Deloitte’s October 2026 study of 1,000 change leaders reveals how much pressure organizations are putting on their people. 84% said employees were expected to take on transformation work on top of their existing workloads. 61% felt AI transformation was moving faster than employees could handle. More than half said nobody in their company had the authority to slow or pause an initiative when employees ran out of capacity.
Microsoft’s 2026 Work Trend Index surveyed 20,000 AI-using knowledge workers across 10 countries. 45% said focusing on current goals felt safer than redesigning their work with AI. Only 13% said they’d be rewarded for reinventing work with AI if the results fell short.
Microsoft calls this tension the Transformation Paradox. Employees are ready to reinvent how they work, but the metrics, incentives, and norms around them keep reinforcing the old way.
Microsoft is right about incentives, and I think the paradox reaches even further. When I put this fall’s research next to the conversations I have with leaders, I see five pressures that feed one another. Each one makes the others harder to solve, which is why fixing them one initiative at a time rarely holds.
Five Pressures That Feed One Another
The Stability Illusion
The first pressure shapes whether we notice change at all. Most of our lives stay familiar while the world around us shifts. We wake up in the same homes, drive the same routes, and sit in the same meetings. Someone uses AI to draft an email or summarize a call, and it feels like a handy shortcut. The everyday experience is incremental, even when the underlying capabilities are moving fast.
That familiarity can make significant change hard to see. Deloitte’s March 2026 research found that 44% of leaders and managers named change that doesn’t feel relevant or personal as a leading barrier, and 30% of workers said they didn’t understand how a change related to them. People respond to what they can see and connect to their own work. A leader’s job is to help them see emerging change early, without leaning on fear or manufactured urgency.
The Exponential Perception Gap
Even once people see change coming, they tend to judge its pace by looking backward, and most planning works by that same extrapolation. We look at last year, estimate next year, and set targets accordingly. That approach works well when change is gradual, but it gets shakier when capabilities accelerate.
Stanford’s 2026 AI Index offers a glimpse of how differently people picture the near future. Experts expected generative AI to assist with 18% of U.S. work hours by 2030, while the public estimated 10%. Whoever turns out to be right, the gap shows that people inside the same organization may be planning for very different futures. The advantage will likely go to the organizations that notice quickly when their predictions miss and adjust with confidence.
Responding quickly also means rethinking how organizations introduce new tools in the first place. Many organizations still roll out new technology along the traditional adoption curve. A few enthusiasts go first, the majority follows over the next year or two, and the stragglers catch up eventually. That pacing worked when a new system arrived every few years and stayed put once it landed. Today, capabilities compound faster than any single rollout can spread, so the people at the back of the curve can find themselves a little further behind with each new wave. Over time, that widening gap quietly slows the momentum of the people around them. Leaders need to treat adoption as a continuous rhythm the whole organization moves through together, with learning built into the work every day.
The Performance Paradox
Leaders who do see the acceleration coming still have to ask people to change inside a system built for today’s results, and this is where Microsoft’s research lands hardest. Picture a regional operations manager on a Monday morning. Her dashboard shows quarterly revenue, customer satisfaction scores, and a cost target she’s already behind on. Her inbox holds an invitation to lead an AI pilot, a request to map her team’s workflows for redesign, and a reminder that her own AI training is due Friday. By 9:30, two people on her team have stopped by to ask whether the new tools mean their jobs are changing. She has the same eight hours she had last year, and every one of them is already spoken for.
Each expectation is reasonable on its own. Together, they compete. Learning takes time, experiments carry risk, and redesigned processes often slow productivity and put targets at risk while people learn the new way of working, long before performance improves. When people protect today’s numbers over tomorrow’s capabilities, they’re responding rationally to the system they work in. I’d treat that behavior as useful information about the system.
The Role Evolution Gap
While the scorecards stay anchored in the old way of working, the jobs underneath them are already changing. AI is creating new kinds of responsibility inside existing roles. BCG’s September 2026 research with approximately 50 AI front-runner companies identified five emerging role archetypes. Citizen builders create their own apps, automations, and agents. Workflow designers decide which parts of a process belong to AI and which belong to people. AI guardians set the boundaries for safe and responsible use. Domain anchors turn specialized expertise into something the whole organization can use. Agent shepherds oversee groups of AI agents, including their permissions, performance, and shutdown.
BCG also found that about half of the firms were pushing decisions further down, and roughly three-quarters emphasized judgment over activity when evaluating work. Responsibility is moving, and tool training only covers part of that shift. Someone has to decide who can redesign work, who owns AI-generated outcomes, and how people build the judgment their new responsibilities require.
The Capacity Paradox
All of those new responsibilities land on people whose days are already full. AI is supposed to give people time back, but in practice it often fills that time with more work. Researchers Aruna Ranganathan and Xingqi Maggie Ye of Berkeley Haas saw this during an eight-month study of a U.S. tech company with about 200 employees. As they reported in Harvard Business Review, people used AI to work faster, then used that speed to take on more tasks and stretch their workday longer. Often, nobody asked them to. Once AI made more feel possible, they simply did more.
This pattern has a long history. In 1865, economist William Stanley Jevons noticed that as steam engines used coal more efficiently, Britain ended up burning more coal, because cheaper power led people to find more uses for it. Ninety years later, C. Northcote Parkinson captured a similar idea in his famous rule that work expands to fill the time available. AI follows both patterns. When a report that used to take three days takes three hours, leaders start asking for it every week, and the time it saved disappears before anyone can use it to learn or rethink the work.
Anthropic saw the same thing inside its own company. As employees handed more work to AI agents, more of their own time went to managing those agents. The share of work sessions spent setting up and monitoring agents rose from about 10% in fall 2025 to 22% in spring 2026. The agents could produce more, but the people overseeing them became the bottleneck.
