AI is moving fast. Faster than most organizations can comfortably process. In just a short span, we have moved from early generative AI experiments to agentic AI, autonomous agents, orchestration layers, and an expanding ecosystem of tools that promise exponential gains. The technological curve is steep, and it is accelerating.
Yet for all this progress, a familiar pattern persists inside many companies. AI is still approached primarily as a technology initiative rather than a business or people transformation. Leaders ask what models to deploy, what platforms to standardize on, or how to “get AI into production,” while overlooking a more fundamental question: what problem are we actually trying to solve, and who needs to change their behavior for AI to matter?
As organizations double down on technology-first innovation, the gap widens. Investments grow. Complexity increases. Adoption stalls. Instead of advantage, AI quietly becomes another cost line which is impressive on slides, underwhelming in reality. The issue is not the sophistication of the tools. It is the absence of a human and business-first approach to change.
This is why 2026 must mark a turning point. Not another year of pilots, platforms, or proof-of-concepts, but a year of real organizational transformation. A shift from asking how advanced our AI stack is to asking how ready our people are to work differently. From measuring deployment to measuring learning, agency, and impact.
This reminds us of the recent Data Innovation Summit panel discussion on Workforce 3.0, focused on upskilling, reskilling, and organizing for AI-driven innovation, where the panelists Louise Vanerell, Alexandra M. Davis, Kajsa Norin and Jakob Ökvist explored resistance to change, incentives, fear of failure, and the lived realities of enterprise adoption, a consistent message emerged. While the session mostly addressed skills, tools, and transformation, what it truly exposed was a deeper tension. The gap between what technology enables and what human systems are prepared to absorb.
This article distils that conversation into a broader reflection on where many organizations find themselves today, why so many AI initiatives stall despite heavy investment, and what leaders can do to move from symbolic adoption to meaningful change.
The Skills That Matter Most Are Not Technical
When asked to name the single most important skill for AI innovation, none of the panelists began with programming, statistics, or machine learning theory. Instead, the answers clustered around qualities traditionally associated with entrepreneurship rather than enterprise employment.
Curiosity. Creativity. Fearlessness. A willingness to try, fail, and try again
This is telling. For more than a decade, organizations have framed digital transformation as a skills gap problem. Train more data scientists. Upskill developers. Hire AI specialists. Yet the panel’s opening exchange highlighted a different truth. Technical capability is rarely the binding constraint. Behavioral capability is.
In many organizations, the underlying systems reward predictability, risk minimization, and adherence to established processes. AI, by contrast, thrives on exploration. It introduces uncertainty, partial solutions, and outcomes that cannot always be guaranteed in advance. The friction arises not because people lack intelligence, but because the organizational environment discourages the very behaviours AI adoption requires.
When panelists described the skills most needed today, they implicitly described a workforce that behaves more like entrepreneurs inside the enterprise. People who are comfortable experimenting without certainty. People who see failure as information, not incompetence. People who are motivated by solving problems rather than protecting roles.
This reframing matters. If leaders continue to treat AI as a technical rollout rather than a behavioral shift, no amount of training budget will produce sustainable change.
Resistance to Change Is Not Irrational
Resistance to AI is often discussed as an obstacle to be overcome, a mindset problem that employees need to “get over.” The panel offered a more nuanced interpretation. Resistance is deeply human, and often rational given the context people operate within.
When organizations announce new AI initiatives from the top, they frequently start with solutions rather than problems. A new platform is selected. A vendor is announced. A roadmap is published. Employees are then asked to adapt, often without having been asked where friction actually exists in their daily work.
As one panelist put it, this is like trying to build a house by starting with the roof. The structure may look impressive, but it lacks foundations rooted in lived reality.
People resist not because they dislike innovation, but because they recognize when change is being imposed rather than co-created. When tools are introduced without addressing real pain points, adoption becomes performative. Employees comply outwardly while quietly reverting to familiar methods that actually get work done.
Turning resistance into agency requires listening before prescribing. It requires leaders to ask not “How do we deploy AI?” but “Where do people feel stuck, overloaded, or constrained today?” Only when solutions emerge from these answers does AI become an enabler rather than a burden.
Bottom-Up Innovation Already Exists, If Leaders Are Willing to See It
One of the most striking insights from the panel was the recognition that AI innovation is already happening in many organizations, just not where leadership is looking.
In one example, service technicians with no formal programming background had built their own bots to solve practical problems in the field. These were not sanctioned projects. They were not part of a strategic roadmap. They emerged because individuals were curious, empowered by accessible tools, and motivated to make their own work easier.
This pattern is more common than most executives realize. Across industries, employees experiment quietly with AI to automate reports, summarize documentation, generate drafts, or analyze data. These efforts rarely appear in dashboards or steering committees. Yet they often represent the most authentic form of AI value creation.
