The public conversation about AI and work is often trapped between optimism and fear. One side emphasizes productivity, new jobs, and human augmentation. The other warns about displacement, wage pressure, and automation at scale. Both perspectives contain truth, but neither is sufficient for institutional strategy. The more practical question is not whether AI will affect work. It already is. The question is how governments, employers, universities, and civic institutions should prepare people for a labor market where work is increasingly performed by human-AI teams.
Microsoft’s WorkLab has argued that organizations are moving toward a future in which employees manage teams of AI agents and leaders think about new human-agent ratios.1 The World Economic Forum’s *Future of Jobs Report 2025* identifies AI and information-processing technologies as major drivers of labor-market transformation, while also emphasizing the importance of analytical thinking, resilience, flexibility, leadership, and lifelong learning.2 The International Labour Organization has similarly warned that generative AI is more likely to transform tasks than eliminate entire occupations in the near term, with clerical and administrative work especially exposed to change.3
This means the workforce issue cannot be solved by teaching people to use one AI product. The task is broader: redesigning roles, skills, training systems, career pathways, and institutional support around the reality that AI will increasingly participate in research, drafting, coding, customer support, compliance, program delivery, operations, and management.
The wrong frame: automation versus jobs
The question “Will AI take jobs?” is emotionally powerful but strategically incomplete. Work is made of tasks, and tasks change before occupations disappear. A policy analyst may spend less time preparing first drafts and more time validating evidence. A program officer may spend less time compiling reports and more time interpreting implementation signals. A procurement specialist may spend less time searching contracts and more time managing supplier risk. A teacher may spend less time creating generic lesson materials and more time adapting instruction to learners.
This task-level view is important because it shows that AI’s impact will be uneven inside the same job. Some tasks will be automated. Some will be accelerated. Some will become more valuable. Some will require new oversight. The workforce strategy, therefore, should not be built only around job replacement forecasts. It should be built around task redesign and capability development.
| Old workforce question | Better workforce question |
|---|---|
| Which jobs will AI replace? | Which tasks will be automated, augmented, or made more valuable? |
| How many people know how to prompt? | How many people can use AI responsibly inside real workflows? |
| How do we give everyone access to tools? | How do we build role-specific competence, judgment, and safeguards? |
| How do we train once? | How do we create continuous learning as tools and workflows change? |
The difference matters for policy. If leaders focus only on job loss, they may miss the need to upgrade millions of existing roles. If they focus only on productivity, they may ignore transition costs, worker voice, and inequality. A serious workforce strategy must do both: build capability and protect people.
The new skill premium
The first wave of AI literacy often focused on prompting. Prompting still matters, but it is too narrow. As AI systems become more integrated into workflows, the valuable worker will need a wider set of skills.
The first is judgment. AI can produce fluent output quickly, but fluency is not truth. Workers must evaluate whether an output is accurate, complete, lawful, ethical, and appropriate for context. This is especially important in policy, education, health, finance, legal, and public administration environments.
The second is process thinking. Human-AI teams require people who can describe work clearly enough to redesign it. They must identify inputs, decision points, dependencies, risks, exceptions, and handoffs. This is the foundation for using AI in a workflow rather than as a separate writing assistant.
The third is verification. Workers need methods for checking claims, tracing sources, comparing outputs, testing edge cases, and documenting uncertainty. In research and policy settings, verification becomes a core professional skill.
The fourth is supervision. As agents and copilots take on more work, people will need to review intermediate outputs, approve actions, manage exceptions, and intervene when systems drift. This is a new managerial skill even for non-managers.
The fifth is ethical and civic reasoning. AI may make work faster, but faster is not always better. Workers need to understand when automation could harm fairness, access, privacy, accountability, or trust.
The most resilient workers will not be those who simply use AI more often. They will be those who can decide when AI should be used, how it should be supervised, and what human responsibility must remain.