Three Capacities Every Leader Should Watch
These pressures point to a distinction most transformation plans skip. Organizations have three different kinds of capacity, and AI affects each one differently.
Production capacity is how much work the organization can generate. AI expands it quickly, and it’s the easiest one to measure. We can count reports, tasks completed, throughput, and hours saved.
Absorption capacity is the organization’s ability to recognize, evaluate, and make use of new information and capabilities. It depends on human attention, judgment, and time to learn. An organization can have more information than ever and still struggle to turn it into good decisions.
Adaptation capacity is the ability to change direction, redesign work, build new skills, and move resources when conditions shift. It lives in leadership flexibility, employee participation, and room to experiment.
Production capacity determines how much work an organization can do. Absorption capacity determines how much of it the organization can meaningfully process. Adaptation capacity determines how well it can change.
AI can raise the first without touching the other two. When production climbs and the rest stay flat, overload follows, and the very attention people need to adapt gets used up reviewing more output. That’s the Anthropic finding in miniature, and it’s why adoption rates alone tell leaders so little about readiness.
Where Do the Saved Hours Go?
If AI saves an employee five hours a week, someone decides where those hours land. They can go to more deliverables, better customer experiences, lower costs, learning, process improvement, innovation, or a little breathing room. Each is a legitimate choice. The trouble starts when nobody makes the choice on purpose and the hours quietly fill with more of the same work.
I use the PATH Framework to help leaders make those choices deliberately. It works for a defined initiative, and it works as an ongoing loop.
Present. Start with an honest look at today. What’s changing outside the organization, and what should stay steady inside it? Which assumptions are aging? Where are people already stretched thin? Just as important, name what’s working and what will hold steady. Customers will still choose the people they trust, and a seasoned manager’s sense of when to push and when to listen will still matter. The relationships that carry a client through a hard quarter, a team’s shared sense of purpose, and the expertise people have spent years building all come with you into whatever comes next. Naming those anchors out loud gives people solid ground to stand on while the rest of the work evolves.
Assess. Look at what’s driving the change, along with readiness and capacity. Is the pressure coming from a competitor’s move, a shift in customer expectations, a new regulation, or a new capability? Which timelines are truly critical, and where is there room to pace the work? How are current KPIs shaping behavior? Which of BCG’s emerging responsibilities are already showing up informally? Where are production, absorption, and adaptation out of balance?
Transition. Redesign the work itself, and plan honestly for what that costs. Build the downtime of learning into the schedule and adjust targets to reflect a temporary drop in productivity. Where the load is too heavy to carry internally, bring in outside help so your people have room to adapt. Protect time for learning and experimentation, define who owns AI-generated outcomes, and give someone real authority to pause an initiative when capacity runs out. Deloitte’s finding that most companies lack that authority is one of the easiest gaps to close.
Harness. Measure what the new capabilities actually deliver. Track better decisions, stronger customer outcomes, and growing employee capability alongside activity counts. Then return to Present, because the environment will have moved again.
The point of the loop is to make change part of how the organization allocates resources, measures success, and makes decisions. Asking people to carry transformation on top of a full job indefinitely is how good initiatives wear people out.
Better Questions for the Next Leadership Meeting
Most AI dashboards track how many employees use the tools, how many processes are automated, and how many hours are saved. Those are useful operational measures. But a company can score well on every one of them and still struggle to make decisions, or post record productivity while its people quietly burn out.
The questions I’d bring to the next leadership meeting go deeper.
- Are our people better equipped to recognize change when it’s still early?
- Can we respond quickly when our assumptions prove wrong?
- Do employees have the time and authority to improve how their work gets done?
- Are new responsibilities backed by the right skills, governance, and accountability?
- Are our efficiency gains creating room for learning and better decisions?
People closest to the work often see the inefficiencies, dependencies, and unintended consequences that never reach an executive dashboard. When they push back, it’s worth asking what that pushback reveals about trust, workload, incentives, or risk. Their insight is essential to redesigning work well.
Technology will keep moving, sometimes gradually and sometimes all at once. The organizations best positioned for that future will be the ones that build the ability to recognize meaningful change, make deliberate choices, develop their people, and adapt while protecting what creates value.
Oh, and One More Thing…
Want to see how your own organization stacks up? My free Calculators, Games, and Assessments library has 26 tools and five leadership games, and you’ll never fill out a form before you see your results. Lead a team through eight quarters of competing initiatives in the Change Load Juggler and find out how long your capacity and trust hold up. Play detective in Adoption or Transformation?, where every dashboard is green but something doesn’t add up. If you’d rather get straight to the numbers, the Operational Efficiency ROI calculator shows where your saved hours are likely to land. Each one takes just a few minutes, and fair warning, the games are hard to play only once.
Sources
- Anthropic, “How employment is changing at Anthropic,” October 16, 2026 (Wilmers, Catipon, Massenkoff, McCrory)
- Boston Consulting Group, “Five Ways That AI Front-Runners Change How Work Gets Done,” September 24, 2026 (Goel, Bedard, Kropp, Sim, Kaffe)
- Deloitte Insights, “The 4:1 change management return most organizations are leaving on the table,” October 19, 2026 (Dake, Domergue, Ahn, Cantrell)
- Deloitte Insights, “Staying relevant in a world that won’t sit still,” March 3, 2026
- Harvard Business Review, “AI Doesn’t Reduce Work, It Intensifies It,” February 9, 2026 (Ranganathan and Ye)
- Microsoft WorkLab, “Agents, human agency, and the opportunity for every organization,” 2026 Work Trend Index, May 5, 2026
- Stanford HAI, “The 2026 AI Index Report”
- William Stanley Jevons, The Coal Question, 1865
- C. Northcote Parkinson, “Parkinson’s Law,” 1955
