Artificial intelligence adoption is accelerating, but meaningful enterprise transformation remains difficult. Many organizations now provide employees with generative AI tools, yet relatively few have redesigned their operations deeply enough to produce measurable company-wide value. McKinsey’s 2025 research found that nearly nine out of ten surveyed organizations were regularly using AI, but most had not embedded it sufficiently into workflows to generate material enterprise-level results. Earlier McKinsey research also found that workflow redesign was the organizational practice most strongly associated with financial impact from generative AI. The adoption problem is frequently misunderstood. Leaders assume that employees resist AI because they lack training or technical ability. Those factors matter, but resistance can also come from deeper psychological concerns.
Employees may believe AI threatens:
Competence: “Will the expertise I built over many years still matter?” Autonomy: “Will an algorithm begin controlling how I perform my job?” Relatedness: “Will AI weaken collaboration, trust, mentorship, or my place on the team?” Security: “Is the company preparing to improve my work or eliminate my role?” Fairness: “Will AI be used to evaluate me without transparency?” Identity: “What is my professional value if software can perform part of my work?”
Wharton researchers describe these concerns through an AWARE framework:
Acknowledge the psychological impact. Watch employee coping behaviors. Align support systems with employee needs. Redesign roles for human-AI complementarity. Empower employees through transparency and participation. The framework is valuable because it shifts AI adoption away from a narrow technology question and toward a leadership and organizational-design challenge.
A successful AI adoption strategy should therefore combine six connected systems:
Business strategy: Identify problems worth solving. Workflow redesign: Rebuild processes rather than adding AI to inefficient work. Human adoption: Protect competence, autonomy, trust, and belonging. Skills development: Provide role-specific learning and practice. Governance: Define safe, accountable, and transparent use. Measurement: Track utilization, quality, financial impact, employee experience, and risk. The objective is not to force every employee to use AI constantly. It is to create an environment where people understand when AI is useful, when human judgment is essential, what the rules are, how success will be measured, and how their roles can become more valuable rather than merely more automated.
The AI Adoption Paradox Artificial intelligence has become one of the most heavily discussed business technologies in modern history. Executives are approving AI budgets. Software vendors are adding copilots and agents to nearly every enterprise product. Employees are experimenting with public AI tools. Boards are asking management teams to explain their AI strategies. Investors increasingly expect companies to describe how AI will improve productivity, margins, innovation, or customer experience. Yet widespread interest has not automatically produced widespread operational transformation.
This creates an AI adoption paradox:
Organizations can have access to advanced AI while remaining unable to change how work is actually performed. A company may purchase hundreds or thousands of AI software licenses while only a small percentage of employees use them regularly. Another company may report dozens of AI pilots while none has progressed into a dependable, governed production workflow. A third company may celebrate high usage statistics even though employees are using AI for low-value activities such as rewriting emails, summarizing documents, or generating generic presentation text. These activities may save time, but they do not necessarily transform the economics of the business. Meaningful adoption occurs when AI becomes part of a repeatable workflow that improves an important outcome.
Examples include:
Reducing the time required to resolve customer-support cases. Helping sales teams qualify opportunities more accurately. Detecting financial anomalies before they become losses. Accelerating product research and development. Improving the quality and consistency of legal or compliance reviews. Reducing administrative work for doctors, engineers, teachers, or consultants. Helping managers identify operational bottlenecks. Increasing the speed at which employees can retrieve institutional knowledge. Allowing small teams to manage workloads that previously required much larger teams. Improving decisions without removing appropriate human responsibility. The difference between superficial adoption and operational adoption is workflow integration. A chatbot that employees occasionally visit is a tool.
An AI capability integrated into a governed process, connected to trusted information, monitored for performance, and supported by trained employees can become infrastructure.
Why AI Adoption Is Harder Than Installing Ordinary Software Many organizations approach generative AI as though it were another conventional software implementation. They select a vendor, purchase licenses, conduct introductory training, publish a policy, and expect employees to begin using the new system. This approach often fails because generative AI is not merely another application. It can affect the substance of knowledge work itself. Traditional enterprise software usually helps people perform a defined process. Accounting software organizes transactions. Customer relationship management systems store sales information. Project-management software tracks assignments and deadlines.
