More than half of the companies that cut jobs for AI now wish they had not. The real story is not about the technology. It is about what happens when organizations stop asking the harder questions. 

In 2025, 55,000 U.S. workers lost their jobs because their employers believed AI would replace them. Not because AI had proven it could. Because they believed it eventually would. 

By the end of that year, Forrester’s research found that 55% of those same employers regret the decision. Half are expected to quietly rehire by the end of 2026, often at lower wages, often offshore, and almost always under a different job title. The technology that was supposed to render human work redundant turned out to need a great deal of human work to function at all.

This is the story that most post-mortems get half right. They focus on the capabilities gap: AI was not ready, the tasks were too complex, and the institutional knowledge was irreplaceable. All of that is true. But the deeper failure was not technological. It was a failure to ask the right question before making an irreversible decision. 

The right question was not: Can AI do this? It was: Should we, and if so, for whom? 

What Actually Went Wrong 

Between 2022 and 2024, a Swedish fintech eliminated approximately 700 customer service positions, replacing them with an AI assistant it publicly described as doing the work of 700 people. The announcement was received as proof of concept for AI-driven workforce reduction at scale. 

By early 2025, internal reviews and customer feedback told a different story. Satisfaction ratings had fallen. Customers were frustrated by robotic responses, inflexible scripts, and the particular indignity of having to repeat their problem to a human after the bot had failed to resolve it. Their CEO eventually said it plainly: “We focused too much on efficiency and cost. The result was lower quality, and that’s not sustainable.” The company began rehiring. 

The technology that was supposed to render human work redundant turned out to need a great deal of human work to function at all.

This company was not uniquely reckless. The pattern it established has since been documented across enough organizations that Forrester built it into their 2026 workforce predictions as a structural expectation rather than an outlier. Advanced AI agents currently achieve around a 58% success rate on single-step tasks. For multi-step tasks requiring critical thinking, contextual judgment, and nuanced problem resolution, that figure drops to approximately 35%. Companies made massive workforce decisions based on projections of AI capability that the technology has not yet reached and, in most cases will not reach on the timelines assumed. 

What filled the gap was not AI. It was lower-paid workers, often offshore, managing the interactions that the system could not handle. The cost savings that looked compelling on an investor slide were quietly offset by customer churn, brand damage, loss of institutional knowledge, and the expense of rebuilding the human capacity that had been discarded. 

The Question Nobody Asked

There is a specific kind of organizational failure at work here, and it is worth naming directly. It is not naïvety about technology. It is the subordination of strategic judgment to the appearance of innovation. 

Some organizations cut staff because AI genuinely enabled them to deliver better outcomes with leaner teams. A smaller number did it because investor pressure, earnings calls, and the gravitational pull of a dominant narrative made it feel like the thing that forward-looking companies do. Forrester’s own language is pointed: “Too often, the C-suite lays workers off for the future promise of AI.” Not the current capability. The future promise. 

The question was not: can AI do this? It was: should we, and if so, for whom? 

That distinction matters enormously, because it reveals what the decision was actually optimizing for. When organizations cut human capacity ahead of proven AI capability, they are not optimizing for better outcomes for clients. They are optimizing for a metric: cost-per-head, margin improvement, the optics of being seen to lead on AI transformation. The client, whose experience had degraded, was not the primary consideration. The shareholder presentation was. 

This is the AI regret story that is not being told clearly enough. It is not primarily a story about technology that overpromised. It is a story about organizations that abdicated the leadership responsibility to ask what they were actually optimizing for, and whether the clients and customers they serve would agree that it was the right thing. 

Augmentation Is Not a Softer Version of Replacement 

The industry has largely settled on augmentation as the corrective narrative: use AI to handle routine, high-volume tasks so that humans can focus on the judgment-intensive work that machines cannot do. This is directionally right. But it carries its own risk of being applied superficially, as a reframing exercise rather than a genuine strategic rethink. 

Augmentation done well asks: how do we use this technology to deliver a meaningfully better experience to the people we serve? Augmentation done poorly asks: how do we use this technology to reduce our cost base while maintaining the appearance of quality? These are not the same question, and the organizations currently rehiring the staff they let go have learned, expensively, that the answer to the second question tends to produce the outcomes Klarna experienced. 

