Precisely how AI will impact the workforce in the short to medium term is unclear. Many layoff announcements are better understood as AI-washing: ordinary restructuring initiatives packaged in AI language to reassure investors that leaders have a grip on the coming technological change. Goldman Sachs estimates that AI reduced monthly U.S. payroll growth by only around 16,000 jobs over the past year, adding roughly 0.1 percentage points to the unemployment rate—a drag that includes slowed hiring as well as outright job elimination. These numbers may well increase, but for now, we are much earlier in the process than the prevailing narrative suggests.
The fact that we are early, however, and that the scale of layoffs is smaller than the news cycle implies, does not mean all these layoffs are wise. Many companies are already making real workforce changes in response to AI—and they are frequently getting them wrong. An international survey of 1,000 business leaders published in April 2025 found that 55% of businesses that laid off workers due to AI implementations admitted they had made wrong decisions around staffing. In a more recent survey of 600 HR leaders who made AI-driven layoffs in the 12 months to February 2026, only 8.4% said the restructuring had delivered as promised and that they would do the same again. Costly reversals are already happening. Gartner predicts that half of companies that cut customer service staff due to AI will rehire workers. Forrester, meanwhile, expects that half of all AI-attributed layoffs will be reversed, although the replacement positions may be offshored or come with lower salaries.
These figures are not outliers. They are supported by other reporting that point in the same direction: companies are often not getting AI rightsizing right.
The problem is not simply that companies are moving too quickly or cutting too deeply. It is that many are approaching AI-driven restructuring from the wrong starting point: headcount rather than work. A more durable approach begins by understanding how AI changes individual tasks and business processes, then redesigning the organization around the capabilities—human and technological—needed to execute them. As we will argue, that requires companies to rethink how they define work, connect AI decisions to business strategy, build a realistic understanding of what the technology can and cannot do, and preserve enough flexibility to change course as both the technology and the evidence evolve.
Where Companies Go Wrong
Most companies that have already attempted AI-driven workforce changes are making one or more of a recognizable set of mistakes.
- Anticipatory cuts. Many companies are restructuring their workforces based on forecasts and market narratives about future AI capabilities rather than on evidence from their own active implementations. A recent survey of over 1,000 global executives found that only 2% of organizations had made large headcount reductions based on actual AI implementation, while 60% had laid off people or reduced hiring with an expressed connection to AI. Block’s elimination of nearly half its workforce in February 2026 is the highest-profile example: CEO Jack Dorsey attributed the cuts to AI but framed them as a bet on where the technology is heading rather than what the technology could do for the company right now. (He also said some of the redundancies were the result of overhiring due to the pandemic.)
- Cutting without understanding AI capabilities. Even when companies move beyond anticipation to decisions rooted in current AI implementations, layoffs are often made without adequate understanding of AI’s capabilities and limitations in context. 54% of HR leaders said they would have made better choices with a more informed understanding of what AI could do. Nearly a quarter admitted that layoff decisions were made without first testing the business case. And 55% said their layoffs were not worthwhile because AI required more human oversight than expected.
- Poisoning the internal culture. A survey of 5,400 U.S. employees found that engagement drops to 44% at organizations where layoffs (for any reason) have occurred, compared to 51% overall, and that 58% of unaffected employees become more likely to look elsewhere. Framing cuts as a response to the emergence of AI can undermine remaining employees’ psychological safety, leading them to disengage from their jobs and to avoid experimenting with AI-driven productivity. AI-washing risks generating the same cultural damage without delivering any corresponding transformation at all.
A recent Gartner survey found that 80% of companies made reductions to their workforces when implementing new “autonomous business capabilities,” but that there was no correlation between lower employee numbers and return on the investments. Until organizations measure success by decision quality, organizational capability, and learning rather than by cost savings and headcount, they will continue to make the same mistakes.
A Rightsizing Playbook
No single organization has fully cracked AI-era rightsizing, and the companies experimenting most seriously with new workforce models are themselves operating under considerable uncertainty. But their approaches, taken together, point toward a more disciplined way of making these decisions—one that replaces top-down headcount targets with a closer examination of how work gets done, where AI can genuinely create value, and where human capabilities remain essential. The following principles offer a framework for doing that.
Redefine the unit of work.
When organizations begin a rightsizing process, they typically set targets for the number of positions to be cut, with the goal of replacing the core function of the human role with an AI alternative. But most white-collar roles involve multiple tasks. Workers at every level typically communicate, collaborate, and make the kinds of judgments that are hard to reduce to algorithmic functions.
Eliminating roles rather than tasks will often mean that the AI replacement will only be able to perform the most visible parts of a job. The organization will find itself needing to hire staff back either to fill newly constituted roles that cover the discovered gaps, or to fill the old role when the visible part of the job cannot function without the invisible parts.
The scale of this failure shows up in the survey data: one in three HR leaders reported losing critical skills and expertise along with their laid-off employees. The failure to predict this loss tells us something important: organizations that did not know which skills they were losing never fully understood what the eliminated roles actually did. It is therefore vital to decompose roles into their component tasks and to understand which activities can be replaced by AI in an overall business process and which cannot. Only then is it possible to determine what the new workforce should look like. A headcount reduction plan can then be designed to ensure workflows are not disrupted.
Citigroup’s approach of targeting specific processes for improvement is an example of this in practice. The bank started with an initial list of 50 processes flagged for greater automation. Rather than starting from a target headcount reduction, Citi identified where technology and process redesign could change the work, and made staffing decisions on that basis. The analysis came first; the restructuring followed.
Develop a business strategy, not an AI strategy.
