Everyone in the agent ecosystem talks about democratization. OpenClaw is open-source. Claude Code runs on a laptop. Google just announced Antigravity 2.0 at I/O to let anyone build agents through a managed API. The tools are technically available to anyone with an internet connection. A Rest of World investigation published today argues that framing is dangerously incomplete.

The piece introduces a term worth paying attention to: agentic inequality. Not inequality of model access, which is the problem most discussions fixate on. Inequality of agent quality, integration depth, and operational reliability. The gap between having a chatbot and having an autonomous system that negotiates contracts, manages procurement, and makes financial decisions on your behalf.

The Scaffolding Problem

“Access to a base model is not the same as access to a reliable agent,” Matthew Sharp, a research affiliate at the Oxford Martin AI Governance Initiative, told Rest of World. “Every layer above the model, including scaffolding, tool integration, security, workflow design, and supervision, reintroduces skill and capital barriers.”

This is the core insight. A startup founder in Bengaluru named Raman Choudhary told Rest of World he replaced an engineer, a researcher, and a content specialist with a Claude Code agent for a few hundred dollars a month, saving 1.5 to 2.5 million Indian rupees ($15,700 to $26,000) per year in salaries. But Choudhary is a former software engineer who can configure and supervise these tools. He acknowledged the model “isn’t built for Indian workflows, including regional payments, tax systems, and local languages.”

The numbers back up the structural nature of this gap. World Bank data published in May 2026 found that ChatGPT visits per internet user in high-income countries are roughly 50 times those in low-income countries. Less than 5% of the population in low-income countries had basic digital skills in 2023. High-income countries hold 77% of global data center capacity. Low-income countries hold less than 0.1%.

The Investment Funnel

The capital concentration makes the access gap self-reinforcing. According to Rest of World’s analysis of the AI funding landscape, U.S. AI firms attracted 75% of all AI investment in 2025, roughly $194 billion, citing OECD data. The top 10 global AI investors led $96 billion in funding rounds for U.S. companies last year, compared to $1.9 billion across all other countries combined.

That capital buys compute, talent, and the engineering layers that turn foundation models into production-grade agent systems. McKinsey’s Global Institute projects AI agents and robots could generate $2.9 trillion in annual U.S. economic value by 2030. Note the geography qualifier: U.S. economic value. The report’s own framing acknowledges that this automation dividend accrues where the infrastructure exists to capture it.

India’s Billion-Agent Bet

India is attempting the most ambitious countermeasure. The government plans to deploy personal AI agents to 50 million Hindu pilgrims at the Kumbh Mela festival next year, and eventually to all 1.4 billion citizens. The Kumbh Doot framework is a voice-first agent operating in 20+ Indian languages, coordinating with civic, transport, health, and commercial systems.

Ramesh Raskar, an MIT associate professor involved in the initiative, told Rest of World: “When every one of us has our own AI agent that can talk to each other, transact with each other, and create economic opportunities for us, then we can participate as first-class citizens in the agentic economy.”

The catch is the same one that haunts every government-led tech deployment at scale. Sharp warned that “the same infrastructure can become a surveillance layer if the data flows, defaults, and oversight are wrong.” The Kumbh Doot system plugs into Aadhaar (India’s biometric ID system), the UPI payment gateway, and DigiLocker, a government verification system. The white paper promises no pilgrim will be “surveilled, profiled, or tracked beyond what they consent to.” Whether that holds at scale with 50 million users is an open question.

More fundamentally, governments and companies can revoke access at any time. Rest of World notes this dependency risk explicitly: agent-dependent nations become vulnerable to the geopolitical decisions of a handful of American and Chinese firms.

The Compounding Clock

Nick Srnicek, a senior lecturer in digital economy at King’s College London, framed the trajectory in structural terms: “We will see new inequalities of access, scale, quality and trust,” he told Rest of World. “Agentic inequality can harden into systems of dominance.”

The compounding dynamic is what makes this different from previous technology gaps. A well-resourced firm integrates agents into proprietary data, procurement, and customer operations. Those agents improve with use. The firm moves faster, bargains better, avoids costly mistakes. That advantage accumulates. A startup without the engineering team to build reliable agent scaffolding falls further behind with every quarter, not because the model is unavailable, but because the operational layer that makes it useful requires capital and expertise that concentrate in the same places they always have.

The Platform Question

For anyone building in the agent ecosystem, this raises a practical question: who are you building for? OpenClaw’s open-source model makes agents technically accessible to anyone who can run a server. Google’s Antigravity lowers the bar further with managed cloud infrastructure. But neither addresses the scaffolding gap: the tool integrations, workflow design, local-language support, and supervision layers that Sharp identified as the real barriers.

The $2.9 trillion McKinsey figure assumes organizations “prepare their people and redesign workflows.” Most of the world’s organizations cannot do that. Not because they lack ambition, but because the capital, talent, and infrastructure required to turn a foundation model into a production agent system remain concentrated in a small number of countries and companies. The agentic economy is arriving. The question is how many people actually get to participate in it.