The Board
In 2025, 20% of EU enterprises with 10 or more employees used artificial intelligence in their operations, up from 7.7% in 2021. Denmark leads at 42%, Romania trails at 5.2%. And in February 2026, a study of nearly 6,000 executives across the US, UK, Germany, and Australia found that roughly 90% of them report zero measurable impact from AI on either employment or productivity over the past three years.
That gap between adoption and impact is the story of work in 2026. Everyone is buying the tools. Almost nobody has changed how work actually gets done.
The Solow Paradox is back, and this time it brought receipts
In 1987, economist Robert Solow wrote that "you can see the computer age everywhere but in the productivity statistics." Thirty-nine years later, the same sentence works if you swap "computer age" for "AI." The NBER working paper behind the CEO survey is blunt: 69% of firms actively use AI, but executives report using it an average of just 1.5 hours per week. Two-thirds of corporate leaders have access to the most powerful productivity tools ever built and use them less than they use their email signatures.
US nonfarm business productivity grew 2.2% in 2025, a healthy number by historical standards. But the San Francisco Fed estimates that AI's contribution to total factor productivity growth in 2025 was 0.01 percentage points. Not a rounding error. Barely a decimal.
The prediction crowd remains optimistic. Goldman Sachs projects a 15% boost to labor productivity when AI is "fully adopted." But "fully adopted" is doing a lot of work in that sentence. And the executives themselves, when asked to forecast, predict AI will boost productivity at their firms by just 1.4% over the next three years.
The productivity revolution is real. It is also, for now, happening to about 5% of companies. BCG's research classifies just 5% of firms as "future-built" for AI. Those firms expect double the revenue increase and 40% greater cost reductions than the rest. The other 95% are running pilots, buying licenses, and waiting for something to click.
Most companies don't have an AI strategy. They have an AI budget.
Stanford's 2025 AI Index found that 78% of organizations reported using AI in at least one business function in 2024, up from 55% a year earlier. That adoption curve would normally take enterprise technology half a decade. But adoption and integration are different things. One is a purchase order. The other is an operating model.
The companies getting value from AI share a pattern that BCG, McKinsey, and the NBER data all point to: they changed how work is organized. McKinsey's research found that more than 70% of the skills employers seek today are used in both automatable and non-automatable work. The skills endure. The workflows around them do not.
Shopify's CEO made this concrete in April 2025 with an internal memo that went viral: before asking for more headcount, teams must demonstrate why AI cannot do the work. That memo was read as a threat. It was more accurately a process change. Every hiring request now starts with a question about whether the role needs to exist in its current form, or whether the work can be split differently between human and machine.
Whether you agree with the policy or not, the underlying logic is spreading. The question companies are learning to ask is "which parts of this job are better done by a person, and which parts can a machine handle?"
The middle of the org chart is thinning
Gartner predicts that by the end of 2026, 20% of organizations will use AI to flatten their structures, eliminating more than half of their middle management positions. A Korn Ferry survey found that 41% of employees say their companies have already reduced managerial layers.
The reason is structural. A 2025 Harvard Business School study found that 6 in 10 managers spend more than half their time on administrative tasks: scheduling, reporting, performance tracking, routing approvals, bridging communication between teams. These are exactly the tasks that AI project management and analytics tools now handle with increasing reliability.
What is disappearing is the coordination layer. The manager whose primary value was knowing what each person on the team was doing and reporting it upward. AI dashboards, automated status reports, and project tracking tools do that work now, often better, because they update in real time.
What is not disappearing is the judgment layer. The manager who decides which project to prioritize when resources are constrained. The one who notices that a team member is burning out before they say so. The one who makes the call on a difficult tradeoff between speed and quality. AI cannot do those things, and claiming it can is a good way to learn a painful lesson in six months.
The companies that cut management indiscriminately are already finding this out. The ones that are doing it well are splitting the old "manager" role into two parts: the coordination work gets automated, and the judgment work gets assigned to fewer, more senior people with wider spans of control. LinkedIn data from early 2026 shows job postings with "manager" in the title declining while postings for "lead" and "principal" roles, which carry strategic responsibility without traditional people-management duties, are growing.
The Klarna lesson: cutting first, thinking second
The most instructive case study of 2025 was a reversal. Klarna's CEO announced that the company's AI customer service agent was doing the work of 853 employees and that headcount had dropped by 40%, from 5,000 to roughly 3,000. The company was the poster child for AI-driven efficiency.
Then customer complaints increased. Satisfaction scores dropped. Customers reported that automated responses were generic, repetitive, and unable to handle anything that required judgment. Klarna's CEO publicly acknowledged that the company "went too far" and began hiring customer service agents again.
Klarna is part of a pattern. Gartner predicted in February 2026 that by 2027, half of companies that cut customer service staff because of AI will rehire for similar functions, often under different job titles. Forrester's 2026 predictions go further: 55% of employers already regret laying off workers for AI capabilities that, in their words, don't exist yet.
