Key takeaways
- A study of more than 21,000 US companies found that the heaviest AI spenders grew their headcount by about 10%, and their entry-level headcount by 12%, in the two years after adopting AI.
- Most of that data comes from the years when AI assisted employees with tasks, before agentic AI could take over whole job functions. And the heaviest adopters were mostly fast-growing, venture-backed tech companies.
- Where a company redesigns entire functions around AI agents, a team’s output stops depending on its size: companies keep their people and grow without hiring at the same pace.
- What AI leaves to humans is the most demanding work, which is why people enjoy their jobs more, at the expense of a higher cognitive load. Working with AI can also push us into constant multitasking: attention is becoming the real bottleneck of this transition.
- Our role moves from doing the work to defining it, judging it and staying accountable for it.
A few days ago I saw a chart on social media, taken from a Financial Times article: companies that adopt AI hire more people, and especially more young people at entry level. I have to admit it surprised me. It goes against most of what I hear from the leaders we work with, and against what I see in my own work with AI. So I did what I always do when a finding feels a bit too comforting: I went back to the source and challenged it.
The optimistic story
The research comes from Ramp, a US corporate card company, and Revelio Labs, a workforce data firm. They linked the real AI spending of more than 21,000 US companies to their employment records, from January 2021 to February 2026. The companies that invested most heavily in AI grew their headcount by about 10% in the two years after they started, and their entry-level headcount by 12%. Companies that only played a little with AI showed no significant change. On Ramp’s chart, the gap keeps widening month after month, but the further right you go, the fewer companies the estimate rests on.
![]()
Source: Ramp Economics Lab and Revelio Labs (2026), A New Look at AI’s Impact on Jobs
This optimistic story deserves to be taken seriously, because it is the one the big consulting firms have told for years. Back in 2017, McKinsey concluded that there would be enough work to maintain full employment until 2030 under most scenarios. The logic is a familiar one: AI makes people more productive, productivity lowers costs, lower costs create demand and growth, and growth creates jobs. It is roughly what happened with electricity, computers and the internet.
What struck me is how much the tone has changed. My impression was that BCG and McKinsey had stopped saying that AI will create more jobs than it destroys. When I went back to their latest reports, I found a more subtle picture. McKinsey still projects slightly more jobs created than lost in the US by 2035, but it now describes this as possibly the largest workforce transformation in US history, with around 11 million people who may need to change occupation entirely. BCG now speaks of AI reshaping half of US jobs within two to three years, and possibly eliminating one in ten over a longer horizon, while warning that this is not an unemployment forecast. And in McKinsey’s latest State of AI survey, more managers expect AI to reduce headcount in their organisation over the coming year than a year ago. The promise of net job creation is still there, but it has become quieter, more conditional and further away.
Why I think this data tells yesterday’s story
Three things in the Ramp study made me cautious.
The first is timing. To see what happens two years after a company adopts AI, you need companies that adopted at least two years before the data ends, so before early 2024. At that time, AI was mostly helping employees write, analyse, code and search faster. It assisted people with tasks; it did not yet run job functions. Agentic AI, the kind that plans and executes a whole workflow across systems, has only started to scale in large companies over the past year. The study captures the era of augmentation, and we are now entering the era of automation.
The second is who adopted. The researchers are transparent about it. The early adopters were larger, more engineering-heavy and more than three times as likely to be venture-backed, and they were already growing much faster than the rest before they touched AI. Ramp even notes that who funded a company predicts its AI adoption better than its sector, and that the gains by sector are concentrated in the information industry. The authors compare early adopters with later adopters to correct for this, which is good practice, but the population remains dominated by tech companies and start-ups. When a start-up raises money, it hires, with or without AI.
The third is depth. Ramp’s lead economist said himself that for many firms, adoption so far has been pretty minimal. Paying for a few AI subscriptions is very different from redesigning how work gets done. The authors write that their results do not mean AI mechanically creates jobs.
And there is already a signal pointing the other way. In August, the Stanford Digital Economy Lab found that employment of 22 to 25 year olds in the most AI-exposed occupations is now 19% below where it would have been, mainly because companies hire fewer of them, and mostly in jobs where AI replaces tasks rather than supports them.
The same team, far more volume
Most of our corporate clients are currently experimenting with AI and running pilots, but few are already at the stage of redesigning entire workflows with agentic AI.
This is where a conversation with Michael Chauliac, a close friend and our AI mentor, changed my perspective. His company, harperOS, deploys an AI operating system across a company: it builds the agents and trains the whole team to run on them. That goes way beyond giving employees access to AI tools.
