The most revealing thing about the state of AI in the workplace right now is not the technology itself. It is that the co-founders of the world's most important AI company cannot agree on what their own product is going to do to your headcount.
Dario Amodei, chief executive of Anthropic, has said publicly and repeatedly that artificial intelligence could drive unemployment to 20% within five years - a level not seen since the Great Depression. His co-founder Jack Clark, speaking this week at the Semafor World Economy Summit in Washington alongside more than 450 chief executives and government leaders gathered for the IMF and World Bank spring meetings, took a rather different line. Clark, who leads a 30-person internal think tank studying AI's effects on work, would go no further than noting "some potential weakness in early graduate employment" in certain sectors. Mass unemployment, he insisted, is a choice, not an inevitability.
Both men are, in their own way, right. And that is precisely the problem for every employer trying to make sensible decisions about their workforce right now.
The honest answer to the question in this headline is: we do not yet know. What we do know - from a growing body of surveys, economic data and real-world deployment evidence - is rather more useful. The companies getting this right are not the ones waiting for the debate to be settled. They are the ones who have stopped asking whether AI will change everything and started asking how to manage that change without leaving their people behind.
Start with the case for optimism. A new survey by the Federal Reserve Bank of Atlanta, drawing on more than 700 corporate executives across sectors ranging from finance to manufacturing, found that companies reported productivity gains of 1.8% from AI in 2025, with larger increases expected this year. The overall impact on headcount, the researchers found, was "close to zero." Most businesses are getting more out of the same number of people. Nobody is being marched out of the building.
Now the case for concern. Morgan Stanley surveyed 935 executives across four countries and found something rather different: AI adoption had already led to the elimination of 11% of jobs at the companies studied, with a net 4% decline in headcount globally. The researchers called it an "unexpected outcome," given that earlier analysis had predicted AI would be a net positive for employment. Clearly, that was too optimistic.
The picture that emerges from these conflicting datasets is not one of catastrophe or of uncomplicated good news. It is one of uneven impact - concentrated in particular roles, particular sectors and particular seniority levels. Large companies, the Atlanta Fed found, plan to reduce employment by 0.8% this year directly because of AI. Entry-level white-collar roles are showing the earliest signs of pressure. Graduate hiring in knowledge-intensive industries is beginning to soften, as Clark himself acknowledged. The displacement, when it comes, is not arriving as a sudden shock but as a slow tide - easy to miss until it is already at the door.
Here is something counterintuitive that should keep every HR director up at night. Your retention figures may look healthy not because your workforce is engaged, but because they are frightened.
ManpowerGroup's 2026 Global Talent Barometer found that while AI usage among workers jumped 13% to reach 45% of the workforce, confidence in using the technology fell by 18% over the same period. The result is a phenomenon that researchers have started calling "job hugging" - 64% of workers say they intend to stay with their current employer not because they are thriving, but because they are seeking stability in uncertain times. They are staying put and holding on.
Chart 1 of 3
AI adoption jumped 13 percentage points in one year, reaching 45% of workers globally. Over the same period, confidence in using the technology fell sharply by 18 points - the first overall decline in worker confidence in three years.
Source: ManpowerGroup Global Talent Barometer 2026 (13,918 workers across 19 countries, fieldwork September-October 2025)
ADP's survey of 39,000 workers globally puts numbers on the underlying malaise: fewer than one in five employees are fully engaged, and only 22% feel secure in their jobs. MetLife's 24th Annual Employee Benefit Trends Study, published in March, found that 61% of employees are worried about the ethical and safety risks of AI - bias, misinformation, lack of accountability - up five percentage points from a year ago. Yet 80% of employers report that AI tools are now part of everyday tasks.
The gap between those two numbers is the crisis hiding inside the productivity story. Employers have deployed the tools. They have not, in most cases, done the harder work of managing the human response to them.
ADP's research surfaces one particularly striking paradox: daily users of AI - the employees most fluent in the technology - were four times more likely than non-users to say they felt less productive than they could be. The tools are being used. The integration is not working. And in the absence of clear signals from employers about what the technology means for their futures, workers are filling the silence with anxiety.
The split between Amodei and Clark on unemployment is not a minor technical dispute. It reflects a genuine uncertainty at the frontier of the technology - and it matters for employers because it means there is no authoritative voice to defer to.
Amodei's view rests on a conviction that AI will advance much faster than most people expect. If he is right, the transformation of knowledge work will be compressed into a very short window, leaving little time for labour markets or policy responses to adjust. Clark's counterargument - that big changes in employment "take a long time to filter through to the economy" - is more historically grounded, and arguably more useful for practical planning purposes.