The pattern is consistent. Companies announce AI-driven workforce reductions. The reductions produce short-term cost savings and medium-term quality problems. The quality problems lead to rehiring, sometimes at lower salaries, sometimes offshore, but rehiring all the same. I think the lesson is that most companies are replacing people before they have actually redesigned the work, and that sequence matters enormously.
The productivity evidence is real but smaller than the hype
When you look past the macro data and into specific task-level studies, the picture gets more interesting. The most surprising finding of 2025 came from METR, an AI safety research organization. In a randomized controlled trial with 16 experienced open-source developers completing 246 tasks, developers who used AI coding tools took 19% longer to complete tasks than those who did not. Before the study, the same developers predicted that AI would make them 24% faster. The gap between perceived and actual productivity is worth sitting with.
Other evidence points in different directions. Industry surveys report that developers using AI tools save an average of 3.6 hours per week. Brookings research found that firms adopting AI saw roughly 6% higher employment growth and 9.5% more sales growth over five years. These are not trivial numbers. They are also not the revolution that the marketing materials promise.
The honest summary: AI produces measurable productivity gains in specific, well-scoped tasks. It produces uncertain or even negative effects when deployed without restructuring the work around it. And it produces the largest gains for less experienced workers doing routine tasks, while experienced workers in complex domains see smaller or no benefits.
The roles that are actually growing and actually shrinking
The World Economic Forum's Future of Jobs Report 2025 projects that by 2030, 170 million new roles will be created globally while 92 million will be displaced, a net gain of 78 million jobs. The fastest-growing roles are big data specialists, AI/ML engineers, and cybersecurity analysts. The fastest-declining are clerks, cashiers, and administrative assistants.
In absolute numbers, though, the biggest job growth is in roles that AI cannot touch: farmworkers, delivery drivers, construction workers, nurses. The data center buildout alone has created 216,000 construction jobs since 2022, according to labor market tracking data.
For junior developers and entry-level knowledge workers, the picture is genuinely difficult. Workers aged 22 to 25 in AI-exposed roles have seen a 16% drop in employment. Companies are finding that AI coding assistants can handle work that used to go to less experienced engineers. The traditional entry-level pipeline, where you learn by doing simple work and gradually take on complexity, is being compressed.
Europe is regulating first and figuring out the rest later
The EU's AI Act entered into force in August 2024, with full applicability by August 2026. By this summer, employers using AI in HR processes, which includes resume screening, candidate chatbots, performance management, and video interview analysis, will face high-risk classification. Fines can reach 35 million euros or 7% of global turnover.
This is happening in parallel with the EU Pay Transparency Directive, which requires transposition into national law by June 7, 2026. The combination means that by the end of this year, European companies will need to explain both how they pay people and how their algorithms make decisions about people. Most are not ready for either.
The agentic hype is real but the deployment gaps are bigger
The latest wave of AI enthusiasm centers on "agentic AI," systems that don't just respond to prompts but autonomously plan, execute, and coordinate tasks. A PwC survey from May 2025 found that 79% of organizations already run AI agents in production. IDC expects AI copilots to be embedded in 80% of enterprise workplace applications by the end of 2026.
The gap between adoption and reliability is where things get interesting. Gartner predicts that over 40% of agentic AI projects will be scrapped by 2027. Eighty percent of organizations have reported risky agent behaviors, including unauthorized system access and improper data exposure. CNBC reported in March 2026 that this "silent failure at scale" represents a category of business risk that most companies have no framework for managing.
How to build teams when the tools change every quarter
The hardest question for any leader building a team in 2026 is how to plan for capabilities that change faster than org charts can adapt. I think the evidence points to a few principles.
Firstly, hire for judgment. Build teams around ambiguity, context sensitivity, and novel problem-solving. Secondly, redesign work before reducing headcount. Cut the work first, then adjust the team. Thirdly, invest in AI literacy. The tools are useless if the people using them cannot distinguish between useful output and confident-sounding nonsense. Fourthly, protect the entry-level pipeline. If AI handles the tasks that junior employees used to learn from, the company saves money now and creates a talent crisis in five years.
The uncomfortable middle ground
The future of work with AI is a slow, uneven, often clumsy process of figuring out which work humans should do, which work machines should do, and how the two fit together in organizations that were designed for neither.
The companies that are getting it right are the ones that treat AI adoption as an organizational design problem. The companies that are getting it wrong are optimizing for a version of AI that does not exist yet while dismantling the human systems that currently hold their operations together.
Ninety percent of executives say AI has had no measurable effect on their companies. Five percent of companies are capturing almost all the value. Fifty-five percent of those who laid off workers for AI already regret it. Those three numbers, together, tell you everything about where we actually are.