Generative AI can participate in activities previously associated with human expertise, including:
Writing. Coding. Analysis. Research. Design. Translation. Planning. Forecasting. Customer communication. Knowledge retrieval. Decision support. Ideation.
Document review. Data interpretation. Workflow coordination. Because these activities are closely connected to professional identity, AI adoption can feel more personal than previous technology changes. An employee may not feel threatened when a new expense-management platform replaces a spreadsheet. The same employee may feel deeply threatened when AI drafts reports, produces financial analysis, communicates with customers, or recommends decisions that were previously considered evidence of human expertise. This is why an AI program can trigger emotional and political responses even when leaders describe it as a productivity initiative. The technology may be presented as assistance, but employees may interpret it as evaluation, surveillance, deskilling, displacement, or a signal that leadership no longer values their expertise.
The Psychological Foundations of AI Resistance The Wharton framework draws attention to three fundamental psychological needs: competence, autonomy, and relatedness. When AI strengthens these needs, employees are more likely to adopt it. When AI frustrates them, resistance becomes more likely.
1. Competence: “Am I Still Good at My Job?”
People generally want to feel capable. Professional confidence often develops through years of education, repetition, feedback, judgment, and accumulated experience. Generative AI can destabilize this confidence surprisingly quickly. A marketer may see an AI system generate ten campaign concepts in seconds. A software developer may watch a coding assistant produce a working function almost instantly. A lawyer may see a system summarize hundreds of pages of documents. A designer may generate dozens of visual directions from a short prompt. A financial analyst may observe AI drafting an initial market assessment that once required hours of research.
The employee may appreciate the speed while also wondering:
Was my expertise less valuable than I believed? Will the company still need as many people in my role? Will younger employees using AI outperform experienced employees? Will management expect twice as much output without increasing compensation? Will I be blamed when AI makes an error? Will my ability to perform the work independently deteriorate? Resistance is not always irrational. It may be a response to unresolved questions about professional value. Leaders should not dismiss these concerns by telling employees that “AI will not replace you, but someone using AI will.” That phrase may be intended to encourage adoption, but it can also communicate a threat: adopt immediately or become economically irrelevant.
A more constructive message is:
AI will change parts of the work. We will involve you in deciding where it helps, where it creates risk, which human capabilities become more important, and what support you need to succeed in the redesigned role. That message does not promise that nothing will change. It provides employees with a credible path through the change.
2. Autonomy: “Will I Still Control How I Work?”
Employees are more likely to accept tools that increase their agency. They may resist tools that feel mandatory, intrusive, inflexible, or controlling.
Autonomy concerns arise when organizations:
Require employees to use AI without explaining why. Measure adoption only through usage volume. Insert AI recommendations into decisions without allowing human challenge. Use AI to monitor employee communications or behavior. Standardize creative or professional work too aggressively. Restrict employees to poorly designed tools while demanding ambitious results. Change performance expectations without consultation. Allow algorithms to influence scheduling, evaluation, promotion, or discipline without transparency. An AI system can technically save time while psychologically reducing ownership. For example, a sales representative may resent a system that automatically determines which customer to contact, what to say, when to follow up, and how the interaction will be scored. The same representative may welcome an AI assistant that prepares account research, highlights relevant developments, drafts optional messages, and leaves the final strategy to the human. The technology may be similar. The design philosophy is different.
One system commands. The other supports.
3. Relatedness: “Will AI Weaken My Relationships at Work?”
Work is not only a collection of tasks. It is also a social environment where people gain recognition, learn from colleagues, develop friendships, receive mentorship, and feel part of a group. Poorly designed AI adoption can weaken this environment. Employees may interact less with experienced colleagues because they ask AI instead. Managers may replace conversations with automated summaries. Junior employees may lose opportunities to learn through observation and participation. Teams may rely on machine-generated communication that is efficient but impersonal. Organizations may reduce entry-level roles that historically served as training grounds for future experts and leaders. These changes can produce a workplace that is faster but less connected. Relatedness matters because adoption often spreads socially. Employees are more likely to experiment when trusted colleagues demonstrate useful applications, share mistakes, explain limitations, and solve problems together. Peer learning may be more persuasive than another mandatory online course.
