Enterprise AI Readiness Is a Workforce Capability Strategy,

Not a Learning Initiative

Most organizations are still putting AI readiness in the wrong budget under the wrong owner. They purchase model access rights, establish tool policies, conduct training sessions, and measure completions. Such a process views AI as a problem of content. Yet facts suggest otherwise. Adoption of AI is possible only when a business translates use of models into altered work, designated skill requirements, new role definitions, manager responsibilities, and staffing considerations. Various entities ranging from OECD to DOL, NIST, IMF, ILO, Stanford, the EU, and the World Economic Forum all describe various aspects of such an operating reality. Taken collectively, they prove that AI readiness should be integrated into the process of workforce planning related to job redesign and governance, not into separate learning campaigns.

Innovation in this area is rapid. DOL explains that guidance on AI literacy would continue to evolve with advances in AI technologies and changes in labor market. NIST describes the AI RMF as a living document and suggests that future revisions to the Generative AI Profile will introduce new risks and mitigating actions. IMF demonstrates that development of new skills takes just months within labor markets, not decades. A proper organization-level approach must include regular updates to the plan. The areas most affected by changes are vendor control mechanisms, modeling constraints, job posting indicators, and regulatory obligations for use of AI.

The objective is simple. To go from AI excitement to readiness of the workforce through skill architecture, role segmentation, leadership and practitioner career tracks, and quantifiable skill lift in live workflows. The initial step to achieving the objective is developing clear definitions. DOL describes AI literacy as a foundation for the development of competencies for ethical and effective use and evaluation of AI systems with a particular emphasis on generative AI. The EU AI Act requires that providers and users of AI take appropriate measures to ensure a certain level of AI literacy among staff members and others involved in activities, based on technical expertise, experience, education, training, and context of use. Enterprise-wise, the term means technical proficiency that enables a particular role to evaluate results of modeling, understand its constraints, engage in cooperation with data and risk experts, and apply proper response in practice.

The training frame is too small

The problem of skill gaps is currently among the top constraints for businesses. As stated in the report of the World Economic Forum, 63 percent of surveyed employers regard skills as a critical challenge for the transformation of their businesses between 2025 and 2030. In its 2026 briefing paper, OECD stresses that internal AI skills gaps are widespread barriers and skills requirements are constantly evolving. Also, as noted in a study of 51 successful implementations by Stanford, most of the effort was required in managing change, improving data quality, and re-designing work, but not in creating the model. All those statements can be viewed as related. Training is important, although the scope is too narrow. A firm must choose work to be changed, who will do it, how judgment will be required, and how risk will be managed.

Another issue that changes in favor of enterprises is the legal floor. The EU AI Act directly imposes the literacy obligation on the provider and deployer of AI systems. Moreover, as stated in DOL's 2027 guidance, programs must promote AI skill development of their workforce in their setting. Besides, OECD provides another point that has implications for firms – namely, almost all employees require some literacy to use responsibly, while professionals and IT staff require deeper expertise, and leaders need to have knowledge of strategy and governance of AI systems. Clearly, training alone cannot fulfill such complex expectations. Firms should develop a workforce literacy program by role, depth, and accountability.

The change does not apply to occupations, but to specific tasks of those workers. The International Labour Organization updates its assessment methodology and conducts an analysis at the 6-digit occupation level, covering almost 30,000 tasks. It is concluded that approximately 25 percent of employees work in an occupation exposed to some extent to GenAI technology, yet the probability of transformation exceeds that of redundancy by two to one. Also, the DOL guidance emphasizes the same idea, albeit from a different angle – employees acquire literacy better by working in their daily operations. If firms approach the matter of literacy via training tools, they miss the actual target – the operating unit.

Current skill needs and future skill needs are not the same thing

The criteria for the present state of readiness include broad literacy and task-relevant judgment. According to the 2025 OECD brief on the skill gap, the share of jobs requiring AI-related skills is very low (about 1 percent according to the estimate provided). However, the majority of jobs exposed to AI will require literacy in how to interact with AI. Further, the 2026 OECD brief clarifies that digital and data professionals require high-level expertise in the area; other employees will require basic AI literacy. Also, as AI becomes involved, some of the additional skills required become those related to problems solving, critical analysis, teamwork, and leadership. DOL notes that responsible utilization requires data protection, risk assessment, policy awareness, and human accountability of results. Present-day enterprise readiness starts here.