Training must move from generic AI literacy to role-based capability
Many organizations are responding to AI with broad awareness sessions. These are useful as a starting point, but they rarely change performance. A lawyer, nurse, procurement officer, teacher, researcher, grant manager, and policy adviser do not need identical AI training. They need shared foundations and role-specific practice.
A better model has three layers. The first layer is universal AI literacy. Everyone should understand basic concepts: what generative AI can and cannot do, why hallucination occurs, how data privacy works, what kinds of tasks are appropriate, and when human review is required.
The second layer is role-based workflow training. People should practice AI use in the actual work patterns of their role: drafting a policy memo, reviewing a contract, preparing a lesson plan, summarizing stakeholder feedback, analyzing program data, or responding to a service request. This is where abstract literacy becomes usable skill.
The third layer is advanced workflow leadership. Selected staff should learn how to redesign processes, evaluate AI tools, manage agentic workflows, document controls, and support colleagues. These people become institutional multipliers.
| Training layer | Target group | Core outcome |
|---|---|---|
| Universal AI literacy | All staff and learners | Safe, informed, responsible use. |
| Role-based workflow training | Specific professions or teams | Better performance inside real tasks. |
| Advanced workflow leadership | Managers, analysts, digital teams, fellows | Capacity to redesign, evaluate, and govern AI-enabled work. |
This model is also relevant for universities and fellowship programs. Students and early-career professionals should not only learn AI as a technical subject. They should learn how AI changes policy analysis, research practice, organizational design, public service delivery, entrepreneurship, and institutional strategy.
Equity and access cannot be an afterthought
AI workforce strategy is also an equity strategy. Workers with access to high-quality tools, supportive managers, good data, and opportunities to redesign their roles may become more productive and valuable. Workers in under-resourced institutions may face the opposite: pressure to use AI without training, surveillance without empowerment, or displacement without transition support.
Countries and regions with weaker digital infrastructure may also fall behind if AI adoption depends only on private purchasing power. The divide will not be only between those who have internet access and those who do not. It will be between those who can use AI inside meaningful workflows and those who only have surface-level access.
This has implications for public policy and funders. Workforce programs should support AI capability in public institutions, small businesses, schools, civic organizations, and local governments, not only in large firms. Training should be connected to real opportunities: internships, applied projects, sector partnerships, entrepreneurship support, and recognized credentials. Public investment should help people move into higher-value roles rather than simply adapt to faster automation.
What institutions should do now
Institutions should begin with a task exposure map. Rather than guessing which jobs are at risk, they should map the tasks in priority roles and classify them as automatable, augmentable, judgment-intensive, relationship-intensive, or compliance-sensitive. This gives leaders a practical basis for training and redesign.
They should then create role-specific AI playbooks. A playbook should explain approved use cases, prohibited uses, data rules, review standards, example workflows, escalation points, and quality expectations. This makes AI use less dependent on individual improvisation.
Third, institutions should redesign performance expectations. If AI reduces time spent on routine drafting, what should that time be used for? Better analysis? More stakeholder engagement? Faster service? Higher-quality review? Without explicit expectations, productivity gains can disappear into busyness.
Fourth, leaders should involve workers in implementation. Workers understand where processes break, where AI might help, and where it could create harm. Participation improves design and reduces resistance.
Finally, workforce strategy should be linked to governance. Training without guardrails creates risk. Governance without training creates paralysis. The two must be built together.
Conclusion: the future worker is a responsible orchestrator
AI will not affect all workers equally, and it will not transform every institution at the same pace. But the direction is clear. More work will be supported by systems that draft, analyze, retrieve, recommend, coordinate, and act. The workforce challenge is to prepare people not only to use these systems, but to work with them responsibly.
This requires a new public and institutional agenda: role-based training, workflow redesign, worker participation, equity-focused access, and governance-linked capability building. The goal is not to turn every worker into a programmer. The goal is to help people become better analysts, supervisors, designers, communicators, and decision-makers in environments where AI is part of the team.
CentPol’s position is simple: the future of work should not be left to software adoption alone. It must be shaped through deliberate workforce strategy, institutional design, and public-interest investment.