4. Security: “Is This a Productivity Tool or a Layoff Strategy?”
Employees often evaluate AI announcements through the lens of job security.
When leaders say they are introducing AI to “increase efficiency,” employees may hear:
Reduce headcount. Increase workloads. Eliminate junior positions. Outsource professional judgment to software. Monitor performance more closely. Justify future restructuring. Leadership credibility depends on whether the company communicates honestly. Not every organization can promise that AI will never affect employment. That promise may be unrealistic.
Companies can still explain:
Which tasks are expected to change. Which roles are likely to grow. Which capabilities will become more important. Whether productivity gains will be reinvested, used for growth, or used to reduce costs. What retraining and internal mobility options will be offered. How decisions affecting employees will be made. Whether AI will be used in performance evaluation. What appeal and oversight mechanisms will exist. Uncertainty is often less damaging than secrecy.
5. Fairness: “Who Benefits from AI?”
Employees will judge adoption partly by how benefits and burdens are distributed. They may resist if executives receive strategic benefits while employees receive additional monitoring, pressure, or responsibility.
They may also resist if:
AI access is limited to favored departments. Training is available only to senior staff. Productivity targets increase immediately. Employees are expected to experiment on personal time. AI-generated errors become the employee’s responsibility. Savings flow entirely to shareholders while workloads intensify. Certain groups are disproportionately evaluated or displaced. Adoption improves when employees see a fair exchange. For example, if AI reduces administrative work, the organization should be clear about what employees gain from the saved time. They might gain more flexible work, greater focus on meaningful tasks, better development opportunities, lower burnout, or participation in new revenue-generating activities.
The AWARE Framework for Human-Centered AI Adoption Wharton’s AWARE framework gives leaders a practical way to address resistance. A: Acknowledge the Psychological Impact The first step is to admit that AI adoption affects more than efficiency. Leaders should discuss professional identity, skills, control, security, collaboration, and ethics openly. This does not require dramatizing every concern. It requires treating employee reactions as legitimate data.
Useful questions include:
Which parts of the job do employees believe define their expertise? Which AI capabilities feel most threatening? Where do employees expect AI to improve their work? Which decisions should remain human-controlled? What risks are employees seeing that leadership may have missed? How could the technology affect team relationships? Which groups may experience the change differently? What would make employees trust the implementation process?
Organizations can gather this information through:
Anonymous surveys. Team discussions. Interviews. Focus groups. Office hours. Union or employee-representative consultations. Workflow observation. Pilot retrospectives. Internal communities of practice. The objective is not to give employees veto power over every technological change. The objective is to understand the human system before redesigning the technical one. W: Watch Coping Behaviors
Employees react to AI in different ways. Some responses are visible. Others remain hidden.
Constructive coping behaviors may include:
Experimenting with prompts. Sharing useful workflows. Building reusable templates. Asking for training. Testing outputs carefully. Identifying new use cases. Collaborating across departments. Reporting errors and risks. Helping colleagues learn.
Unproductive or risky coping behaviors may include:
Avoiding AI-related assignments. Quietly refusing to use approved tools. Using unauthorized public tools. Hiding AI use from managers. Copying confidential information into unapproved systems. Accepting AI output without verification. Sabotaging implementation efforts. Inflating usage statistics without meaningful application. Withdrawing from team discussions. Treating AI as a shortcut rather than a professional tool. Hidden usage deserves particular attention. Wharton reports that more than half of employees in the cited research said they would use AI tools without formal approval, while nearly one-third concealed their use from employers.
This behavior is often described as shadow AI. Shadow AI should not be treated only as an employee-discipline problem. It may indicate that demand for AI is stronger than the approved technology environment can support.
