After ten days with my family, a few beaches in the Dominican Republic, a stop in Orlando, and a quick detour to the Unleash AI Summit in Las Vegas, I’m back at the keyboard. And the first thing waiting for me was a week’s worth of research that deserves more than a LinkedIn post.
In 1987, the economist Robert Solow, who I have read much of this work, made an observation that became famous precisely because it was so obvious: “You can see the computer age everywhere but in the productivity statistics.”
Companies had spent billions on computers. They’d restructured entire departments, laid off typists and filing clerks and switchboard operators. And productivity hadn’t moved. The technology was real but the transformation wasn’t.
We’re watching the same movie again, only the stakes are higher and the speed is faster, and the people losing their jobs this time around are losing them to a future that hasn’t “quite” arrived yet and is definitely not spread “equally”.
What the data actually says this week
On March 24, 2026 (by birthday), a working paper from the National Bureau of Economic Research, drawing on a survey of 750 CFOs conducted with Duke University and the Federal Reserve Banks of Atlanta and Richmond, found that AI-driven job cuts in 2026 are expected to be roughly nine times higher than last year, up from about 55,000 jobs to an estimated 502,000.
Let’s sit with that for a moment, because the number is easy to read past. Last year, a major corporation cutting 100 jobs for AI-related reasons was a headline. This year, that same company is on track to cut 900. Multiply that across thousands of companies making similar decisions simultaneously, and you’re talking about half a million people losing their jobs this year alone, based largely on what AI is projected to do, not what it’s currently proven to do. These aren’t layoffs driven by poor performance or declining revenue. They’re preemptive cuts, bets placed on a technology that’s still finding its footing, which is a bit scary to watch.
And that’s what makes the second number in the same report so important. Workers using AI tools say that time spent on certain tasks has actually increased by up to 346%. Not decreased, but increased. Instead of finishing a task in one hour, some workers are now spending more than four hours on the same task because of AI, whether that’s verifying output, correcting errors, re-prompting to get usable results, or managing the coordination that AI-assisted workflows create. The tool we implemented promised to save time but for many workers right now, it’s consuming more of it.
The reason is almost always the same: the organization adopted AI without embodying it.
This is the distinction that explains everything happening in the data right now. Adoption means the tool got installed, the license got purchased, and someone sent a company-wide email announcing that AI is now available. Embodiment means the way people think about the work actually changed, the workflows changed, the judgment about what’s worth doing changed. One is a procurement decision and one is a transformation. Most companies did the first and called it the second, and the 346% is the outcome.
We didn’t automate the work. We automated on top of the work. Your employees are living that difference every day, and the productivity numbers are telling us exactly what it costs.
The next day, March 25, ADP Research released its “Today at Work 2026” report, based on responses from more than 39,000 workers across 36 countries. Only 22% of workers strongly agreed their job was safe from elimination. Among front-line workers, that dropped to 18%. Even at the C-suite, only 35% felt secure. Nearly four out of five people going to work every day right now don’t believe their job is safe, and they’re doing that work inside organizations where AI is often making their days longer, not shorter. That is not the picture of a workforce being liberated by technology but one being put in crisis mode.
Also, this week March 24, Tufts University’s Digital Planet lab released its first American AI Jobs Risk Index, mapping vulnerability across every major occupation, city, and state in the country. Their projection: 9.3 million U.S. jobs face displacement risk over the next two to five years, with somewhere between $200 billion and $1.5 trillion in annual household income at risk. Writers are at 57% vulnerability. Computer programmers at 55%. And the cities building AI, Silicon Valley, Boston, Seattle, Washington D.C., face the highest displacement risk from the very technology they’re creating.
Harvard Business School published research this month showing that job postings for roles built around repeatable, automatable tasks are down 17%, while postings for roles requiring human judgment and human-AI collaboration are up 22%.
The pattern is clear and consistent: if your work can be described as a checklist, you’re exposed. If your work requires real context and judgment built over years, AI is more likely to make you more valuable, not less. The workers getting hurt most right now are entry-level employees, with a 16% employment decline among 22-to-25 year olds in AI-exposed roles.
We’re eliminating the starting line and then wondering why we can’t find experienced people to promote. That’s not a technology problem but a shortsightedness problem.
And yet a February 2026 NBER working paper found that 90% of executives say AI has had zero impact on employment at their own firms, even as AI-related layoff announcements accelerate across the economy. AI-related stocks have accounted for roughly 75% of S&P 500 returns since ChatGPT launched, which creates a powerful incentive to frame any cost-cutting as AI-driven, whether it actually is or not. Even Sam Altman has publicly acknowledged that companies are blaming AI for layoffs that have nothing to do with it. The narrative has gotten ahead of the reality, and the gap between them is where people’s livelihoods are disappearing.