Asking “What can AI do to reduce our headcount?“ starts from the wrong place, treating AI as an isolated cost-reduction tool that is disconnected from the business’s broader strategy. Instead, businesses should treat AI as an integral part of the general suite of capabilities available to them to deliver on their goals, then determine the optimal combination of AI and human resources required to achieve them. Sometimes the answer will dictate direct reductions in headcount and costs. But in others, resources freed up by AI will be best deployed by employing more people in areas where human involvement creates value that AI cannot, or by redirecting the efforts of existing staff to different tasks.
Redeployment of staff is not cost-free, typically requiring significant investment in reskilling, retaining institutional knowledge, and supporting company culture and morale. But Fortune reports that Walmart has gambled that redeployment will be a better investment than hiring replacements or ceding the work entirely to AI: the company has invested nearly $1 billion in AI training for its 1.6 million U.S. workforce, and CEO John Furner expects the company to “have roughly the same number of people we have today” in three to five years.
JPMorgan Chase’s strategy provides an example of this approach. Rather than starting from the goal of using AI to reduce headcount, the bank began by looking at how it could most effectively redeploy staffing resources that had been freed up by the efficiencies created by AI implementations. This led to reductions in some areas of the workforce and increases in others. For instance, according to a CNBC report, while operations roles were reduced by 4% and support roles by 2%, client-facing and revenue-generating teams expanded by 4%. The firm’s overall headcount has remained steady at just over 300,000. Maintaining a stable headcount should not be a business goal in its own right. Rather, total staffing numbers should fit the needs of a cohesive business strategy employing humans and AI to create value where appropriate. In the case of JPMorgan Chase, CEO Jamie Dimon expects total headcount to fall over the next five years.
Understand what AI can do.
Leaders need to get AI-savvy without delay. That means understanding the technical capabilities of this technology in sufficient depth to make informed decisions about potential implementations in the organization’s unique context.
A 2023 study of thousands of board-level executives found that just 8% disclosed having a substantial level of conceptual knowledge regarding any dimension of AI technology. More recent evidence suggests that competence at the highest levels is growing, but only slowly. A 2025 Gartner survey of 456 CEOs shows that CEOs, at least, are not confident in the AI competence of their senior leadership teams, with only 44% deeming even their own CIOs to be AI-savvy. At the board level, the number of members disclosing AI expertise grew only from just 1.5% to 2.7%.
When leaders do not understand what AI can do, they either delegate workforce decisions to technical teams who lack the business context to make them well, or they make those decisions themselves based on flawed assumptions.
Some organizations are taking concrete steps to close the gap. Before developing its generative AI strategy, Schneider Electric held 56 workshops over four weeks that brought AI experts together with executives across 15 functions, grounding AI literacy in specific business contexts rather than abstract training. Others are investing in more traditional AI literacy training programs: UBS, for instance, has sent senior leaders to Saïd Business School at the University of Oxford for an AI Senior Leadership Journey, while IKEA set an initial target of AI literacy training for 500 leaders, a target that has since been expanded. Approaches like these are a prerequisite for the kind of task-level analysis that responsible rightsizing demands.
Understand what AI cannot do.
While it is true that some AI shortfalls are just the limitations of current models, certain boundaries are highly likely to endure. Some functions are inherently human because they require moral agency and accountability. Choosing values, exercising ethical judgment, contextualizing decisions within relationships and institutional history, and knowing when to override a recommendation are not tasks that can be delegated to algorithms, however sophisticated. Understanding where these boundaries lie before eliminating human roles is essential to avoid decisions that will later have to be rolled back.
In most organizations, accountability functions are carried out by middle managers. These individuals bridge strategy and execution, own implementation decisions, and serve as the organization’s conscience, overseeing the gap between what leadership intends and what happens on the ground. As we have argued elsewhere, automating this layer destroys a critical ethical infrastructure that organizations cannot afford to lose. Yet a 2025 Korn Ferry survey found that 41% of companies have already reduced managerial layers (in part because of AI).
The importance of middle management and other human tasks is illustrated by the survey findings above that many companies underestimate the need for human oversight. They discover that the oversight capability they needed was exactly the judgment and translation capacity they had been removing.
Designing Work for Reversibility
When roles are eliminated and experienced employees leave, they take with them institutional knowledge, working relationships, and contextual understanding that cannot easily be replaced. Rehiring is not the reverse of firing—it is slower, more expensive, and the replacement rarely brings the same depth of organizational understanding. Given the pace at which AI capabilities are evolving and the uncertainty about which forms of restructuring will work and which will not, organizations need workforce structures that can flex as the picture changes. That means using attrition and internal redeployment before layoffs, phasing cuts rather than making them in a single wave, and treating reversibility as a design criterion rather than an afterthought.
Citigroup’s approach illustrates what phased, reversible restructuring looks like in practice. The bank began its transformation in 2022, and it is still underway in 2026. While a major focus of the transformation was always more common global data and aggressive adoption of technology (including machine learning and blockchain), Citi wisely incorporated generative AI into its plans in 2023. Its headcount goals are ambitious—in early 2024 it announced plans to cut up to a fifth of its workforce over three years, while reducing costs by $2.5 billion—but it is making the reductions relatively slowly. Thus far it has trimmed about 15,000 employees from an initial base of 240,000.
Citi is also reducing its reliance on contractors. Roughly half of the company’s 50,000-strong tech workforce were contractors when the initiative began, and its goal during the transformation is to reduce that proportion to 20%. Citi tries to fill open positions with existing employees whenever possible, with around 50% of job openings filled by current staff.