Employees may turn to unauthorized tools because:
Approved systems are unavailable. Procurement is too slow. Policies are confusing. Internal tools perform poorly. Employees fear being judged for using AI. Managers discourage experimentation. Useful workflows have no official path to approval. A mature organization studies shadow AI to understand unmet demand while still enforcing necessary security and privacy standards. A: Align Support Systems Generic training is rarely enough. A one-hour webinar explaining prompts may increase awareness but does not necessarily change work. Support should be tailored to employee roles, experience, risk exposure, and workflow.
A useful learning system has multiple layers. Foundational AI Literacy
Every employee should understand:
What generative AI can and cannot do. Why outputs may be inaccurate. How confidential information should be handled. What company policies permit. When human verification is required. How bias and inconsistency can appear. How to identify unsafe or inappropriate uses. Role-Specific Training Employees should learn how AI applies to their actual responsibilities. A finance team requires different examples from a marketing team. A software engineer needs different safeguards from a human-resources manager. A healthcare employee requires different controls from a retailer.
Training should use realistic tasks, approved data, and actual business scenarios. Practice Environments Employees need safe places to experiment.
These may include:
Sandboxed systems. Approved sample data. Prompt libraries. Test workflows. Internal hackathons. Guided experiments. Simulation exercises. Team-based problem-solving sessions. Peer Support
Organizations can establish:
AI champions. Departmental coaches. Communities of practice. Office hours. Internal discussion channels. Demonstration sessions. Case libraries. Reusable workflow repositories. Wharton highlights PwC’s use of experimentation sessions and internal peer “activators” as an example of combining practice, autonomy, and social learning. Manager Training Managers need more than end-user skills.
They must learn how to:
Select valuable use cases. Redesign responsibilities. Evaluate AI-assisted work. Set realistic performance expectations. Manage employee anxiety. Identify risk. Encourage experimentation without sacrificing accountability. Measure outcomes instead of activity. Escalate technical or ethical concerns. An organization cannot build an AI-enabled workforce with managers who do not understand how AI changes work. R: Redesign Roles for Human-AI Complementarity Many AI programs underperform because organizations automate isolated tasks without redesigning the surrounding process.
This can produce what might be called digital clutter. Employees perform the old workflow, use the new AI tool, verify the output, transfer information between systems, correct mistakes, and complete additional reporting requirements. Instead of saving time, AI adds another layer of work. McKinsey’s research indicates that workflow redesign is strongly associated with the financial impact organizations obtain from generative AI. Yet only a minority of organizations reported fundamentally redesigning even some workflows. A better approach begins with the full workflow.
Leaders should map:
The desired business outcome. Every major step required to achieve it. The people and systems involved. Delays, duplication, and error points. Tasks suitable for automation. Tasks suitable for AI assistance. Decisions requiring human authority. Data and integration requirements. Risk controls. Performance measures. Then the organization can assign work according to comparative strengths.
AI Is Often Strong at:
Processing large volumes of information. Retrieving relevant documents. Producing first drafts. Classifying or organizing content. Identifying patterns. Summarizing. Generating alternatives. Translating. Performing routine calculations. Monitoring for defined conditions. Automating repetitive digital actions.
Humans Remain Essential for:
Accountability. Ethical judgment. Contextual understanding. Empathy. Negotiation. Relationship building. Leadership. Strategic prioritization. Ambiguous decisions. Exceptions. Sensitive communication. Creative direction.
Final approval in high-risk situations. Complementarity does not mean preserving every existing task. It means designing roles in which AI absorbs appropriate work while humans gain responsibility for higher-value judgment, relationships, innovation, oversight, or decisions. E: Empower Employees Through Transparency and Participation Employees are more likely to support an AI program they helped shape. Participation can occur at several levels.
Employees can:
Identify repetitive work. Recommend high-value use cases. Test vendors. Design prompts and workflows. Define quality standards. Evaluate risks. Participate in governance committees. Create training materials. Mentor colleagues. Report performance problems. Recommend when AI should not be used. Wharton points to organizations including BNY, Colgate-Palmolive, and Johnson & Johnson as examples of broadening access or involving employees in identifying AI opportunities.