The challenge for leadership is not to invent innovation, but to recognize it, amplify it, and learn from it. Doing so requires abandoning the assumption that transformation must always be centrally orchestrated. Instead, leaders must create conditions where experimentation is safe, visible, and connected to broader learning.
The Myth of a Generic AI Strategy
A recurring frustration voiced during the discussion was the reliance on generic frameworks when it comes to AI. Organizations adopt early-standardized AI strategies, maturity models, and policies because they feel safe. Everyone else is doing it. Consultants recognize it. Regulators tolerate it. The problem is that these frameworks rarely reflect how work actually happens inside a specific organization.
An AI strategy that works for a digital-native company will not translate directly to a regulated industrial enterprise. Even within the same organization, different departments face fundamentally different constraints, incentives, and workflows. Treating AI adoption as a uniform process inevitably leads to superficial outcomes.
The panelists emphasized that meaningful progress requires starting from the bottom. Understanding local contexts. Acknowledging differences. Allowing solutions to vary while maintaining shared principles.
This is uncomfortable for leaders accustomed to control. It requires accepting that there is no single “right” way to adopt AI, and that coherence emerges through learning rather than enforcement.
Why Identifying Change Agents Matters More Than Buying Tools
Another theme that surfaced repeatedly was the importance of recognizing individuals who naturally drive change. Every organization has them. People who experiment early. People others turn to for advice. People who bridge technical understanding with business intuition.
Most organizations, however, do not know who these people are.
Traditional HR metrics focus on roles, tenure, and performance within predefined processes. They rarely capture behaviors associated with innovation. As a result, tools are often distributed evenly or according to hierarchy rather than readiness.
When AI platforms are rolled out en masse, the assumption is that everyone will adopt at roughly the same pace. Predictably, this does not happen. Early adopters thrive. Others feel overwhelmed or disengaged. Usage metrics disappoint. Leaders conclude that “people are not ready.”
The alternative is to identify and support those who already exhibit the behaviors needed for AI-driven change. Give them tools first. Learn from how they use them. Let their success stories inform broader rollout. This approach respects the natural diffusion of innovation rather than fighting it.
Fearlessness Requires Psychological Safety, Not Motivational Speeches
Throughout the discussion, the word “fearlessness” appeared repeatedly. Yet fearlessness does not emerge from exhortation. It emerges from safety.
Organizations often claim they want employees to experiment and fail, while simultaneously punishing mistakes through performance reviews, budget scrutiny, or reputational damage. The result is predictable. People nod enthusiastically in meetings and quietly avoid risk in practice.
True psychological safety requires aligning incentives with rhetoric. If experimentation is valued, it must be reflected in how success is measured, how careers progress, and how leaders respond when things do not work as planned.
Failing early, in controlled environments, should be celebrated as learning. Failing late, after large investments and forced rollouts, is far more costly. Yet many organizations invert this logic, discouraging small failures while tolerating large, silent ones.
Vanity Metrics Are the Enemy of Learning
Few topics generated as much shared frustration as metrics. The desire to measure AI progress is understandable. Boards want evidence. Executives want reassurance. Yet many commonly used metrics create a false sense of success.
Counting the number of AI tools deployed, licenses purchased, or models in production says little about actual value. These metrics reward activity rather than impact. They encourage rushed deployment rather than thoughtful adoption.
Worse, they can actively discourage experimentation. If teams are evaluated based on usage targets or production milestones, they may prioritize visibility over usefulness. Tools get shipped that nobody uses. Dashboards glow green while real work remains unchanged.
Meaningful metrics are harder. They require asking uncomfortable questions. Are problems actually being solved? Are people working differently? Are decisions improving? These outcomes are less easily quantified, but far more indicative of progress.
The Hidden Cost of Silent Failure
One of the most insightful observations from the panel concerned a particular kind of failure that rarely gets discussed. Silent failure.
This occurs when AI projects technically succeed but fail socially. A system works as designed. It is deployed on time. It meets specifications. And yet, adoption remains low. People avoid it. Workarounds persist. The organization moves on without acknowledging the gap between promise and practice.
Silent failure is dangerous because it reinforces the illusion of progress. Leaders believe they are advancing. Employees become cynical. Future initiatives face greater resistance.
Avoiding silent failure requires confronting adoption honestly. Not as a communications challenge, but as a design challenge. If people are not using a tool, it is rarely because they are lazy or fearful. It is because the tool does not fit their reality.
Centralization Versus Decentralization Is a False Dichotomy
Large organizations often struggle with where AI should live. Centralized to control cost and risk. Decentralized to encourage innovation and relevance.
The panel highlighted examples of organizations navigating this tension by providing shared frameworks rather than centralized solutions. Common infrastructure, standards, and guardrails coexist with local autonomy. Teams are free to experiment within boundaries that ensure coherence and compliance.