For each job function, his team interviews the top performers and builds a personal agent from the way they perform their job. That agent is connected to the company’s systems, its knowledge and its processes. Each employee then works with the agent of their function, and the agents of different functions talk to each other. So there are two conversations running in parallel, one between the humans and one between the agents. The work turns from executing to defining outcomes and validating output, a very different way of operating. A function’s output stops depending on how many people it has: the same team carries far more volume, and each person’s job moves from doing the work to directing it.
He usually rolls this out in about four months for mid-sized companies of 20 to 500 people. He tells me people are usually excited about their new way of working. What changes is the hiring plan: the company grows without adding people at the same pace.
In that operating model, companies keep their people and put the new capacity into growth, instead of hiring more people to do the same work. Most companies are not there yet, and such a redesign of operating processes with agentic AI is much easier for small and medium-sized companies than for big corporations, because they are more agile. I think this will give them a real competitive advantage in the years to come.
BCG’s June survey of almost 12,000 workers puts numbers on this divide, and I find it one of the most interesting findings of the year. BCG compared what makes the bigger difference: better access to AI tools, or a clear AI strategy. Among people with neither, 55% report a measurable impact on the business. Give them better tools without a clear direction, and it only rises to 60%. Give them a clear strategy, even with limited access to tools, and it jumps to 80%. In other words, a clear strategy adds about 25 points, better tools about 5. The same survey shows that two thirds of employees receive little or no guidance on what to do with the time AI saves them, and more than half do not reinvest that time in more strategic work.
![]()
AI is not a productivity tool you simply put at the disposal of employees. It has to be part of a clear strategy focused on the outcomes you want. Attention follows intention, for a person and for a company alike. Tools do not create direction. They multiply whatever direction is already there, and when there is none, they multiply the noise.
The joy paradox
The same BCG survey tells a second surprising story, about the people who stay. Two thirds of regular AI users say AI has made them more satisfied with their work, and 41% say it has increased their cognitive load. BCG calls it the joy paradox: AI makes work better and harder at the same time.
![]()
Michael Chauliac put it in one line in a recent post: “When AI does the work, the job that is left is harder.”
From a neuroscience perspective, this makes a lot of sense to me. Most respondents say AI has taken over the simpler tasks and left them with the more complex ones. When the execution goes to the machine, what is left for us is the work that needs judgement, decisions, creativity and relationships. It is the most interesting work, and also the most demanding for the prefrontal cortex, the brain’s CEO. And when a CEO has to review much more high-quality work, produced by extremely intelligent team members, she tends to get tired. Our capacity for that kind of deep, focused work is limited, and AI is increasing the demand for it.
AI can push us into multitasking
There is a third effect I only noticed recently, in my own work. When I build with Claude, I sometimes have to wait ten minutes before it produces the output I want. During those ten minutes, I don’t wait. I usually open a new stream of work, sometimes two, while Claude works on the first one, although I have built dozens of slides explaining why I shouldn’t. Then I come back when the first output arrives. And what I notice is that my attention on that first task is reduced, because I carry some attention residue from the second one.
Sophie Leroy, a researcher at the University of Minnesota at the time, named this effect in 2009. When we switch away from an unfinished task, part of our attention stays stuck on it, and our performance on the next task suffers. Working with AI creates exactly these conditions: several unfinished tasks, open in parallel, each one waiting for a machine.
![]()
Sources: NeuroMindfulness® Institute; Leroy, S. (2009), Organizational Behavior and Human Decision Processes
I think this is one of the biggest risks of this transition. We are designing organisations where one person directs several agents running in parallel, with a brain that was never built to hold that many threads at once. We know that multitasking quickly exhausts the brain. Making it our default way of operating tires the executive function, and when the executive function is tired, it becomes easy to lose our critical thinking when we review what AI produces, and to give away our agency to AI.
From step by step to loops
Working with AI also asks for a shift in mindset, and I have lived a version of it before. When I trained as a scientist, I learned to work in a very linear way, building things step by step, paying attention to details in a linear fashion. When I joined BCG, I was quite surprised to discover, in the training for new joiners, what they called the analytical loop. You start by building the storyline, making assumptions about what the answer might be. Then you work out which analyses you need to validate those assumptions. From the outcome of those analyses, you refine the storyline, and you go round the loop again.