The numbers, at a glance (for your presentations)
9x — AI-driven job cuts in 2026 are projected to be nine times higher than 2025, rising from roughly 55,000 to an estimated 502,000. (NBER/Duke CFO Survey, March 24, 2026)
346% — Workers using AI tools report that time spent on certain tasks has increased by up to 346%. AI is making work slower, not faster, for a significant share of the workforce. (NBER/Duke CFO Survey, March 24, 2026)
22% — Only 22% of workers globally strongly agree their job is safe from elimination. Among front-line workers, that drops to 18%. (ADP Research “Today at Work 2026,” March 25, 2026)
9.3 million — U.S. jobs projected at risk of AI-driven displacement over the next two to five years, representing between $200 billion and $1.5 trillion in annual household income. (Tufts Digital Planet, March 24, 2026)
57% / 55% — Vulnerability rates for writers and computer programmers, the two highest-risk professional occupations in the Tufts index.
17% down / 22% up — Job postings for automatable roles fell 17% after ChatGPT’s launch, while postings for human-AI collaboration roles grew 22%. (Harvard Business School / HBR, March 2026)
16% — Employment decline among 22-to-25 year olds in AI-exposed roles. (HBR, March 2026)
90% — Share of executives who say AI has had zero impact on employment at their own firms, even as AI-related layoff announcements accelerate. (NBER, February 2026)
The question nobody is asking
Every organization I work with right now is asking the same question: is AI ready for the work? They’re evaluating tools, running pilots, benchmarking outputs, measuring adoption rates. That’s the wrong question, and the 346% is what happens when you skip the right one.
The right question is whether the work, workforce, culture ready for AI?
Because the transformation sequence that actually produces results runs in one direction: Mindset first, then Human, then Journey, then Technology. Always in that order, never the reverse. What most organizations have done is start at the end, deploy the technology, and hope the rest of the sequence fills in behind it and it doesn’t. You can’t layer new tools on top of unchanged thinking and call it transformation. That’s frosting on a moldy cake. It looks like progress until someone takes a bite.
The belief gap is the real crisis underneath the jobs data. The belief gap is the quiet, pervasive conviction inside most organizations that AI doesn’t really apply to my job, or that nobody has actually given me permission to change how I work, or that the right move is to wait until things settle down before engaging. That belief doesn’t announce itself. It shows up as slow adoption, as workarounds, as people using AI to do the same work the same way but with an extra verification step added on top. It’s the reason the 346% exists. People aren’t failing to use the tools but they’re using the tools without changing the thinking, and the two things are not the same.
Adoption is the start. Embodiment is “on”. Most organizations are still in rehearsal, and the productivity data is the proof.
What leaders get wrong, and what to do instead
The companies cutting jobs right now are mostly betting on future AI capability, not current AI performance, and the data suggests that bet is running well ahead of reality. When you eliminate people before the tools are ready, you lose the institutional knowledge, the client relationships, and the judgment that takes years to build. And if the productivity gains don’t materialize on the timeline you promised your board, you can’t easily rehire what you let go.
Here’s the uncomfortable truth about that:
You’re cutting the people who carry the knowledge that the AI still needs to learn from.
The deeper problem is that most leaders are trying to manage an AI transformation without first going through one themselves. You cannot delegate a mindset shift. You cannot send your team to a prompt engineering workshop and call it change leadership. When leaders visibly model the new behavior, the organization follows. When leaders announce the new behavior and return to their old habits, the organization reads the truth in what you do, not what you say. If you’re not using AI tools yourself, regularly, deliberately, and talking openly about where they help and where they fall short, you’re not leading a transformation. You’re sponsoring one. Those are not the same thing.
The sequence matters more than the tools. Here’s where to start:
Go first, visibly and specifically. Identify three ways you are personally using AI in your own work right now, not your team, you. Share what’s working and what isn’t in your next leadership meeting. The belief gap closes faster when leaders model the answer than when they mandate it.
Audit before you cut. Before making any further AI-related headcount decisions, find out where your teams are actually spending time with AI tools right now. Ask them directly: is this saving you time, or adding steps? Make your workforce decisions based on that real data, not on what the vendor promised in the demo.
Name what only humans can do. Before eliminating any role, identify what institutional knowledge, relationships, or judgment that person carries that no AI tool currently replicates. If you can’t answer that question clearly, you’re not ready to eliminate the role.
Fix the workflow before you add more tools. If you’ve already deployed AI that’s increasing workload rather than reducing it, pause and redesign the process first. More tools on top of a broken workflow produces a more expensive broken workflow.
Measure embodiment, not adoption. Stop tracking license utilization and prompt volume. Start tracking whether the work itself has changed: are decisions getting made faster, are your people spending more time on the things that actually require them, are client outcomes improving? That’s the measurement that tells you whether transformation is actually happening.