Participation changes the employee’s role from recipient to co-designer. That distinction matters. People tend to resist transformation imposed on them more strongly than transformation built with them.
The Enterprise AI Adoption System The AWARE framework addresses the human side of adoption. Organizations should place it inside a broader operating model. A complete AI adoption system contains six connected layers. Layer 1: Business Value AI projects should begin with a business problem, not with a technology demonstration.
Useful questions include:
What important outcome needs to improve? Where is the organization losing time or money? Which processes frustrate customers or employees? Where are decisions delayed by inaccessible information? Which workflows contain excessive manual work? Where does quality vary significantly? Which opportunities cannot be pursued because the organization lacks capacity? What new product or service could AI make possible?
A clear use case should include:
Target user. Business problem. Current workflow. Proposed AI contribution. Expected benefit. Required data. Human oversight. Risks. Success measures. Owner. Layer 2: Technical Readiness The organization must determine whether the environment can support the use case.
Considerations include:
Data quality. Data access. System integration. Identity and permission controls. Model selection. Reliability. Latency. Scalability. Vendor dependency. Security. Auditability. Monitoring.
Cost. A brilliant employee-adoption program cannot rescue a tool that is slow, inaccurate, inaccessible, or disconnected from actual work. Layer 3: Workflow Design The company should redesign the process around the target outcome. This may require eliminating steps, changing decision rights, connecting systems, or redefining roles before AI is introduced. Dell and Moderna are cited by Wharton as examples of organizations focusing on process simplification, collaborative redesign, or closer integration between people and technology functions. Layer 4: Human Adoption The organization applies the AWARE principles.
It assesses:
Competence. Autonomy. Relatedness. Trust. security. fairness. readiness. manager capability. learning needs. employee participation. Layer 5: Governance and Risk AI adoption without governance can create legal, financial, operational, security, and reputational exposure.
The U.S. National Institute of Standards and Technology provides a voluntary AI Risk Management Framework intended to help organizations incorporate trustworthiness into the design, development, use, and evaluation of AI systems. NIST has also released a generative AI profile addressing risks specific to generative systems.
An enterprise governance model should define:
Approved and prohibited uses. Data-handling requirements. Human-review thresholds. Documentation standards. Testing procedures. Risk ownership. Vendor evaluation. Incident response. Model monitoring. Audit requirements. Employee responsibilities. Customer disclosure rules.
Escalation pathways. Governance should make responsible adoption easier, not merely create another approval maze. Layer 6: Measurement and Learning AI implementation should be managed as a learning system.
The organization needs to know:
Who is using the system? For which workflows? How frequently? What outcomes are improving? Where is quality declining? How much time is genuinely saved? Are employees satisfied? Are customers affected? Are new risks emerging? Is the use case economically sustainable? Should it be expanded, revised, or discontinued? Measurement should extend beyond license activation or prompt volume.
High usage can coexist with low value. Low usage can indicate either resistance or a poorly designed tool. The metric must match the business objective.
A Practical 90-Day AI Adoption Roadmap Days 1 - 15: Diagnose Begin with discovery.
Actions:
Select one business function or workflow. Interview employees and managers. Map the current process. Identify pain points. Document unauthorized AI usage. Review current policies. Assess available data and systems. Identify psychological concerns. Establish baseline performance measures. Appoint an accountable business owner.
Deliverable:
A documented adoption opportunity containing the business problem, workflow, employee concerns, technical requirements, risks, and expected value. Days 16 - 30: Design Create the future workflow.
Actions:
Decide which tasks AI will automate. Decide which tasks AI will assist. Define human approval points. Establish quality standards. Select a limited pilot group. Create role-specific training. Define permitted data. Establish risk controls. Build employee feedback channels. Set measurable pilot objectives.