This approach requires trust. It assumes that people, when given the right tools and constraints, will make sensible decisions. It also requires leaders to resist the temptation to over-engineer governance before learning has occurred.
Incentives Shape Behavior More Than Strategy Documents
When discussing why people hesitate to step outside their comfort zones, incentives emerged as a critical factor. Most employees operate under constant pressure to deliver efficiently, meet targets, and avoid mistakes. Experimentation threatens all three.
If learning new tools slows output, even temporarily, it can feel risky. If failure is remembered longer than success, caution becomes rational. If promotions reward stability rather than exploration, curiosity fades.
Organizations that want AI-driven innovation must confront these incentive structures directly. Allowing time for learning. Recognizing experimentation in performance discussions. Protecting those who take thoughtful risks.
Without these changes, strategy documents remain aspirational artifacts rather than drivers of behavior.
Learning Requires Permission to Be Inefficient
One of the most understated yet powerful insights from the panel concerned learning speed. Organizations often expect people to adopt new tools without any drop in productivity. This expectation is unrealistic.
Learning is inherently inefficient. It involves trial and error, confusion, and slower execution. When organizations fail to account for this, they inadvertently discourage learning altogether.
Many corporate learning systems fail not because content is poor, but because there is no space to apply it imperfectly. Employees return from training to the same pressures, the same deadlines, the same metrics. The safest choice is to revert to familiar methods.
Creating space for inefficiency is not indulgent. It is an investment. Organizations that recognize this are more likely to see sustained capability growth rather than episodic enthusiasm.
Forced Adoption Undermines Autonomy
The panel also touched on the growing trend of forcing AI usage, particularly in technical roles. Mandating specific tools, discouraging alternative approaches, or tying compliance to performance reviews may drive short-term adoption metrics, but at significant cost.
Autonomy matters. People care not just about outcomes, but about how they achieve them. When organizations dictate tools rather than problems to solve, they risk disengagement and reduced ownership.
AI should expand agency, not constrain it. The goal is better outcomes, not uniform workflows. Leaders who focus on quality and impact rather than tool usage are more likely to unlock genuine productivity gains.
AI Is a Means, Not the Value
Perhaps the most important philosophical thread running through the discussion was the distinction between using AI and creating value. The two are not synonymous.
There is intrinsic value in exploring AI. Learning how it works. Understanding its limitations. These activities build capability over time. But they are not ends in themselves.
Organizations that conflate AI usage with success risk losing sight of the problems they exist to solve. The most effective leaders hold both perspectives simultaneously. Encouraging exploration while remaining grounded in business outcomes.
This balance is difficult, but necessary. Without it, AI becomes either a toy or a burden.
Comfort Zones Are Psychological, Not Structural
When panelists reflected on their own behavior, a shared pattern emerged. Discomfort precedes growth. Initial resistance gives way to curiosity. Boredom follows mastery. The cycle repeats.
This mirrors what organizations experience at scale. Transformation is not linear. It involves oscillation between excitement and anxiety. Leaders who expect constant momentum will be disappointed. Those who normalize discomfort as part of progress are more likely to sustain change.
Ultimately, moving beyond comfort zones is less about structural reorganization and more about mindset. About accepting uncertainty as a companion rather than an enemy.
Toward a More Human-Centered AI Transformation
The panel concluded not with answers, but with an invitation. To step into discomfort. To question assumptions. To listen more than prescribe. To recognize that AI transformation is as much about unlearning as learning.
Workforce 3.0 is not defined by job titles or tools. It is defined by behaviors. By curiosity over compliance. By experimentation over execution alone. By courage over certainty.
Organizations that embrace this reality will not eliminate resistance, fear, or failure. They will integrate them into a healthier system of learning. One where technology amplifies human potential rather than exposing organizational fragility.
In the end, AI will not replace the need for leadership. It will make it more visible.
10 AI Adoption Steps for AI Success in 2026
So finally. Here are 10 adoption steps you can consider to turn your AI initiatives into success.
- Start with business and people problems, not AI capabilities, platforms, or vendor roadmaps.
- Map where AI experimentation is already happening informally inside the organization and actively support those early adopters.
- Create explicit space for experimentation where short-term inefficiency and failure are accepted as part of learning.
- Align incentives, performance metrics, and career progression with curiosity, learning, and responsible risk-taking.
- Measure adoption and impact, not deployment – distinguish clearly between AI that exists and AI that is actually used.
- Reject generic AI strategies in favor of approaches grounded in your organization’s specific culture, constraints, and workflows.
- Equip leaders to model curiosity and uncertainty rather than certainty and control.
- Make failures visible, discussable, and bounded – treat them as data, not defects.
- Accept uneven adoption speeds across teams and functions instead of forcing uniform rollout.
- Track success through improved decisions, behaviors, and outcomes – not vanity metrics tied to licenses, models, or tools.