However, the future-state needs become harder to project. According to WEF, about four in ten skill sets among existing workers are going to get changed or completely become obsolete by 2030; 59 out of every 100 workers will need re-skilling by that time. Also, seven in ten companies expect that they will have to hire new staff trained for different skills. Fifty percent expect internal movement between declining and growing occupations as well. The IMF notes that now, every tenth job posting in an advanced economy demands at least one novel skill. Furthermore, demand for novel skills concentrates in professional, technical, and managerial jobs. Also, according to the report, IT skills are involved in more than half of all such posts, and a rising percentage is due to AI. An enterprise which tries to map only current skill shortages is doomed to create understaffing in the coming years.

As a result, the enterprise should develop a multi-year plan. There is no single document saying exactly what timeframe we are supposed to follow; yet, the trend is quite clear. WEF utilizes a five-year period in employer planning (2025-2030); OECD says that skill needs are constantly changing. Stanford notes that successful deployments have accumulated substantial sunk cost of previous failures and large investment in non-models. In turn, both DOL and NIST emphasize that their guidance will constantly change with the development of AI and work processes in general. The enterprise conclusion is clear – an enterprise needs a rolling workforce plan with a relatively long-term perspective and constant reviews of that plan.

Recommendation one: build a present-to-future skill map from work, not from tools

First, there is no single level of human oversight appropriate for all workflows within a business. Stanford found that the optimal amount of human oversight depends on several factors: task nature, risk tolerance, regulations, and complexity. NIST states that the responsibility of an AI actor will be different across design, development, deployment, operation and monitoring, testing, evaluation, procurement, impact assessment, and governance stages. This finding becomes the basis for a company-wide rule: do not start with a technology stack. Start with the processes. Identify the task, its inputs, criteria for assessing results, review points, escalation process, and the individual or team responsible for taking final action.

This process inventory must distinguish between current and future skills. The former relates to the tool itself, its outputs, data management, escalation, and compliance with company policies. The latter relates to the redesigned role, workflow composition, evaluation methods, vendor review, monitoring, and system governance. According to OECD, most organizations fail to adopt an organized method for identifying AI-related skill needs. IMF proves that new skills are introduced unevenly across jobs and countries. DOL recommends that further learning opportunities allow clear paths toward higher-level skills and AI-related occupations. An organization failing to document the skill transition plan by process will be unable to estimate hiring needs, reskilling opportunities, and prioritization.

The ideal skill inventory should comprise four levels. The lowest level involves literacy required of the broader workforce. The next level includes technical fluency of practitioners who apply AI at work. The third level involves operational fluency of managers who set review standards, assign tasks, and evaluate performance. The fourth level requires specialists and control staff. This four-level classification is consistent with OECD's employee, leader, and digital/data professional categorization; DOL's distinction between baseline literacy and continued learning opportunities; and NIST's separation of actors throughout the AI life cycle. This is not a strict matrix. It is a practical staffing guide.

Recommendation two: write role pathways for leaders, practitioners, and control functions

One such course, “AI upskilling,” misses the mark entirely by addressing a role-differentiated challenge with undifferentiated instruction. According to DOL, foundational literacy is merely step one and calls for clear pathways to deeper technical skills, job-relevant tool application, and AI-based occupations. Continued education should leverage stackable curricula and occupation-specific pathways. According to OECD, there will be separate training tracks for leaders, employees broadly, and digital/data professionals specifically. According to WEF, employers do not intend merely to retool existing workers but to hire, redeploy, and transition to different roles. Role pathways are not optional; they are the bridge between literacy and enterprise labor force supply.

Leader roles require specific knowledge about workflow selection, risk tolerance, review criteria, vendor selection, data considerations, staffing plans, and business-case discipline. DOL indicates that managers and other roles require custom AI literacy for managing team adoption and integration and change management. According to OECD, leaders need strategic appreciation for AI's benefits and risks and governance, data stewardship, workforce preparation, collaboration, communication, and change management skills. In its survey of enterprises, Stanford University found that a co-sponsoring model of business units and tech groups led to success and that departmental programs plateaued when more extensive participation was necessary. Leadership requires a pathway based on the decisions that they need to make.

Practitioner pathways must address the specific jobs at hand. Analysts, recruiters, customer service agents, software engineers, marketing professionals, and operations staff all need different levels of literacy. DOL states that AI literacy should include task integration with live feedback and context-aware delivery. According to Stanford University, successful enterprise applications were either fully automated or included human-in-the-loop, rather than being fully autonomous. This affects the requirements on the person. The practitioner pathway must train people in framing tasks, evaluating output, revising output, escalating edge cases, and documenting activity. This is hands-on knowledge.

