Shaping inclusive economic growth in 2026

About the Center

(Header) About the Center

The digital transformation of European micro-businesses

Focus Areas

(Header) Focus Areas

Scaling impact together

Events

(Header) Events

Building trust in the age of artificial intelligence

Insights

(Header) Insights

(Footer) About the Center

(Footer) Focus Areas

(Footer) Events

(Footer) Insights

PostContent

We stand at an inflection point. AI systems are moving from tools that assist to agents that act, making decisions, executing tasks, and operating at speeds and scales that outpace human oversight. This shift promises extraordinary gains in productivity and efficiency – but it also raises a fundamental question: what does this mean for inclusion? Ultimately, these agentic systems must work for the global majority, not just the privileged few. At the Center for Inclusive Growth, we believe that AI should expand opportunity rather than concentrate it, and vulnerable communities should gain power and agency – not simply access to new tools. Yet we are currently seeing a different story play out. While AI investment has surged past $1.5 trillion globally, according to research by the World Economic Forum and the Schwab Foundation, less than 1% of that investment is directly targeting social impact. Despite advancements in digital adoption globally, two-thirds of the population live in regions where digital infrastructure remains fragile, data systems are incomplete, and the skills needed to steer AI systems are scarce. Without deliberate intervention, agentic AI risks amplifying the very inequalities it has the potential to solve. If we’re not able to build the foundations today that mean agentic AI serves everyone, we’re in danger of heading in a direction that is less inclusive rather than more so. So what can we do to ensure that agentic AI works for the global majority? One key element is in creating adoption pathways (or enablers) from the ‘pre-agentic’ phase to the agentic phase. Doing so would integrate today’s AI tools as proving grounds for agentic AI to succeed, via five pillars we believe are crucial for agentic inclusivity.

Five pillars for inclusive agentic AI Agentic AI differs from earlier generations of technology in both capability and consequence. These systems can plan, prioritize, and execute complex tasks with minimal (or no) human involvement. An agent might negotiate contracts, optimize supply chains, allocate resources, or approve credit. And they can do it within seconds, based on patterns in data that no individual could parse alone. This autonomy introduces new risks. When decisions are made at machine speed, who bears responsibility when things go wrong? When algorithms prioritize efficiency over equity, who speaks for those left out? When systems are trained on incomplete or biased data, whose realities get erased? Our work at the Center for Inclusive Growth has identified five interconnected pillars essential to navigating these risks, with the aim of ensuring agentic AI can serve everybody.

InlineImage

1. Infrastructure for inclusionAgentic AI cannot deliver on its promise if the people it's meant to serve remain offline or digitally excluded – and today, that applies to 2.9 billion people. Infrastructure matters. This means ensuring small businesses can transact digitally, workers can access AI-enabled platforms, and communities can participate in the data economies being built around them.Mobile phone penetration is a critical part of this, with rising levels of mobile phone usage providing opportunities for AI solutions. In Vietnam, for instance, SoBanHang provides a business management mobile app currently serving 500,000 microbusinesses and processing $2.5 billion in transaction value annually. Critically, SoBanHang is now mining this data to profile merchants by skill level to ultimately design potential agents truly tailored to their needs. This demonstrates how today's efficiency tools can become tomorrow’s agentic infrastructure.

InlineQuote

“But we need to make sure everyone has access, so no one gets left behind as agentic AI becomes more mainstream. When we build with inclusion in mind, we can ensure that everyone will have the chance to benefit from future technological advances.” 2. High-quality, hyperlocal dataMany emerging markets are data-rich but information-poor. Enormous volumes of data are generated daily, yet the frameworks and governance needed to transform that data into insights serving local populations often don’t exist.For agentic systems to work equitably, they must be trained on data that reflects real conditions. For women with limited financial histories, informal workers without traditional employment records, or smallholder farmers facing climate volatility, hyperlocal data is vital. Through initiatives like the Artificial Intelligence to Accelerate Inclusion (AI2AI) Challenge, the Center is funding organizations pioneering this localization and helping them scale responsibly and effectively.

