The UAE has launched one of the most ambitious public-sector AI capability programs anywhere. In May, the Cabinet approved a plan to train 80,000 federal employees in Agentic AI, from ministers and senior executives to technical staff and new joiners, as part of a wider effort to move at least half of federal services and operations toward agentic models within two years.
The training design already contains several smart choices. It divides employees into five occupational categories, plans personalized learning paths based on role and competency, and treats capability-building as continuous rather than as a one-time course. Those features reflect how AI skills actually develop.
Still, the difficult part begins after people complete the training. Organizations routinely confuse learning exposure with behavioral adoption. An employee can finish a course, pass an assessment, and return to a workplace where managers discourage experimentation, rules remain vague, and the safest career move is to keep doing the job exactly as before.
Skill Follows Safety
AI training works best when employees know what using AI will mean for their jobs. Someone who fears that efficiency gains will eventually eliminate a role hears a prompt-engineering lesson differently from someone who sees AI as a path to promotion. Someone who worries that using AI makes them look less competent may quietly use it without telling colleagues. Someone who fears punishment for a mistake will avoid experimentation no matter how polished the training portal looks.
This is where organizational psychology becomes operational. Before asking employees to change how they work, leaders need credible answers to basic questions. Which tasks are expected to change? Which decisions stay human? How will mistakes be handled? What happens to time saved? What support will employees receive as roles evolve?
Microsoft’s 2026 Work Trend Index offers a useful global benchmark. Organizational factors such as culture, manager support, and talent practices accounted for 67 percent of reported AI impact in its analysis, versus 32 percent for individual mindset and behavior. Only 13 percent of AI users surveyed said they were rewarded for reinventing work with AI even when results were not achieved.
That gap explains why training can disappoint. Employees may be taught to experiment while the performance system punishes experimentation. The curriculum says innovation; the workplace says do not risk missing your current target.
Turn Training Into Building
The most effective AI learning moves quickly from instruction to supervised practice on real work. Employees should learn the basic capabilities and limits of the tools, then use them to improve a workflow they actually own. A procurement specialist might build a first-pass supplier comparison. An HR employee might create a controlled onboarding assistant. A service employee might design an agent that prepares a case summary before a human decision.
The UAE already has a powerful domestic example of this model. At the national Agentic AI retreat, officials reported that ADNOC had trained 20,000 employees to build job-specific Agentic AI models, with 3,000 active models supporting daily tasks and Agentic AI utilization reaching 80 percent over the previous 90 days. The important detail is the verb build. Employees become more capable when they create and adapt tools around work they understand.
Give Managers a Role in the Learning System
Training also needs a management layer. Direct managers decide whether employees get time to experiment, whether mistakes become learning opportunities, whether AI use is discussed openly, and whether strong use cases spread across the team. If managers receive only the same generic course as everyone else, they remain unprepared for the conversations that determine adoption.
Managers need practical scripts and routines. Ask each team to choose one workflow to improve. Reserve protected time for small experiments. Review one successful and one failed AI use case each month. Clarify which data can enter which tools. Reward employees who document reusable methods and help peers. Those behaviors turn a training program into a learning system.
Measure Capability, Not Course Completion
The final shift is measurement. Completion rates are useful administrative data, but they should sit near the bottom of the dashboard. Better indicators include the share of priority workflows with trained internal owners, the number of employee-built tools that reach production, time from idea to working prototype, reductions in rework, and measurable business outcomes.
The UAE’s scale creates an unusual opportunity. It can demonstrate what workforce AI transformation looks like when training, governance, managers, incentives, and real workflow redesign reinforce each other. The 80,000-person program is a major start. Its long-term value will depend on whether employees leave training with permission, support, and a reason to work differently on Monday morning.