Working with AI is much closer to that loop than to the linear way most of us were trained to work. As Michael puts it, people become the drivers: they define, approve and improve. Our role moves from doing the work to defining it: what expertise the task needs, what the best version looks like, and then judging what comes back with a critical eye and improving it, again and again. And the responsibility stays with us. Nobody can say the AI made them do it. It asks for more critical thinking and more big-picture vision, and less of the careful step-by-step execution on which many of us built our careers. Most people are still used to working in a linear fashion, and I suspect this is part of the reason why simply giving them access to AI tools changes so little.
So, do companies that adopt AI really hire more?
Some do, for now, mostly the fast-growing ones that use AI to grow faster. But I would not draw conclusions from data from the time when AI was an assistant. The question is moving from “does AI make employees more productive?” to “how many people will an organisation need to deliver the same value?”, and, just as important, “what kind of cognitive skills will those people need for the work that is left?” For every leader, it comes down to one very concrete choice: what will you do with the capacity AI gives you back, cut or grow? That is exactly where BCG’s finding matters, because without a clear direction, the capacity simply dissolves into noise.
I don’t have a definitive answer on whether AI will kill jobs or create new ones in the medium term. But what is clear to me is that the new ways of operating in companies that adopt agentic AI at scale force a complete rewiring of the way we used to work. The skills that made us progress in our careers so far are not the ones that will make us thrive in this new world.
This is exactly the question Veronica Brejan and I set out to answer when we created REWIRE at the NeuroMindfulness® Institute. Out of two years of research and experiments with leaders around the world, one conviction took shape: the real advantage, for a company and for any human being, will be the capacity to master attention, to direct it on purpose, to make wise choices and, through those choices, to shape the results we want to see in our work and in our lives. While AI agents take their place beside us, our bet is on the human ability to stay at the driving wheel. We call it the REWIRE loop, and it has five steps:
![]()
Source: NeuroMindfulness® Institute
- Attention: noticing where our attention goes.
- Intention: directing it on purpose, towards what truly matters.
- Choice: making a conscious choice, instead of reacting on autopilot.
- Action: turning that choice into action.
- Results: shaping the results we want to see, then paying attention to them, which takes us back to the start of the loop.
Like working with AI, it is a loop rather than a straight line: every turn sharpens the next one.
To live that loop day after day, we believe five superpowers are critical to thrive in the world of AI:
- Attention Mastery, the foundation of all the others: the ability to notice where our attention goes, direct it deliberately to what matters, and protect it, so we keep our critical thinking when we review what AI produces.
- Inner drive, because AI makes it easy to do more of everything, and drive needs a direction that only we can give it.
- Learning agility, because the tools and the way we work now change every few months, and we learn what we pay attention to.
- Resilience, to sustain a heavier cognitive load without burning out, and it starts with noticing what is happening in us.
- Trust and human connection, because when agents take over the execution, relationships are what stays deeply human, and trust grows from the attention we give people.
Where do you stand?
![]()
If you are curious where you stand on these five superpowers, we have built a short self-assessment. It takes about five minutes. It is free, confidential and GDPR compliant, and you receive a personal report by email with your score on each superpower, a profile for each one and a few reflection questions to carry into your week.
Take the five superpowers assessment
We would love to hear from you if you found the insights in your report valuable. Which superpower surprised you most? Share it in the comments, or simply reply to the email.
Related reading
- Why So Many Capable Leaders Are Losing Their Ability to Think Deeply
- Why Attention Is the New Executive Superpower
- Is emotional intelligence the last frontier of AI?
References
- Kharazian, A., Simon, L. and Stevens, R. (2026). A New Look at AI’s Impact on Jobs: Firm-Level AI Spending and Workforce Adjustment. Ramp Economics Lab and Revelio Labs, 30 June 2026.
- Murray, C. (2026). Heavy corporate AI spenders add staff faster than peers. Financial Times.
- McKinsey Global Institute (2017). Jobs lost, jobs gained: Workforce transitions in a time of automation. December 2017.
- McKinsey Global Institute (2026). Workforce in motion: Skills and pathways to future jobs in the United States. September 2026.
- McKinsey & Company (2026). The state of AI in 2026: On the road to ROI. 25 August 2026.
- BCG Henderson Institute (2026). AI Will Reshape More Jobs Than It Replaces. April 2026.
- Boston Consulting Group (2026). AI at Work 2026: Why Strategy Matters More Than Tools. June 2026.
- Brynjolfsson, E., Chandar, B. and Chen, R. (2026). Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence. Stanford Digital Economy Lab, revised August 2026.
- Leroy, S. (2009). Why is it so hard to do my work? The challenge of attention residue when switching between work tasks. Organizational Behavior and Human Decision Processes, 109(2), 168–181.
This article was first published on LinkedIn.
Leave a Comment