Investing in human jobs will determine who captures AI's gains

The global debate about AI focuses on what the technology can do. The more consequential question for most of the world is what people must build before AI can do it responsibly.

That question is not only about computing power or connectivity. It is about the human layer; the expertise, accountability and institutional capacity that allow AI to be deployed in ways that work and can be trusted. In most developing countries, that layer barely exists. Its absence is both the central obstacle to AI diffusion and the most significant job-creation opportunity of the next decade.

AI does not wait for national strategies or dedicated legislation to be in place. It enters countries through software updates, digital transformation initiatives, vendor platforms, cloud services, procurement decisions and existing institutional workflows. But beneath that is the harder question: Why is AI diffusing so slowly in the places that need it most – and what has to be built before it can spread responsibly?

The 2025 UNDP Human Development Report (HDR) offered a striking finding: Six in 10 survey respondents expect AI to create new job opportunities in low and medium HDI countries, while 70% expect it to increase their productivity. These are not the responses of people who feel left behind. They are the responses of people waiting for an entry point.

The barrier is not connectivity or compute alone. It is human and institutional. The institutions, accountability mechanisms and local capacity that allow AI to be deployed responsibly – and therefore adopted at scale – do not yet exist in most places. Where trust is absent, adoption stalls. Where accountability is weak, harm concentrates among those least able to challenge it.

This is why AI diffusion and the building of trust infrastructure cannot be treated as sequential stages. Waiting for adoption to occur and then retrofitting governance around it leaves the most consequential decisions unaccountable. UNDP’s AI Landscape Assessments (AILA) across 50 developing countries show this pattern repeatedly: Deployment timelines move ahead of governance readiness, and the gap between the two is where trust breaks down. An AI diagnostic tool deployed in a clinic without anyone qualified to evaluate its performance, contest its errors or explain its outputs to patients is not a development gain; it is a liability. The human layer is not a luxury that follows adoption. It is a precondition for it.

When that layer exists, the possibilities become concrete. The health worker who can trust the AI diagnostic because someone has already stress-tested it in her context. The teacher whose AI assistant has been evaluated for her language and her curriculum. The agricultural extension officer whose tool has been audited for the conditions of the farmers it serves. These are not passive recipients of AI. They are the reason the human layer has to be built.

Every major technology that scaled globally did so because a supporting human ecosystem grew around it. Cars did not become universal because everyone learned to build engines. They became universal because a coordinated ecosystem of certified mechanics, safety inspectors, insurers, standards bodies and testing organizations emerged to make them usable and trustworthy in daily life. AI is at that same inflection point now, yet the ecosystem it needs does not yet exist.

What is missing today is a structured secondary market of human expertise focused on AI performance, safety and accountability; one that does not compete with AI developers but makes their tools usable in real institutions and real communities. This market would include people and organizations who test AI systems in local conditions, evaluate performance and bias, integrate tools into legacy workflows, secure AI systems against misuse, and help institutions understand what these tools are doing and where they fail.

The job profiles are concrete: AI auditors and inspectors, assurance engineers, synthetic data developers, AI cybersecurity specialists, red-teamers, and insurers underwriting AI-enabled operations. These roles exist today only in fragments, mainly in high-income countries. What does not yet exist is a clear, scaled pathway for them to emerge globally, particularly in the developing and mid-sized markets where AI adoption is accelerating fastest, and the human layer is thinnest.

The AI supply chain already has a geography. Lower-value activities such as data labelling and annotation are concentrated in low- and middle-income countries, requiring intensive human labour but offering limited returns. Higher-value tasks – model design, deployment, governance – remain largely confined to high-income countries with specialized knowledge and infrastructure.

The opportunity is to move up that chain. Not through redistribution, but through local creation. The human layer that makes AI trustworthy is, by its nature, place-specific. It requires knowledge of local languages, laws, institutions and failure conditions, and that specificity is precisely what makes it impossible to simply import.

An AI auditor evaluating a court system in Port of Spain cannot be substituted by one based in San Francisco. A red-teamer stress-testing a public health AI in Nairobi needs to understand the context that shapes both the tool's outputs and the harms its errors would produce. The pathway upward runs through building that expertise domestically: through certifications, training pathways, and institutional anchors that turn local necessity into local capability.

Cybersecurity offers the clearest precedent. It began as a niche capability accessible mainly to wealthy governments and large corporations. It evolved into a globally distributed skills market worth hundreds of billions of dollars, and millions of jobs. Many countries that did not invent the internet or dominate software development became leaders in cybersecurity services, talent and regulation. The human layer for AI trust and safety is on the same trajectory. The window to shape it, to build the certifications, training pathways, institutional anchors, and professional standards, is open now.

The 2025 HDR identifies a set of emerging roles at the human-AI interface: explainers who evaluate AI outputs before they enter consequential decisions, trainers who customize models and set risk and performance thresholds for specific industries and contexts, and sustainers who ensure organizations adapt as AI systems evolve. In high-income settings, these roles are beginning to appear inside large organizations. In developing country contexts, they take on a different and more urgent form. Someone has to determine whether the AI system can be trusted in the first place. That is an auditor, an inspector, an assurance engineer, a civic technologist who builds the mechanisms that allow citizens to contest AI decisions affecting their lives.

For countries like Trinidad and Tobago, where UNDP recently conducted its AI Trust and Safety assessment, the entry point for jobs is concrete. Trinidad and Tobago brings high human capital, existing institutional depth and deep integration with global markets. Yet when AI tools were deployed across ministries and secondary schools at scale, the governance infrastructure to match them did not yet exist. The roles that emerged from that gap became immediately visible: auditors, assurance engineers, risk profilers, people who could build ministry-level safeguards and protect students from harms that no existing framework had anticipated. These are not drawn from a ready pipeline. They simply do not yet exist at the scale that adoption will demand.

The economic and the moral argument converge on the same point according to the UNDP Trust & Safety Re-imagination Programme. The largest opportunities from AI will come not from automation alone, but from performance-enhancing augmentation – from the human work that makes AI usable, trustworthy, contextually validated and effective in real systems. In developing countries, that augmentation has a specific shape: It is the human layer that allows AI to diffuse at all.

That layer has to be built deliberately. It will not emerge on its own, because the market signals are still weak, the professional standards do not yet exist, and the countries best positioned to build it are not the ones currently setting the agenda This is precisely why UNDP is part of the founding group of organizations building 100 AI Diffusion Pathways by 2030, because the pathway from AI's arrival to its responsible use runs directly through the human layer. And that layer is made of jobs that are waiting to be created.