AI Agents Went Shopping in Abu Dhabi — and Paid in Stablecoins Without Touching a Wallet

An Unusual Morning at Hub71

On an otherwise ordinary day inside Abu Dhabi’s Hub71 technology ecosystem, the usual rhythm of coffee runs and retail browsing took a distinctly futuristic turn. The customers weren’t people. They were autonomous AI agents — software programs acting independently — and they were doing something that has long been considered the final frontier in machine automation: they were opening their wallets. As part of a live test orchestrated by the XDC Network, these agents spent credits on coffee and merchandise from real, operating merchants, and even purchased services from other AI agents. The pilot, reported by Khaleej Times and Zawya, marks one of the first practical demonstrations of machine-led commerce taking place in a sanctioned, regulatory-aware environment. It wasn’t a simulated sandbox exercise; actual merchants fulfilled orders, and stablecoin payments moved across blockchain rails in real time. The significance was not lost on observers. For years, AI agents have been able to search the web, compare products, recommend services, and even draft emails. But when it came to actually paying for something — completing a financial transaction — they hit a wall. That wall, it appears, is beginning to crumble in the capital of the United Arab Emirates.

A Payments Architecture Built for Machines

The technical mechanics of the experiment reveal why this matters. The AI agents transacted exclusively in USD Coin ($USDC), a leading fiat-backed stablecoin, rather than holding the native token of the network they operated on. Normally, on-chain transactions require the payer to hold some amount of the network’s native currency to cover gas fees. The XDC Network eliminated that friction through a relayer — a third-party component in blockchain infrastructure that sponsored the transaction costs on behalf of the agents. As the press release explained, XDC Network covered all network fees, meaning the agents could pay merchants and each other without ever needing to possess any token other than USDC. This subtle architectural choice is far more important than it might seem. It decouples the payment experience from the underlying blockchain mechanics, allowing AI systems to interact with financial infrastructure exactly the way a human would: by simply spending a stable balance, with the technical overhead handled invisibly in the background. The total value moved in the test was described as significant, but the more striking detail is the cost efficiency. Each payment cost a fraction of a cent, a level of micro-transaction viability that makes machine-to-machine payments commercially plausible. When software can transact for fractions of a cent, an entirely new economy opens up — one where billions of machine-driven micro-payments can flow without human accounting intervention at every step.

Closing the Last Mile of Autonomy

Jeremy Noori, Head of Structured Products at XDC Network, framed the achievement in clear terms. In the official release, Noori noted that AI agents had already matured to the point where they could search, compare, and recommend products or services with impressive accuracy. What they could not do, until now, was complete the transaction. “This event closes that gap,” he said. The distinction matters because it separates recommendation engines from economic actors. A chatbot that suggests a coffee shop influences human behavior, but an AI agent that can autonomously order and pay for that coffee is participating in the economy directly. The shift transforms AI from a tool into a counterparty. In this Abu Dhabi test, agents bought items typically associated with human workers — coffee, merchandise — and also purchased services offered by other agents, illustrating the emergence of a self-sustaining machine economy. This is the logic behind XDC Network’s positioning as an “agentic protocol,” a term used in the Zawya coverage of the event. The network is being architracted to handle the specific requirements of autonomous software: low fees, fast settlement, stable-value transactions, and the ability to operate without complex token management. By combining stablecoin payments with fee sponsorship, the test demonstrated a realistic blueprint for how AI agents can engage in commerce without human oversight at the point of purchase — raising both excitement and important questions about accountability.

Putting Guardrails on Machine Spending

Equally notable was the event’s attention to governance. The test may have been a technical demonstration, but it took place alongside a serious discussion about how to supervise machine-led payments. Khaleej Times reported that a panel featuring representatives from XDC Network and MESA delved into the role that banks, stablecoin issuers, and regulators must play in overseeing agent-to-agent transactions. The opening panel, aptly titled “Agents that pay,” included MESA’s Chairman and Founder Dr. Bhaskar Dasgupta, MESA Co-chair Ramy Soliman, and XDC Network executives, signaling that the business community intends to address oversight in parallel with innovation. The venue itself added another layer of significance. Hub71 operates within the Abu Dhabi Global Market (ADGM), an international financial center that has deliberately positioned itself as a regional hub for digital asset activity. ADGM has built a comprehensive framework for distributed ledger technology and stablecoin issuance, giving companies a clear regulatory pathway to experiment with new forms of value exchange. That context is crucial. Innovation in financial technology often runs ahead of regulation, creating a gray zone that hampers institutional adoption. By staging this demonstration inside ADGM’s jurisdiction, XDC Network and its partners are signaling that they want machine payments to develop within a compliant, supervised framework — not in a regulatory vacuum. The panels appear designed to bring banks and policymakers into the conversation early, rather than presenting them with a fait accompli later.

A Preview of the Agent Economy

The broader implications of this experiment extend well beyond a single coffee purchase in Abu Dhabi. The technology industry has spent the past two years racing to build increasingly autonomous AI agents, with major technology companies and startups alike pushing AI systems that can navigate software, manage workflows, and complete complex tasks with minimal human direction. But those agents have largely operated in a digital cage, able to affect screens and databases but disconnected from real-world economic consequences. The inability to pay has been a structural constraint on the entire agent economy. The Abu Dhabi test directly challenges that constraint. Combined with the explosive growth of the stablecoin market — which has become one of the quiet workhorses of the global financial system, settling trillions of dollars annually — the capability to enable autonomous machine payments suggests a future where agents negotiate prices, settle invoices, and reconcile accounts without any human involvement. Analysts have been predicting for years that machine-to-machine payments would eventually become a meaningful share of digital commerce, but concrete demonstrations have been rare. That is why this relatively modest test in Abu Dhabi is generating outsized attention. It is one thing to sketch a theoretical architecture for the machine economy; it is another to watch it operate in a live, regulatory-aware setting with real merchants on the other side of the transaction. The event offers a glimpse of a world where software has purchasing power — not as a metaphor, but as a practical feature of the economy. For enterprises, the implications are significant: supply chains could self-manage procurement, autonomous fleets could pay for their own charging and maintenance, and software systems could source and pay for the digital services they depend on to operate.

From Demonstration to Commercial Reality

It would be premature, however, to declare that the machine economy has arrived. The Abu Dhabi test was deliberately small in value and executed with credits rather than participants’ own funds. It is a demonstration of what is technically possible, not proof of commercial scale. Real-world deployment will require solving harder problems: ensuring that autonomous agents comply with anti-money-laundering rules, establishing legal accountability when a machine makes a flawed purchasing decision, and building identity frameworks that allow counterparties to trust software entities engaged in transactions. The regulatory discussion that accompanied the test suggests that XDC Network and MESA are acutely aware of these hurdles. The path from controlled pilot to mainstream adoption is long, but the direction of travel is increasingly clear. Stablecoin infrastructure has matured; fee sponsorship has solved the gas-fee problem that historically made micro-transactions uneconomical; and the AI industry’s relentless push toward autonomous agents shows no sign of slowing. What happened in Abu Dhabi was a small transaction with outsized symbolic weight. It demonstrated that the foundational components — stablecoin payments, sponsored network fees, and agent marketplaces — can be assembled into a working system in which software pays for goods and services with minimal human input. The machinery of the machine economy is being built, and it took its first paid coffee break in the capital of the UAE.

This article draws on reporting from Khaleej Times and Zawya regarding the XDC Network’s agentic commerce demonstration at Hub71 in Abu Dhabi.