Deliverable:
A pilot design with clear roles, responsibilities, safeguards, training, and success metrics. Days 31 - 60: Pilot Run the workflow with a controlled group.
Actions:
Train participants. Allow guided experimentation. Hold weekly feedback sessions. Monitor errors and workarounds. Measure time, cost, quality, and satisfaction. Compare results with the baseline. Document unexpected uses. Address psychological and operational concerns. Revise prompts, integrations, and policies. Share lessons transparently.
Deliverable:
Evidence showing whether the use case improves the target outcome and whether employees can use it safely and effectively. Days 61 - 75: Redesign Do not scale the original pilot automatically. Refine it.
Actions:
Remove unnecessary process steps. Improve system integration. Clarify responsibilities. Adjust training. Strengthen controls. Correct incentive conflicts. Establish support resources. Identify which employee groups require additional help. Determine whether the economic case remains valid.
Deliverable:
A production-ready operating model rather than a larger experiment. Days 76 - 90: Scale Selectively Expand only when evidence supports expansion.
Actions:
Add users gradually. Certify managers or champions. Publish approved workflows. Create reusable templates. Monitor outcome metrics. Review incidents. Track employee confidence. Maintain a feedback channel. Conduct periodic risk reviews. Reassess the process as models and business needs change.
Deliverable:
A governed, measurable, repeatable AI-enabled workflow with clear ownership.
How Small and Medium-Sized Businesses Can Approach AI Adoption Large enterprises can create AI centers of excellence, specialized governance teams, and extensive training programs. Smaller companies need a simpler model.
A practical small-business approach is:
Choose One Painful Workflow Do not begin with an enterprise-wide AI strategy.
Select one process such as:
Preparing proposals. Answering repetitive customer questions. Summarizing meetings. Researching potential customers. Drafting marketing content. Organizing internal documents. Reviewing invoices. Preparing job descriptions. Producing first drafts of reports. Use an Approved General-Purpose Tool Avoid building custom systems before understanding employee needs. Select a reputable tool with appropriate business privacy and administrative controls.
Define a Simple Use Policy
Employees should know:
Which information may be entered. Which information is prohibited. Which outputs require review. Who is responsible for final decisions. How errors should be reported. Measure One or Two Outcomes
For example:
Time per proposal. Customer response time. Number of qualified leads researched. Editing time. Error rate. Employee satisfaction. Improve the Workflow Before Expanding Once the company has evidence, it can add integrations, automation, custom data, or agents. Small businesses should resist the temptation to adopt ten AI tools simultaneously. A focused implementation often produces more value than a large collection of disconnected subscriptions.
Common AI Adoption Mistakes Mistake 1: Buying Licenses Before Identifying Workflows Technology purchasing should follow use-case selection. Otherwise, employees receive a tool without knowing where it creates value. Mistake 2: Measuring Adoption by Login Counts Activity is not transformation. Measure operational outcomes. Mistake 3: Treating Training as a One-Time Event AI capabilities and workflows evolve continuously. Learning must be ongoing. Mistake 4: Ignoring Managers Managers translate strategy into daily behavior.
An uninformed manager can suppress experimentation or create unsafe pressure. Mistake 5: Automating a Broken Process AI may accelerate inefficiency. Simplify the workflow first. Mistake 6: Avoiding Honest Discussions About Jobs Employees will fill silence with speculation. Communicate what is known, what remains uncertain, and how decisions will be made. Mistake 7: Creating Governance That No One Can Navigate Overly complicated approval systems encourage shadow AI. Controls should be understandable and proportionate to risk. Mistake 8: Allowing Unequal Access A small executive group should not capture all learning opportunities.
Broader participation increases trust and generates more use cases. Mistake 9: Expecting Immediate Financial Returns Adoption includes experimentation, process redesign, training, integration, and iteration. Not every pilot will succeed. Mistake 10: Treating Employee Resistance as Disobedience Resistance may reveal poor communication, weak tools, hidden risks, unrealistic expectations, or flawed process design. Listen before escalating.
How to Measure AI Adoption Properly A balanced measurement system should include five categories.
