Ajay Banga does not speak like a development banker.
He speaks like a businessman who has stared at a demographic cliff and decided it is actually a launching pad.
His framing, repeated in every forum he enters, is stark: 1.2 billion young people will enter the workforce over the next decade and a half, against a projected supply of 420 million jobs.
As he told the Atlantic Council on the eve of the 2026 Spring Meetings, the fundamental question is whether those young people can find dignity and hope where they are.
The World Bank Group has reorganized its entire institutional identity around that sentence.
Under Banga, job creation is no longer a sectoral outcome.
It is the primary measure of development success, the lens through which every lending package, every project approval, and every scorecard item is evaluated.
Three pillars, five sectors, and one silent assumption
Banga’s jobs doctrine rests on three pillars: infrastructure, human capital, and catalytic finance designed to unlock private sector participation.
These pillars are meant to drive job creation in five sectors that do not rely on relocating jobs from the developed world: infrastructure and energy, agribusiness, primary healthcare, value-added manufacturing, and tourism.
The Bank has announced specific targets: 300 million Africans gaining access to productive electricity by 2030, nine billion dollars per year flowing into agribusiness, and 1.5 billion people gaining access to better primary healthcare.
Sub-Saharan Africa, which currently imports more than 70 percent of all pharmaceuticals it consumes, is now central to the jobs agenda.
What is not stated publicly, but felt acutely inside the Bank’s operational machinery, is the silent assumption on which this entire architecture depends.
The three pillars are a supply-side story: mobilizing capital, reforming policy environments, building infrastructure.
None of that translates into jobs unless the human beings meant to fill those roles can perform them, adapt to change, and keep delivering results once the project closes and the consultants fly home.
That last mile of human activation is not in the strategy documents.
It is assumed to follow automatically from investment, and that assumption is where strategies go to die.
AI is already in the room, and it is already a problem
Artificial intelligence is not an emerging concern for some future workforce.
It is an operational reality for the workforce Banga’s strategy must activate right now.
Researchers who study the builders of these systems describe what they call the San Francisco Consensus: a convergence of belief that within three to six years, AI will fundamentally transform every aspect of human activity, including work, driven by a non-human intelligence with better reasoning skills than humans across virtually every domain.
Banga acknowledged this at the Atlantic Council, noting that his 780 million job gap calculation is pre-AI, without elaborating.
In high-income countries, AI primarily displaces knowledge workers in existing formal employment.
In lower-income countries, AI is arriving simultaneously with, and in some cases faster than, the formal employment structures it might otherwise disrupt.
For a young person in Lagos, Dakar, or Kinshasa entering a labor market that cannot absorb her, the question is whether AI will serve as a tool that expands her capabilities or arrive as a system controlled by distant institutions, concentrating advantage where it already exists.
Almost none of AI use is spoken about openly, because institutional culture treats AI use as inauthenticity: workers who disclose AI assistance risk having their work dismissed, while those who conceal it carry the ethical burden alone.
Researchers call this the humanitarian AI paradox, in which individual adoption races ahead of institutional readiness and most usage happens as shadow activity outside any governance or support.
The performance-learning gap that can hollow out an investment
Multiple independent studies from the OECD’s Digital Education Outlook 2026 conference confirm that generative AI can dramatically improve task performance while simultaneously undermining learning.
In one United States study, students who used large language models wrote better essays, but 80 percent could not remember what they had written about afterwards.
In a Turkish study, students using AI for mathematics did better on exercises but worse when their underlying mathematical thinking was assessed directly.
The OECD researchers named the mechanism “metacognitive laziness”: when learners delegate cognitive work to an AI tool, productive friction disappears, and performance metrics improve while actual capability erodes.
For the Bank’s jobs agenda, this is not an abstract educational concern.
If workers operating new infrastructure, digital health systems, or pharmaceutical manufacturing lines are using AI in ways that produce polished outputs while eroding their own understanding, the return on capital is built on foundations that will not hold.
In crisis contexts, that erosion undermines the ability of local teams to detect when guidance is wrong, challenge flawed assumptions from distant headquarters, and adapt when something goes wrong at three in the morning in a fragile district health facility.
The agentic turn and what it demands of human systems
Unlike previous waves of automation, agentic AI does not merely respond to queries.
It pursues goals expressed in natural language, breaking them into sub-tasks, retrieving context from past interactions, and executing each step in sequence.
When a colleague writes, “Please prepare a follow-up to all partners from yesterday’s briefing, attach the slide deck, and propose two dates for a technical consultation,” the agent retrieves the participant list, identifies the right slide deck, checks scheduling constraints, and drafts a message calibrated to the relationship.
The workforce implications for the Bank’s target countries remain unresolved.
AI systems are trained overwhelmingly on data from high-income countries, in English and Chinese, and OECD researchers show that AI interventions produce small or negative learning effects in other languages.
Workers in francophone West Africa, in DRC, in rural Indonesia face a structural disadvantage in the AI transition that mirrors exactly the disadvantages the Bank’s jobs agenda is designed to address.
Who controls the natural language interface, and in whose language it works, determines who benefits from the jobs agenda.
What actually works at the last mile
The Bank’s implementation record contains a lesson its own economists know well: capital investment without human activation does not produce outcomes.
A district immunization officer who does not understand why the new cold chain works differently will not use it reliably.
A pharmacist with no peers to consult about locally manufactured medicines will default to the imported product she already trusts.
The evidence for what changes this is specific.
In Nigeria, 4,300 health workers were mobilized into a functioning peer learning network within two weeks.
Within four weeks, they had collectively produced more than 400 root cause analyses of immunization challenges.
Within six weeks, they had discovered that their assumed root causes were wrong, shifted their activities, and begun finding children who had escaped every previous vaccination effort.
More than half of those workers continued to collaborate and report results after the funding closed, at 90 percent lower unit cost than conventional technical assistance and seven times faster implementation than groups supported through technical assistance or training.
This was a six-week deployment of The Geneva Learning Foundation’s peer learning-to-action model, in partnership with UNICEF and NPHCDA, the Nigerian Ministry of Health.
What this model does, which conventional training cannot, is treat the existing workforce as a knowledge system rather than a deficit to be corrected.
Knowledge flows horizontally among peers in local languages, driven by the intrinsic motivation of people solving problems they face every day.
When AI enters this system, it handles what peers have neither the time nor bandwidth for at scale: matching practitioners with complementary experience, surfacing patterns across thousands of generated plans, highlighting what is working in one district that another has not tried.
The governing principle is that AI should augment human connection, not replace it.
The scorecard question Banga has not yet answered
Banga told the Atlantic Council that the Bank is moving from inputs, how many projects and how many dollars, to outcomes: jobs, growth, and development-oriented results.
He has reduced the Bank’s performance scorecard from 155 items to 22, and he is right to make that shift.
The concrete question his team now needs to answer is what sits between the capital investment and the jobs outcome.
The evidence points to a functioning human learning system, operating at scale and low cost in local languages, with AI embedded as a co-worker within structures of human accountability and peer governance rather than deployed as a black box from above.
Infrastructure without the human systems to operate and adapt it is a liability.
Building those human systems means building something that can absorb, govern, and amplify AI rather than be eroded by it.