What this means if you’re early in your career
The entry-level roles that used to be the first rung on the ladder are shrinking fastest, and that’s worth taking seriously. But the Harvard Business School research is specific about what’s growing: roles that require human judgment, communication, and the ability to direct AI rather than simply operate it.
The students and early-career professionals who will thrive aren’t the ones who avoided AI out of anxiety, or the ones who outsourced their thinking to it entirely. They’re the ones who moved through fear into genuine agency with these tools, who understand what AI can do well and what it can’t, and who can make a clear case for what they personally bring to work that a prompt cannot replicate. That combination is rarer than most hiring managers will admit, and it’s exactly what’s gaining value right now.
Don’t wait for your institution to teach you this, PLEASE. The fluency you build through direct, deliberate practice with these tools is itself a differentiator, and the thirty days you spend paying close attention to where AI helps and where it falls short will tell you more about your own future than any report or index can.
The story underneath the numbers
The gap between what companies are saying about AI and what’s actually happening inside their organizations is enormous right now, and it’s going to get more visible before it gets smaller. The 346% task-time increase is the clearest signal of that gap. It tells you that millions of people are working harder, not because AI failed to show up, but because their organizations deployed it without changing anything else.
This is what I’ve spent thirty years watching organizations get wrong about technology, and it’s the same pattern every time. The technology gets treated as the answer before anyone has agreed on the question. The sequence gets inverted. The hard work of changing how people think gets skipped in favor of the easier work of changing what tools they use. And then, when the results don’t materialize, the organization either doubles down on the technology or abandons it, and either way the people in the middle pay the price.
The organizations that will look different in five years aren’t the ones that moved fastest on AI. They’re the ones that understood the sequence: that you change the mindset before you change the toolset, that you build the belief before you build the workflow, that you make heads count before you count heads. They’re the ones building what I call changefulness, the organizational muscle to keep adapting without waiting for certainty, to stay in motion when the ground is shifting, to treat disruption not as something that happens to you but as something you’re capable of leading.
That capacity doesn’t come from a software license. It comes from leaders who go first, who model the discomfort of learning something new in public, who ask harder questions about their organizations than the quarterly earnings call requires. It comes from the deliberate, unglamorous work of changing how people think before expecting them to change what they do.
Robert Solow would recognize this moment. The computers are everywhere. The productivity still isn’t moving. And the people making the decisions that will define the next decade of work have a narrow window to ask the question they keep skipping: not whether AI is ready for the work, but whether the organization is ready to actually change.
The technology will keep improving whether you ask that question or not. The window to build the foundation it requires is the part that closes and is requiring immediate attention.
Jason Averbook is a globally recognized thought leader, advisor, and keynote speaker focused on the intersection of AI, human potential, and the future of work. He is the Senior Partner and Global Leader of Digital HR Strategy at Mercer, where he helps the world’s largest organizations reimagine how work gets done — not by implementing technology, but by transforming mindsets, skillsets, and cultures to be truly digital.
Over the last two decades, Jason has advised hundreds of Fortune 1000 companies, co-founded and led Leapgen, authored two books on the evolution of HR and workforce technology, and built a reputation as one of the most forward-thinking voices in the industry. His work challenges leaders to stop seeing digital transformation as an IT project and start embracing it as a human strategy.
Through his Substack, Now to Next, Jason shares honest, provocative, and practical insights on what’s changing in the workplace — from generative AI to skills-based orgs to emotional fluency in leadership. His mission is simple: to help people and organizations move from noise to clarity, from fear to possibility, and from now… to next.
You can email at jasonaverbook@gmail.com or send message at LinkedIn to connect.
We've never had a clean answer for how domain expertise moves from generation to generation without the apprenticeship layer — and entry-level roles, as frustrating as they were for the people in them, were doing that work quietly in the background.
The 346% number tells you something important: organizations treated AI deployment as a task swap, not a workflow redesign. So now you have humans absorbing both the original cognitive load and the AI's failure modes. That's not transformation. That's addition disguised as subtraction.
The harder question underneath yours: if we cut the people carrying the knowledge AND we didn't build the systems to capture it first, what exactly is the AI learning from in five years? We may be automating the surface while hollowing out the foundation.
The 346% task time increase stat hit hard because I've seen a version of this myself.
I run AI agents for content, automation, code. On paper my output tripled. In practice I spent evenings reviewing agent work, weekends debugging edge cases, mornings clearing notification queues from overnight runs. The cognitive load didn't shrink, it shapeshifted. Your point about automating on top of work instead of transforming it is the real diagnosis.
I ended up building a wellbeing system into my agent setup (https://thoughts.jock.pl/p/ai-productivity-paradox-wellbeing-agent-age-2026) because without it the agents just kept producing and I kept reviewing.