1. Usage
Active users. Frequency of use. Repeat usage. Workflows used. Departmental participation. Approved versus unauthorized use.
2. Business Outcomes
Revenue improvement. Cost reduction. Cycle-time reduction. Higher conversion. Faster product development. Lower error rates. Better service resolution. Increased capacity.
3. Quality
Accuracy. Completeness. Consistency. Customer satisfaction. Rework. Human override rate. Hallucination or failure rate.
4. Employee Experience
Confidence. Perceived usefulness. Trust. Workload. Autonomy. Skill development. Job satisfaction. Psychological safety.
5. Risk
Privacy incidents. Security events. Policy violations. Biased outcomes. Incorrect customer communication. Intellectual-property exposure. Regulatory exceptions. Unreviewed high-impact decisions. A company should not declare success simply because employees use AI.
The relevant question is:
Are employees using AI in ways that improve business outcomes while preserving quality, accountability, security, and trust?
Key Takeaways
AI adoption is a transformation challenge, not merely a software deployment. Employee resistance may reflect threats to competence, autonomy, relatedness, security, fairness, or professional identity. Leaders should acknowledge concerns openly rather than relying on slogans or mandatory training. Shadow AI is both a risk and a signal that employees have unmet demand for useful tools. Generic training should be replaced with role-specific learning, guided experimentation, peer support, and manager education. Workflow redesign is more important than placing AI on top of existing processes. Human-AI complementarity requires deliberately assigning work according to the strengths of people and machines. Employees should participate in selecting use cases, testing workflows, defining standards, and identifying risks. Governance should create clarity and safe pathways, not an administrative barrier that drives usage underground. Adoption metrics must include business outcomes, quality, employee experience, and risk, not only license activation. Organizations should begin with a small number of high-value workflows, test them carefully, and scale only after obtaining evidence. The greatest AI advantage may belong to organizations that redesign themselves effectively, rather than those that purchase the most advanced models.
Frequently Asked Questions
1. Why do employees resist AI?
Employees may resist because they fear losing expertise, control, job security, privacy, professional identity, or relationships. Resistance may also result from poor tools, unclear policies, inadequate training, unrealistic expectations, or distrust of leadership.
2. Is AI resistance mainly a skills problem?
No. Skills matter, but resistance can also be psychological and organizational. An employee may understand how to use AI while still believing it threatens their role or autonomy.
3. What is the AWARE framework?
AWARE is a leadership framework for reducing resistance to generative AI:
Acknowledge psychological impact. Watch coping behaviors. Align support systems. Redesign roles for complementarity. Empower employees through participation and transparency.
4. What is shadow AI?
Shadow AI refers to employees using unauthorized or undisclosed AI tools for work. It can create privacy, security, compliance, and intellectual-property risks.
5. Should companies ban public AI tools?
Companies should apply controls based on risk. Blanket bans may sometimes be necessary for highly sensitive work, but they can also push use underground. A better long-term strategy is to provide approved alternatives, clear policies, training, and monitoring.
6. How should a company choose its first AI use case?
Select a workflow that is repetitive, measurable, important, sufficiently documented, and manageable in risk. The use case should solve a real business problem rather than merely demonstrate AI capability.
7. How much training do employees need?
Training requirements vary by role. All employees need foundational literacy, while employees handling sensitive, technical, regulated, or high-impact work need deeper role-specific instruction.
8. Should AI automate entire jobs?
Most near-term implementations are more practical when they automate or augment tasks within jobs. Entire workflows may eventually change significantly, but organizations should evaluate accountability, exceptions, customer impact, and human judgment carefully.
9. How can managers encourage adoption without forcing it?
Managers can demonstrate relevant use cases, provide safe practice environments, invite employee participation, recognize learning, explain expectations, and measure outcomes instead of raw usage.
10. How should companies communicate about job impact?
Leaders should communicate what tasks are changing, which skills will matter, how productivity gains will be used, what retraining is available, and how workforce decisions will be made. They should avoid promises they cannot guarantee.