“It’s about moving beyond generic solutions so that every small business, worker, or farmer can shape and benefit from the next generation of technology on their own terms.”&nbsp;3. AI governance that keeps humans accountableAs AI systems gain autonomy, accountability becomes existential. Agentic AI makes decisions at superhuman speed and scale, often in ways that are opaque even to their creators. Without robust governance, these systems risk operating beyond the reach of the communities they affect most.Governance must ensure that AI acts within clearly defined guardrails, that decisions are traceable and auditable, and that humans remain unambiguously accountable for outcomes. This means transparency about how agents are trained, what data they use, and how they prioritize competing objectives. It means designing systems that can be questioned, redirected, or overridden when necessary.For small businesses and vulnerable workers, this translates into a non-negotiable principle: even when a machine makes the decision, a human must own the consequences. 'Humans in the loop' is not simply a technical safeguard, it’s the mechanism by which accountability is preserved.When an AI agent denies a loan, misallocates resources, or produces a harmful outcome, there must always be an identifiable person or institution answerable for that result. The governance frameworks we establish now will determine whether that chain of accountability holds as AI systems grow more autonomous and whether the communities most affected by those decisions have any meaningful recourse.

4. Pathways for adoption that ensure usage is as important as innovationEven the most sophisticated AI agent is useless if people don't trust it, or don't know how to steer it. Too often, the focus falls on building and deploying AI solutions, while the question of whether people actually use them – effectively and on their own terms –&nbsp;is treated as secondary. Ensuring genuine, sustained usage is just as important as the innovation itself.This is where the ‘pre-agentic’ phase becomes strategically important. Today's AI tools are proving grounds. They show small business owners that technology can simplify daily tasks, reduce administrative burdens, and free up time for higher-value work. They build the fluency and confidence needed to question AI recommendations, redirect agents when needed, and halt processes that don't serve their interests.Participation defines this pillar of ‘inclusion by design’. Vulnerable communities must be involved as co-designers and beneficiaries of AI solutions, not merely end users of systems designed elsewhere.

5. Skills pathways that democratize agencyEnsuring people use AI is one thing; ensuring they can shape it is another. None of the other pillars fully deliver unless the workers, entrepreneurs, students, and civil society organizations who live and work alongside agentic systems have the skills to engage with them critically and confidently.AI literacy must become as fundamental as financial literacy. But the goal goes beyond basic understanding. People need to know what AI can and cannot do, how to evaluate its recommendations, and when to trust or distrust its outputs. This way they become active agents in how these systems evolve, rather than passive recipients of decisions made for them.

Building the foundations nowThe age of agentic AI is unfolding in real time, in local shops and health clinics, in supply chains and credit applications. For the global majority, today's reality prioritizes efficiency and optimization. AI tools that help people work better, faster, and with less friction.Inclusion in the era of agentic AI will be measured by whether we use this pre-agentic phase wisely. That means acting concretely across all five pillars: building the digital infrastructure that connects the excluded; investing in hyperlocal data that reflects real lives; establishing governance frameworks that keep humans accountable; creating adoption pathways that prioritize usage alongside innovation; and developing skills that give people genuine agency over the systems shaping their futures.The foundations we lay today will decide whether agentic AI becomes a force for convergence or divergence, whether it expands opportunity for the global majority or leaves them further behind.At the&nbsp;Center&nbsp;for Inclusive Growth, we place inclusion by design at the heart of everything we do. And, now, with the rapid pace of innovation in the agentic space, each sector, developer and innovator must do so as well.

We stand at an inflection point. AI systems are moving from tools that assist to agents that act, making decisions, executing tasks, and operating at speeds and scales that outpace human oversight. This shift promises extraordinary gains in productivity and efficiency – but it also raises a fundamental question: what does this mean for inclusion?

Rethinking inclusion in the age of agentic AI