Artificial intelligence has moved beyond the adoption question. Across workplaces and educational institutions, people are already using AI to write, analyze information, summarize documents, generate ideas, automate processes and make decisions. The more important leadership question increasingly concerns where AI belongs within the work people perform.
Diane Hamilton captured this shift in a recent Forbes article, “How AI Has Changed The Speed At Which Organizations Must Adapt.” As AI continues to remove friction from finding answers and automating work, Hamilton argues, “the advantage may increasingly belong to people and organizations that recognize sooner when the thing they need to learn has changed and are willing to ask a different question.”
For leaders, asking a different question requires looking beyond efficiency alone because efficiency and human development can move in different directions. Technology can make completing a task easier while simultaneously removing experiences through which people learn to think, solve problems, exercise judgment and work with others. The friction paradox suggests the next challenge for leaders is deciding when AI should remove friction to improve efficiency, when friction should remain to develop human capability, and when work should be redesigned so AI strengthens rather than substitutes for human thinking, judgment and relationships.
Remove Friction When It Improves Efficiency
The case for using AI to eliminate unnecessary friction continues to strengthen. New research from the Federal Reserve Bank of St. Louis shows how widely AI has already penetrated American work. Between August 2024 and May 2026, the percentage of adults using generative AI increased from 45% to 62%, while workplace use increased from 33% to 45%.
Yet the more interesting discovery involves how AI is being used. The researchers found at least 20% of workers use AI in more than 80% of occupations, while fewer than 3% of individual job tasks have AI adoption rates above 50%. AI use is therefore widespread but shallow, suggesting workers are selectively applying the technology to portions of their jobs rather than handing entire roles over to AI.
This distinction provides leaders with an important starting point. AI can remove administrative burden, accelerate information retrieval, support drafting, organize data and reduce repetitive cognitive work while preserving the human contribution surrounding those activities. The leadership challenge involves identifying friction that consumes human capacity while contributing little to meaningful human capability.
Microsoft Research offers additional evidence of this opportunity. Researchers examining digital activity across several large international companies found heavy AI users experienced a 21.2% increase in productivity-oriented application activity and a 7.1% increase in communication activity. Yet the researchers also observed a shift toward more individual and documentation-focused work, raising questions about how efficiency gains could eventually affect interpersonal communication and information sharing.
Efficiency therefore remains valuable, while providing only one measure of successful AI adoption. Leaders should ask what people will do with the time and capacity AI returns to them. Removing friction creates the greatest value when recovered capacity moves toward higher-value activities such as creativity, problem-solving, collaboration, learning and judgment.
Preserve Friction When It Develops Human Capability
Some friction deserves a different response. Learning something difficult, considering competing arguments, solving an unfamiliar problem and developing an original idea all require cognitive effort. Removing too much of this friction can improve immediate performance while reducing the human effort responsible for developing lasting capability.
A newly published multinational study involving 14,949 higher-education students provides compelling evidence for this distinction. Researchers identified four AI interaction profiles, including “Augmentors,” who combined high AI use with strong cognitive skills and efficiency, and “Substitutors,” who demonstrated lower performance and disengagement. Most importantly, self-testing was the strongest predictor of cognitive skill development, while frequency of AI use was the strongest predictor of study efficiency.
The implication extends far beyond education. Human capability depends less on AI use itself and more on interaction quality, task selection and the structures surrounding its use. The question therefore shifts from how much AI people use to what cognitive work remains theirs to perform.
A global review released by researchers at the University of Sydney and Barker Institute on September 3 reinforces this distinction. Researchers analyzed 271 studies from more than 45 countries and found evidence GenAI can increase engagement, motivation and confidence while helping students complete tasks and improve work quality. Evidence concerning deeper learning, reasoning, self-regulation and long-term development remains considerably less developed.
Productive friction therefore becomes a leadership issue extending well beyond education. Organizations need people capable of thinking when the answer remains unclear, challenging an AI recommendation, recognizing faulty assumptions and making decisions when information conflicts. Leaders who optimize every activity for immediate efficiency may eventually discover they have optimized away some of the experiences through which those capabilities develop.
In my previous Forbes article, “Why The Friction Paradox Is The Leadership Challenge Of AI,” I argued leaders must distinguish between friction impeding performance and friction contributing to human development. The emerging evidence makes the distinction increasingly practical. The objective is to understand the function of friction before deciding whether technology should remove or preserve it.
Redesign Work So AI Strengthens Human Capability
The third leadership decision may ultimately prove the most important. Many activities resist a simple classification as work for AI or work for humans. Instead, organizations will need to redesign tasks and workflows around a changing division of responsibility between people and intelligent systems.
New York Federal Reserve research provides an early indication of this transition. Its August 2026 regional business surveys found AI use among service firms had increased to 61%, compared with 40% in 2025 and 25% in 2024. Among manufacturers, adoption reached 51%, approximately twice the previous year’s level.
Widespread layoffs were uncommon among the businesses surveyed. Some organizations reduced hiring because of AI, while others added workers to help implement the technology, and retraining remained the primary workforce response. Companies were teaching AI literacy, tool use, automation, prompt engineering and responsible practices including output verification, bias awareness and data security.
These findings suggest the future of work may depend less on drawing a permanent boundary between human and machine responsibilities and more on continuously redesigning how they interact. AI might generate an initial analysis while a person evaluates assumptions. It might produce options while a human considers consequences, or automate routine execution while employees spend more time with customers, colleagues and complex problems.
Redesign becomes especially important as AI moves from answering questions toward taking actions. AI systems increasingly possess the ability to browse information, manipulate files, execute code and interact with external systems. The progression toward more capable systems makes deciding where human judgment enters a workflow increasingly consequential.
This shift also changes what leaders should measure. Productivity metrics can reveal whether AI makes work faster, while additional indicators can show whether employees are learning, questioning, collaborating and retaining the ability to perform important tasks independently. An organization could become more efficient in the short term while gradually accumulating dependencies it recognizes only when people need to operate independently of technological assistance.
The objective should therefore be augmentation by design rather than augmentation by assumption. Access to AI alone offers little assurance the technology will strengthen employee capabilities. Leaders need to intentionally determine which parts of a task AI performs, which parts humans retain and where interaction between the two produces an outcome stronger than either could generate independently.
Conclusion
The first phase of generative AI encouraged leaders to experiment. The next phase requires judgment. As adoption spreads across occupations, classrooms and workflows, the leadership advantage will increasingly come from understanding where AI belongs rather than simply increasing how frequently people use it.
The friction paradox provides one way to make those decisions. Remove friction when it consumes capacity while contributing little to capability. Preserve friction when effort builds thinking, judgment, learning or relationships. Redesign work when AI and human capabilities can be combined to create something stronger than substitution or automation alone. The next AI question is increasingly clear: leaders must determine when AI use makes humans more capable.
Self-Reflection Questions
- Where is AI removing friction that consumes time while contributing little to meaningful human capability?
- Which activities require struggle, practice or independent thinking because the process itself develops important capabilities?
- Where are employees using AI to augment their thinking, and where is AI beginning to substitute for thinking people still need to perform themselves?
- How are employees using the time and capacity AI gives back to them?
- Which workflows would benefit from a deliberate division of responsibility between people and AI?
- How does the organization measure human capability alongside productivity and efficiency?
- Before automating another activity, what human capability could disappear when people stop performing it?