11. What is human-AI complementarity?
Human-AI complementarity means designing work so that AI handles suitable information-processing, generation, or repetitive tasks while humans retain responsibility for judgment, relationships, ethics, strategy, exceptions, and accountability.
12. How can organizations know whether adoption is successful?
Success should be measured through business outcomes, quality, employee confidence, customer effects, risk levels, repeat usage, and financial sustainability.
13. Is high usage always a positive sign?
No. Employees may use AI frequently for low-value tasks, or usage may increase while quality declines. Usage must be connected to outcomes.
14. Who should own enterprise AI adoption?
Ownership should be shared. Business leaders should own outcomes, technology teams should own infrastructure and integration, risk functions should establish safeguards, HR should support workforce transition, and employees should help design workflows.
15. How long does AI adoption take?
A focused use case can be piloted within weeks or months. Broader transformation requires ongoing redesign, learning, governance, and cultural change. AI adoption should be treated as a continuing organizational capability rather than a one-time project.
Conclusion
The most difficult part of enterprise AI adoption is not gaining access to artificial intelligence. It is changing the organization around it. Companies must decide which problems deserve attention, which workflows should be redesigned, which responsibilities should remain human, how employees will develop new capabilities, and how trust will be maintained during uncertainty. The Wharton AWARE framework offers a useful reminder that adoption depends on psychological conditions as well as technical capability. Employees need to believe they can become competent in the new environment. They need enough autonomy to exercise judgment. They need to remain connected to colleagues and organizational purpose. They need credible information about how their work will change. They need fair access to tools, training, and opportunities. They need governance that protects them without preventing useful experimentation. AI should not be inserted into an organization as though work were a collection of independent mechanical tasks. Work is a system of people, information, incentives, decisions, relationships, responsibilities, and meaning.
Changing one part of that system changes the others. The companies that understand this will not define AI adoption as the number of licenses purchased, pilots launched, prompts submitted, or agents created. They will define adoption by whether people and machines can work together to produce better results, stronger capabilities, safer decisions, and more valuable forms of human contribution. That is the difference between owning AI technology and becoming an AI-enabled organization.
Relevant Articles and Resources
1. Knowledge at Wharton: “AI Adoption Is a Challenge. Here’s a Solution”
The foundational article introducing the AWARE framework and explaining how competence, autonomy, and relatedness affect employee resistance to generative AI.
2. Harvard Business Review: “Why Gen AI Feels So Threatening to Workers”
An examination of how generative AI can satisfy or frustrate psychological needs and influence employee acceptance, motivation, and well-being.
3. McKinsey & Company: “The State of AI: Global Survey 2025”
Research on the expansion of enterprise AI use, agentic AI experimentation, scaling barriers, and the continuing gap between pilots and measurable enterprise impact.
4. McKinsey & Company: “The State of AI: How Organizations Are Rewiring to Capture Value”
Research emphasizing the importance of redesigning workflows rather than simply deploying generative AI tools.
5. McKinsey & Company: “The Learning Organization: How to Accelerate AI Adoption”
Guidance on organizational learning, frontline experimentation, management barriers, and the conditions required to spread successful AI practices.
6. Microsoft: 2025 Work Trend Index
A global study examining the development of human-agent work, organizational capacity, AI skills, and emerging operating models. Microsoft reports that the study included 31,000 knowledge workers across 31 markets, alongside labor-market and productivity data.
7. NIST Artificial Intelligence Risk Management Framework
A voluntary framework for helping organizations manage trustworthiness and risk throughout the design, development, deployment, and use of AI systems.
8. NIST Generative Artificial Intelligence Profile
A companion resource addressing risks and management considerations that are particularly relevant to generative AI.
9. OECD: “Bridging the AI Skills Gap”
Research addressing the growing need for specialist AI skills, broader AI literacy, workforce development, upskilling, and reskilling.
10. OECD: “Making AI Work: Why Investing in Skills Matters”
A discussion of how skills gaps can slow adoption and why investment in workforce capability is necessary for AI to produce productivity and economic benefits.