PwC’s 2025 Global AI Jobs Barometer, analyzing nearly a billion job postings across six continents, found that industries most exposed to AI are seeing 3x higher revenue growth per employee than the least exposed industries. Not because they cut people. Because they made people more capable. Workers with strong AI skills command a 56% wage premium, more than double the 25% premium from a year earlier. This is what genuine augmentation produces: it compounds human capability rather than substituting for it. 

The organizations producing these results are making deliberate choices about where technology creates value for their clients and where it creates efficiency for their balance sheets, and they are honest about the difference. That kind of honesty requires true leadership. It cannot be automated. 

What AI Cannot Lead 

There is a version of the AI conversation in executive circles that treats leadership itself as eventually automatable: a set of decisions that, with sufficient data and processing power, could be made better by a system than by a person. This version deserves direct challenge, not because it is technologically impossible in some bounded future, but because it fundamentally misunderstands what leadership is. 

Leadership at the executive level is not primarily a decision-optimization problem. It is the exercise of judgment in conditions where the available data is incomplete, the stakeholder interests are genuinely in conflict, the ethical dimensions are not reducible to a framework, and the consequences of being wrong are carried by real people. That is the terrain where leaders earn their place. And it is precisely the terrain where AI, in its current form and for the foreseeable future, cannot function independently. 

Deloitte’s 2026 Global Human Capital Trends report, drawing on 9,000+ business and HR leaders across 89 countries, is clear on this: technology can accelerate analysis and clarify uncertainty, but it cannot replace human purpose, values, and judgment behind choices. Sixty percent of executives now regularly use AI to support their decisions. That is appropriate. What is not appropriate, and what the AI regret data makes vivid, is using AI as a substitute for judgment about whether a decision should be made at all. 

AI can tell you what the data suggests. It cannot tell you what you owe the people your decision will affect, employees and clients included. 

The five capabilities that IE Business School researchers identify as irreplaceable at the leadership level are not soft skills in any meaningful sense of the term. Instinct rooted in lived experience. Intuition that integrates emotional intelligence with contextual judgment. Imagination that envisions what has not yet existed. Integrity that distinguishes what is possible from what is right. Identity that provides the stable moral foundation from which hard decisions can be made without losing the trust of the people who depend on them. None of these are on the near-term AI capability roadmap. All of them are on the job description of every effective executive. 

The Leadership Question at the Center of All of This 

As organizations reckon with what went wrong in the first wave of AI-driven workforce decisions, and as they prepare for the next wave of AI deployment across increasingly consequential organizational functions, one question keeps reasserting itself. 

Who is making the decision about how AI is being used, and are they asking the right question? 

Not: what can AI do for our cost structure? But: what should AI do for the people we serve, and are we being honest with ourselves about the difference? 

Not: what will this look like to the market? But: what will this feel like to the client whose experience we are proposing to change? 

Not: are other organizations doing this? But: does it make us genuinely better at what we do? 

These are leadership questions. They require a kind of judgment that is not currently available in any model, however capable. They require a leader who understands the technology well enough to deploy it strategically, who understands their organization’s clients well enough to know where human interaction is irreplaceable, and who has the independence of mind to resist the pressure to follow the herd when the herd is moving in a direction that does not serve the people the organization exists to serve. 

That leader is not interchangeable with a prompt. And finding them is not a transaction. 

Let Us Help You Find a Leader Who Knows the Difference

The organizations navigating AI thoughtfully in 2026 are not the ones with the most sophisticated technology. They are the ones with leaders who ask better questions, who understand the difference between augmenting value for clients and extracting value for shareholders, and who have the judgment to know when the technology serves the mission and when it undermines it. 

At M SEARCH, we have spent 18+ years finding leaders who know the difference between following the pack and leading with purpose. In a moment when the pressure to deploy AI is intense, and the cost of getting it wrong is visible in real-time, that distinction is more consequential than ever. 

If your organization is navigating an executive transition in a world reshaped by AI, or if you are asking what kind of leader you actually need for the next chapter, we would welcome the conversation.