That is the story underneath the 2025 layoff wave. The aggregate numbers will likely vindicate AI in the long run. The Future of Jobs Report 2025 from the World Economic Forum projects a net gain of 78 million jobs globally by 2030, and U.S. labor productivity is measurably trending higher since ChatGPT launched. But the macro story is not the story for any individual company. The question a VP or CFO should be asking is narrower and harder: whose productivity curve are we cutting against, and have we earned the right to use it as our denominator?

For most companies, the honest answer is no.

The productivity case for AI, at the top of the distribution, is serious. Jamie Dimon told Bloomberg TV in October 2025 that JPMorgan spends roughly $2 billion a year developing AI and captures roughly $2 billion in annual savings, a matched 1:1 ratio he called "the tip of the iceberg." At the Cisco AI Summit in January 2025, Goldman Sachs CEO David Solomon said AI can now draft 95% of an S-1 filing in minutes. That work previously took a six-person team two weeks. "The last 5% now matters because the rest is now a commodity," Solomon added.

These are not exaggerations. They are also not your baseline. JPMorgan employs roughly 2,000 AI and machine-learning specialists and has been building AI capability since 2012. Goldman has 11,000 engineers inside a 46,000-person firm. These outcomes belong to institutions with data infrastructure, engineering benches, and workforce redesign capabilities that the median enterprise lacks.

The St. Louis Fed's 2025 research on generative AI productivity is the more representative number for most companies. Using nationally representative survey data, Bick, Blandin, and Deming estimated that self-reported time savings from generative AI correspond to roughly 1.6% of U.S. work hours, which, fed into a standard production model, implies about a 1.1 to 1.3% labor productivity boost since late 2022. Real, but small, and unevenly distributed.

The gap between those two numbers, JPMorgan's 1:1 ROI and the economy's 1.3%, is the gap most executives are waving away when they sign off on headcount decisions.

The implementation gap is now well-documented. Boston Consulting Group's October 2024 survey of 1,000 senior executives across 59 countries found that 74% of companies have yet to demonstrate tangible enterprise-level value from AI. The winners, BCG noted, concentrated on a small number of transformative use cases and invested 70% of their resources in people and process redesign, not tooling.

The Upwork Research Institute's 2024 study of 2,500 workers captured the asymmetry from the other side: 96% of C-suite leaders expected AI to increase productivity, while 77% of employees using AI said the tools had added to their workload. Nearly half (47%) reported not knowing how to achieve the productivity gains their leadership expected. This is not evidence that AI does not work. It is evident that, in most organizations, the capability to convert model output into a clean process change does not yet exist.

The most damning finding comes from Harvard Business Review in January 2026. Thomas Davenport of Babson College and Laks Srinivasan of the Return on AI Institute surveyed 1,006 global executives in December 2025 and found that AI-attributed layoffs were "almost completely in anticipation of AI's impact" rather than tied to measured AI performance. Only 2% of organizations reported large headcount reductions linked to actual AI implementation. Forrester's "Predictions 2026: The Future of Work" report found that 55% of employers who conducted AI-attributed layoffs already regret the decision, and Forrester predicts half of those roles will be quietly rehired by the end of 2026, often offshore or at lower salaries.

Read those two data points together. Permanent workforce decisions are being made on the basis of a promise, and more than half of the executives making them already wish they hadn't.

Klarna is now the reference example, and the facts matter more than the narrative. Between 2022 and 2024, the Swedish BNPL company replaced work equivalent to roughly 700 customer service agents with an AI assistant built in partnership with OpenAI. Its overall headcount fell by about 22%, to around 3,500, due to a mix of AI-driven attrition and hiring freezes.

In May 2025, CEO Sebastian Siemiatkowski publicly reversed course. "We focused too much on efficiency and cost," he told Bloomberg. "The result was lower quality, and that's not sustainable." Klarna began recruiting human agents again, piloting an "Uber-style" remote-flexible model to rebuild the support operation.

The Siemiatkowski quote is the part everyone has seen. The operational lesson underneath it is the part most readers haven't absorbed. What Klarna automated was not uniform work. Customer service is bimodal: a large share of routine queries that AI handles well, and a smaller share of escalations, disputes, and edge cases where a wrong answer becomes a regulatory, reputational, or litigation event. The AI handled the routine volume cleanly. It failed on the smaller share of interactions that carry disproportionate consequences relative to the dollar. Rehiring solves the headcount issue. It does not restore the tenure curve, the institutional memory of recurring fraud patterns, or the relationships with repeat merchants that took a decade to build.

Klarna could rehire because it is a well-known consumer brand in a liquid labor market, and because it was public-facing enough that its reversal became a case study rather than a slow bleed. A regional insurer, a mid-market manufacturer, or a specialty healthcare provider does not have those advantages. The replay value in their labor market is much lower.

The middle of the org chart is taking disproportionate damage. Live Data Technologies found that middle managers comprised about 32% of 2023 layoffs, compared with 20% in 2019, a 60% increase in management's share of workforce reductions. Revelio Labs, which tracks more than 100 million employment profiles, reported a 42% drop in middle-management job postings between April 2022 and October 2025, with no signs of recovery. Gartner projected in October 2024 that 20% of organizations will use AI to flatten organizational structures by the end of 2026, eliminating more than half of their middle-management positions.

As we argued in The Middle Manager Is Breaking. And Your Strategy Is Breaking With Them; that layer is not primarily coordination overhead. It is where strategy gets translated into frontline behavior, where cross-functional conflicts actually get resolved, and where the next cohort of senior leaders learns to operate. Cut the middle rungs and the capability cliff shows up on a multi-year lag, well after the executives who made the call have moved on.

The practical test: if you are the CFO, CHRO, or CEO signing off on AI-justified cuts, there are two questions you should be able to answer in writing before the decision goes through.

One: What is your organization's measured productivity delta from AI, expressed as a percentage of labor hours, and how did you measure it? If the answer involves case studies from JPMorgan, Goldman, or any company outside your industry and maturity band, you are using someone else's denominator. If it involves pilot programs that have not been deployed in production at scale, you are pricing against potential, not performance. Most organizations cannot answer this question rigorously. They do not have the measurement apparatus in place. That is not a reason to guess. It is a reason to treat the absence of data as its own finding: if you cannot measure the gain, you have not earned the right to price against it. The HBR data suggests the vast majority of organizations fall into this category.

Two: Have you separated reversible from irreversible cuts? Contractor ramp-downs, hiring freezes, and attrition-driven shrinkage are reversible in roughly six months. Severing experienced frontline staff and collapsing management layers is not, at least not at the tenure-and-judgment level you had before. Klarna could rehire in a liquid labor market. Most organizations cannot pull 700 experienced people back at the same quality bar.

A useful corollary: inside any function, identify the minority of roles that absorb the majority of edge-case risk. In customer service, that is the escalation tier. In operations, it is the people who recognize recurring failure patterns. In finance, it is the reviewers who catch the exceptions. Those are the last roles to automate, not the first. Klarna's public reversal is what it looks like when that order is inverted.

The aggregate numbers will likely vindicate AI. That is cold comfort for the individual company that cut its most experienced people, flattened its management bench, and degraded the customer relationships that took a decade to build, all on the assumption that a productivity curve it had not yet achieved would arrive in time.

The reckoning, when it comes, will not look like a crisis. It will look like a slow loss of capability: quality drift in the functions that used to catch the hard cases, a thinner pipeline of people who know how to run the business, and a rehire bill larger than the original savings. By the time that pattern is visible on the balance sheet, the executives who made the call will have moved on. The denominator they cut against was never theirs.