Since OpenClaw burst onto the scene as Clawdbot last November, individuals and businesses have embraced artificial intelligence agents to write code, send emails, run a shop, and more. AI agents are forecast to become ubiquitous in the coming years, raising concerns about agentic inequality, and its economic consequences for companies, countries, and people.
AI agents are built on top of large language models, and can reason and take actions to complete tasks on behalf of users. They have been touted as a way to do repetitive and mundane tasks to free up workers’ time for higher-value activities. Many agents still fail at the most basic tasks, and some perform unauthorized actions, yet big tech firms including Google, Amazon, Anthropic, and Perplexity are launching agents that can do increasingly complex tasks autonomously.
Agentic inequality can harden into systems of dominance.”Nick Srnicek, senior lecturer in digital economy at King’s College London
As AI agents become more integrated into the economy, companies and entities that deploy them will benefit disproportionately compared to those that cannot, Nick Srnicek, a senior lecturer in digital economy at King’s College London, told Rest of World.
“We will see new inequalities of access, scale, quality and trust: divides between those who have agents and those who don’t; those who have good agents and those who have bad agents; those who have many agents and those who have few agents; and those who can trust their agents and those who cannot,” he said.
Having access to agents that outpace others means “the outcomes of negotiations and transactions will be structurally biased towards those with greater access,” Srnicek said. “Agentic inequality can harden into systems of dominance.”
AI-powered agents and robots could generate about $2.9 trillion in economic value per year in the U.S. by 2030, McKinsey said in a report last year: “Work in the future will be a partnership between people, agents, and robots — all powered by AI.”
The compounding benefits of early access
Singapore and China have introduced frameworks to regulate AI agents, with a focus on safety and accountability. Local governments in China are boosting so-called one-person companies, or startups with a single founder using AI agents, as part of a broader agent-powered industrial policy. In India, founders are embracing agents to cut costs and scale quickly.
Raman Choudhary began using a Claude Code agent at his startup DentNode, a platform for dental clinics and labs, in Bengaluru last year. The agent reviews code, conducts market research, drafts publicity material, builds financial models, and helps him prepare for partner calls, said Choudhary, a former software engineer.
Those who figure this out early are going to build disproportionately large companies.”Raman Choudhary, startup founder
“Without it, I’d have needed at least one additional engineer, a part-time researcher, and a content or marketing hand,” he said. “In salary terms, that’s 1.5 million to 2.5 million Indian rupees [$15,700–$26,000] minimum per year, plus the overhead of managing those people.” His agentic workflow costs just a few hundred dollars a month, he said.
While he has access to the same Claude model as his counterparts in the U.S., it isn’t built for Indian workflows, including regional payments, tax systems, and local languages. Yet “the leverage here is higher, not lower,” Choudhary said. “Those who figure this out early are going to build disproportionately large companies.”
Adoption of generative AI is growing worldwide, but there is a widening gap between wealthy nations and poorer nations. With agents, there is a risk of “sharper divides, because 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,” he said. These inequalities will appear between countries, but also within wealthy countries, between firms, and between individuals.
A well-resourced firm can integrate agents into its proprietary data, software systems, procurement processes, customer operations, and decision-making workflows. Similarly, a wealthy person can use premium agents to navigate legal systems, improve financial decisions, negotiate contracts, or access public services, Sharp said.
“Better agents may help already-advantaged people and organizations move faster, bargain better, avoid costly mistakes, and accumulate further advantage,” he said. “The risk is not simply that some people use AI and others do not; it is that small differences in agent quality and integration become large differences in outcomes over time.”
Participation as first-class citizens
So-called citizen agents that can help people claim benefits and interact with the state can help bridge the gap somewhat.The Indian government aims to provide personal AI agents to some 50 million Hindu pilgrims at the months-long Kumbh Mela festival next year, and eventually deploy agents to each of its 1.4 billion citizens.
The safeguards around consent, purpose limitation, auditability, and political independence would need to be real.”Matthew Sharp, research affiliate at the Oxford Martin AI Governance Initiative
The Kumbh Doot — doot means messenger in Hindi — agentic AI framework will be a voice-first agent that can operate in more than 20 Indian languages, and coordinate with civic, transport, health, and commercial AI agents. Digi Doot will interact with public and private service providers, and is “designed for inclusion, optimized for users who face the highest friction: rural populations, the elderly, migrants, multilingual users, and those with low literacy,” according to a white paper.
Both systems will tap the country’s digital public infrastructure, including the Aadhaar citizen ID, the UPI payment gateway, the DigiLocker verification system, and the ONDC digital commerce platform.
“It’s about decentralization and democratization of AI,” Ramesh Raskar, an associate professor at the Massachusetts Institute of Technology who leads Project Nanda, a foundational infrastructure for AI agents, told Rest of World.
“Right now, we have no agency, no control over how AI is delivered to us,” said Raskar, who is involved in India’s programs. “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.”
Yet citizen agents raise concerns around security, privacy, and surveillance. “The same infrastructure can become a surveillance layer if the data flows, defaults, and oversight are wrong,” Sharp said. “The safeguards around consent, purpose limitation, auditability, and political independence would need to be real, not merely architectural.”
Governments — and companies such as Anthropic and OpenAl — can also cancel access anytime, leaving agent-dependent nations and citizens vulnerable.
The Kumbh Doot framework is “committed to privacy, dignity, and inclusion,” according to a white paper. “No pilgrim is surveilled, profiled, or tracked beyond what they consent to — and they can revoke consent anytime.” The Digi Doot will ensure “end-to-end privacy with agent actions that are attributable, bounded, and safe,” the proposal says.
Unlike Choudhary, most Indians do not need a sophisticated research agent. They only need a “reliable, safe, low-cost agent that can handle high-friction tasks,” such as filling a form or navigating benefits, Sharp said.
For individuals, small businesses, and nations, this is still a challenge.
“Without deliberate design, premium agents will help already-advantaged users move faster, while everyone else gets weaker, riskier or less-integrated systems,” he said.