Specialist roles include both AI builders and control functions. NIST details these responsibilities: design, development, deployment, operations and monitoring, testing and validation, human factors, domain expertise, impact assessment, procurement, and governance. According to OECD, digital and data professionals should receive technical training with ethical and regulatory awareness and inter-disciplinary collaboration skills. The specialist labor market is tight, and the IMF reports that demand for new skills will be most intense in higher-skill occupations. This implies that enterprises have a business case for moving workers from similar positions into more advanced and technical ones. Specialists' feeder paths should include evaluation, vendor review, model operations, prompt operations, data engineering support, and AI product ownership roles.

Recommendation three: put each line of business on a workforce action plan

A plan for workforce is truly actionable only if it gets adopted by each particular business unit. OECD stresses that workforce development should occur within the overall strategy regarding AI, data and technology, and that any changes in work organization require looking forward to workforce development. DOL states that embedding AI learning in the context of the worker's job, industry, and training regime makes it more actionable. According to Stanford University, the adoption of AI stalled whenever the teams saw the initiative as an IT project but not a project aimed at work redesign and change. The line-of-business workforce plan will take care of all that. Each particular business unit should define its targeted workflow, current gaps in skills, future demands for roles, managers' commitment to the process, ways to retrain the workforce, hiring requirements, and measures of productivity.

The plan should include concrete decisions on buy-vs-build-vs-borrow questions. OECD lists three levers available to institutions: outsourcing, hiring, and training. In this regard, the OECD message about the need for balance is fully applicable to enterprises as well. While outsourcing and/or hiring can solve some immediate problems, internal development is the only way to develop unique judgement and reduce information asymmetries associated with procurement. The WEF Employer Data for 2025 proves the combination of those tools right. Enterprises plan upskilling, hiring, and redeployment all at the same time. Thus, the business workforce action plan should state which gaps can be addressed by hiring, which – by retraining, which can be addressed by outsourcing, and which roles are going to be discontinued.

Manager ownership is non-negotiable. DOL speaks about managers needing special preparation for managing teams and integrating AI. According to Stanford, breakthroughs in cross-functional work started when both the technical and business sponsors of the project became involved, and when the goals set by the company were made a driver of success. NIST lists organizational management, leadership from senior levels, and even the board among the duties that go under governance and oversight. While learning and development teams could contribute to the process, business-line managers had better own it and make changes in workflows and staff composition.

Recommendation four: measure skill lift in production, not in the LMS

Completions Do Not Equal Readiness. According to the DOL, the program should ensure measurable development of AI skills and knowledge within its own organization and should apply outcome-driven iteration to determine whether the participants gain relevant and transferrable skills and competencies. Stanford claims that companies defining relevant KPIs before deployment will be better able to prove value and attract funding. The NIST's Generative-AI profile recommends that organizations apply governance, mapping, measuring, and risk management activities to fit their own goals, risk tolerance levels, and use contexts. What counts is not attendance, but changed performance in live conditions.

What is needed are role- and workflow-specific measures. For the worker, the relevant metrics include output quality, review accuracy, escalation quality, cycle time, rework rate, and safe handling of data. For managers, relevant metrics are adoption in the team, review compliance, incident rate, and time to proficiency in redesigned workflows. Specialists have other metrics to keep in check, including evaluation scope and frequency, vendor reviews, monitoring quality, audit results, and defect or policy breach resolution time. Stanford reports deployment data to provide numerous function-level case examples on KPI selection. In addition, NIST highlights that measurement should leverage domain experts, independent assessments, user feedback, and field data.

Here lies the point. Enterprise AI readiness comes down to the company being able to translate access to the models into the enterprise-level skill requirements, role pathways, and workforce plans for all lines of business. This conclusion follows directly from the available data: Workers will mostly need literacy, not specialist training; Leaders need to develop strategic vision and AI governance capabilities; Practitioners will require practical training in redesigning work and reviewing outputs; Technical and control roles will need advanced role pathway design; and the process will take several planning cycles, connected to the evolution of the company's operating model.

In practice, it means that the company should be able to identify the changes in its workflows, skill gaps at hand, new roles in demand, reskilling options, hiring needs, review guidelines, and relevant production KPIs. Can the company demonstrate that the literacy level corresponds to role pathways, managerial education – to the decision-making processes, and specialist training – to system development and risk management? If not, the organization has just entered the training phase of the workforce challenge.

Next Up: Designing the Human-Ai Enterprise Series