What makes Clark's position interesting for employers is not just the timeline but the framing. He insists that the outcome is, to a significant degree, a matter of choices made by companies and governments rather than a technological determinism beyond human influence. If AI grows the economy substantially - and the evidence that it is doing so is already visible in Anthropic's own trajectory, which saw its annual recurring revenue more than double from $9 billion to over $20 billion in just three months between December 2025 and March 2026 - then there is economic headroom to fund re-skilling, create new roles and support those whose existing positions are displaced.
That is a reasonable argument. It is also, to be blunt, not something most employers can wait for governments to execute on their behalf. The companies navigating this best are not the ones sitting on their hands waiting for policy clarity. They are the ones treating workforce transition as a management responsibility rather than a political problem.
Chart 2 of 3
Employees whose organisations invest visibly in their skills development are 5.3 times more likely to feel secure in their roles - the largest single driver of workforce confidence identified in ADP's global research.
Sources: ADP People at Work 2026; ManpowerGroup Global Talent Barometer 2026; MetLife Annual Employee Benefit Trends Study 2026
Four practices distinguish organisations that are genuinely benefiting from AI from those that have acquired the tools and achieved very little.
The first is redesigning work rather than just adding software. The industries seeing the largest productivity gains - finance and high-skilled services chief among them, according to Morgan Stanley - are those that have rebuilt workflows around AI rather than layering it on top of existing processes. Industry data suggests that deeply integrated companies are seeing labour productivity grow 4.8 times faster than the global average. The organisations that are not seeing those gains are, overwhelmingly, the ones that bought licences and called it a strategy.
The second is making skills investment visible and loud. ADP's research found that employees whose employers invested in their development were 5.3 times more likely to feel secure in their roles. That statistic alone should justify any re-skilling programme. But ADP's researchers add a pointed rider: the investment is not enough on its own. HR needs to actively communicate it. "HR has to think very carefully around how do we make it valuable, clear, and create the space in the organisation for people to spend the time to invest in themselves," said the firm's researchers. Most organisations do neither. Infosys, which is re-skilling all 300,000 of its employees on AI, takes the approach one step further - new graduates are specifically asked to learn their craft without AI tools first, before those tools are introduced, ensuring that the underlying capability being augmented is genuinely there.
The third is putting someone in charge. Daniel Herscovici, the CEO of Plume, speaking at the same summit as Clark, was unequivocal: "Assigning somebody whose job is to wake up every day and address how to implement the infrastructure is quite important." Research supports this bluntly - chief AI officer roles are now present in 61% of enterprises, and companies with dedicated AI leadership are measurably more likely to achieve productivity gains. Without it, AI strategy fragments across departments and produces inconsistent, underwhelming results.
The fourth - and the one that most employers are least willing to act on - is being honest about what the technology changes. Clark's observation that the most valuable workplace skill for the coming decade is not technical proficiency but the ability to "analyse and connect information across many disparate disciplines" has direct implications for hiring, performance management and how organisations define success. Job descriptions written for the previous decade of knowledge work are already measuring the wrong things. So, in many cases, are performance frameworks and graduate recruitment criteria.
Chart 3 of 3
Having AI and using it well are two very different things. The gap between nominal adoption and daily integration is where most employers are currently losing ground.
Source: Gallup workforce survey, February 2026 (23,717 US employees)
Gallup's CEO Jon Clifton, also speaking in Washington this week, offered the most useful single data point for employers trying to navigate all of this. Fifty per cent (50%) of American workers are now using AI. But only 13% are using it on a daily basis. The gap between those two numbers - between nominal adoption and genuine integration - is where the real story of AI in the workplace is playing out.
The companies most likely to reach the superhuman end of the spectrum this headline poses are not the ones that have simply bought the tools. They are the ones that have thought hard about how to make their people feel capable rather than threatened, productive rather than surveilled, and invested-in rather than provisional.
The 20% unemployment scenario has not gone away. Neither Amodei nor any credible economist has ruled it out. But it is not, as Clark insists, the only possible destination. The employers who accept that the outcome is partly in their hands - and act accordingly - are the ones most likely to end up on the right side of whichever version of the future arrives.
Sources: Federal Reserve Bank of Atlanta corporate executive survey, March 2026; Morgan Stanley AI Adoption Survey, February 2026; ManpowerGroup Global Talent Barometer 2026; ADP People at Work 2026; MetLife Annual Employee Benefit Trends Study 2026; CHRO Association Survey, March 2026; Gallup workforce survey, February 2026; Semafor World Economy Summit, Washington DC, 13 April 2026; Derek Thompson / Plain English interview with Jack Clark, March 2026; TechCrunch, Semafor and Nextgov reporting on Jack Clark remarks, April 2026.
