workforce reductions over the coming year. That number deserves scrutiny: a third of employers, surveyed directly, saying reductions are coming. This is not speculation about what AI might do. This is employers stating their plans.
74% of the Gains Are Going to 20% of Companies
The productivity data and the job loss data are both real. What connects them, and what turns them from two separate trends into a single coherent picture, is the distribution finding.
PwC’s 2026 AI Performance study, based on a survey of 1,217 senior executives across 25 sectors and multiple regions, found that the top 20% of AI-ready companies are capturing 74% of all AI-driven economic returns. The most AI-fit companies are achieving a 7.2 times performance boost over their peers, combining AI-driven revenue growth and cost reductions. The full study is available at PwC’s website.
The bottom 80% of companies are not failing to adopt AI because they are incompetent or resistant. They are failing to capture returns because the advantage in AI performance compounds. Companies that built strong data infrastructure, AI talent pipelines, and integration capabilities three or four years ago are now pulling so far ahead that the gap is not closeable through incremental investment. The leading companies are focused on growth, not just productivity. The laggards are chasing efficiency gains that the leaders have already moved past.
This is the mechanism behind the jobs picture. The companies capturing the gains have the resources to invest further, hire the AI talent that remains scarce, and expand into new markets. The companies not capturing the gains are under margin pressure and are cutting headcount to survive. The workers being laid off are not, in most cases, employees of AI winners. They are employees of companies that are losing to AI-equipped competitors and cutting costs to compensate.
Why Experts and the Public See This So Differently
The Stanford index includes a data point that is, in some ways, the most revealing of all: 73% of US AI experts view AI’s impact positively. Only 23% of the public shares that view.
That is not a 50-point gap in opinion. It is a 50-point gap in lived experience. AI experts, by definition, sit inside the 20% of companies and institutions where AI is performing and generating returns. Their salaries have increased. Their tools have improved. Their careers are in demand. Of course they view AI’s impact positively. They are experiencing the upside.
The 77% of the public that does not view AI’s impact positively is not misinformed or technophobic. They are reporting, accurately, that the gains described in the Stanford index are not reaching them. Their jobs are less secure. Their entry-level colleagues are disappearing. Their employers are cutting, not growing. They are experiencing the disruption without the windfall.
This gap matters because it shapes the policy and regulatory environment around AI. When experts consistently characterise public concern as ignorance rather than rational response to documented inequality, they lose the public trust that is required to govern the technology responsibly. The Stanford data does not support the expert framing. It supports the public’s.
What This Means for Workers and Businesses Outside the Top 20%
The PwC data suggests the window for closing the gap between AI leaders and laggards is narrowing, not widening. The compounding advantage of early AI investment means that late adopters face an increasingly steep climb. For businesses, the strategic implication is that incremental AI adoption, adding tools at the margins without rebuilding how work is structured, is unlikely to generate the returns that justify the cost.
For workers, particularly those early in their careers, the Stanford data reinforces the case we have made in our analysis of Klarna’s AI experiment: the roles being automated first are the structured, high-volume, low-judgment tasks that form the base of most career ladders. Building capability in areas that require judgment, accountability, and contextual reasoning is not optional career advice. It is the only documented path to remaining in the portion of the labour market where AI is generating growth rather than eliminating positions.
The Stanford 2026 AI Index does not predict what will happen next. It documents what has already happened. The productivity gains are real. The job losses at the entry level are real. The concentration of returns at the top is real. These are not separate stories. They are the same story, told from three different vantage points.
This article draws on the Stanford HAI 2026 AI Index Report, PwC’s 2026 AI Performance Study, and coverage by MIT Technology Review. Analysis and interpretation reflect the author’s reading of publicly available data and should not be treated as financial or professional advice.