AI Agent Payments: The Race To Give AI A Way To Pay

Ask a chatbot for the best coffee shop nearby and it will tell you. Then comes the wall. “Can you order a coffee for me? It can’t,” Atul Khekade, co-founder of XDC Network, said in an interview. His team ran that exact demo with Claude, Anthropic’s AI assistant, wired up to the blockchain’s new payments connector. This time the agent paid. “It basically does the entire settlement ordering system for you,” he said.

The demo is the centerpiece of XDC AI, an agentic payments framework the enterprise blockchain launched in late July at an event at XDC Tech’s New York office that drew more than 100 bankers, technology partners and investors, according to the company. It combines x402, a machine-payments standard that revives the web’s dormant “402 Payment Required” error code, with gasless settlement in Circle’s USDC stablecoin and a connector for the Model Context Protocol, the interface that lets AI assistants use outside tools.

“AI doesn’t have a transaction layer right now,” Khekade said. Banks and card networks are working on one, he argued, but “for them to come up with that infrastructure overnight is not going to be possible.”

The public record complicates that claim. Over the past year, while banks and retailers were still blocking AI agents from customer accounts, nearly every large payments company started building the plumbing to let them in. The live question is whose rails the agents will end up using.

‘In Pilot Production Already’

XDC’s pitch rests on its history. The company behind the network, founded as XinFin in 2017, launched its proof-of-stake mainnet in June 2019 and has spent most of its life on the unglamorous business of trade finance. Circle announced native USDC on the network in August 2025. Its validator roster includes Deutsche Telekom’s MMS subsidiary, which joined as a standby node operator in 2024, and NTT DOCOMO GLOBAL, added as an institutional masternode validator in July 2026. A joint venture with Japan’s SBI Holdings, set up in December 2023, runs its Asia-Pacific trade finance push.

Khekade says the node count is “almost 200 plus major institutions.” Public announcements tell a more modest story: coverage of the NTT DOCOMO addition put the announced institutional validator roster at more than 20, on a network capped at 108 active masternodes.

The AI framework, he insists, is past the experiment stage. “It’s not POC or something. It’s, it’s in pilot production already,” he said, adding that “insurance companies, banks, fintech providers, airline companies, a lot of them are coming to us now.” None of those customers were named. He expects a “proper Fortune 500 rollout in the next two to three months.”

There is a reason a blockchain wants this lane. On XDC, Khekade said, an address can carry a verified identity on top, “so you’re onboarded basically on the mainstream system with the same, you know, same standards as you do on the banking.” Legacy cores cannot keep up, in his telling: “your main kind of transaction systems are on the mainframe,” layered over for decades, too slow for machine-speed reconciliation. “I don’t think they have time for that.”

‘Only Those Two’

Asked about competition, Khekade kept the field small. “I only look up to someone like, say, ripple, Solana, only those two,” he said. “Maybe Ripple has a framework. I think we launched about the same time.”

Ripple launched first. Its XRPL AI Starter Kit, which bundles an MCP server, Claude skills for wallet and payment operations and the same x402 standard, shipped roughly six weeks before XDC’s New York event. On June 10, Mastercard introduced Agent Pay for Machines, extending its agentic payments program to high-frequency machine-to-machine transactions with more than 30 partners, including Coinbase, Stripe and Cloudflare. Visa’s Intelligent Commerce Connect, announced at its June payments forum, supports four different agent protocols through a single integration.

Big Tech moved even earlier. Google’s Agent Payments Protocol arrived in September 2025 with more than 60 partners, using cryptographically signed mandates to prove a human authorized an agent’s purchase. OpenAI and Stripe launched in-chat checkout the same autumn. And the two open standards at the heart of XDC’s own framework belong to other companies: x402 was open-sourced by Coinbase in 2025, and MCP is Anthropic’s, with payment connectors already shipped by PayPal and a crowd of startups. The moat, if there is one, is not the connector.

Khekade’s answer is that the market is big enough to make the race beside the point. “I wouldn’t call it a competition. I think it’s just the demand of the market are too high for anyone to absorb it,” he said.

‘Probably Not Gonna Happen’

The bull case for putting agents on crypto rails is that the alternative requires banks to serve customers who are not people. Yat Siu, executive chairman of Animoca Brands, said on the On The Margin podcast that “agents with wallets essentially become autonomous economic actors.” An agent needs money to act, and he does not expect a bank to open an account for a piece of software. “Probably not gonna happen, right? So how do they do that? They have a wallet.”

From there, Siu argued, the rest follows. “They can use a stablecoin, right?” “We already have agents that are trading on hyperliquid,” he said, referring to the crypto derivatives exchange. Stablecoins give agents a settlement asset that moves at machine speed, on rails that have already overtaken ACH in raw volume. Khekade makes a parallel argument about the AI industry itself: model companies are raising “hundreds of billions of funding and big valuations with no revenue model figured out” beyond charging for model usage, and a native transaction layer is a way for their agents to earn.

The harder problem is trust. “Our entire financial ecosystem was primarily human-centric,” Chandler Fang, chief executive of t54 Labs, said on the On The Margin podcast. His company builds verification for agent transactions, and he describes the merchant’s side of one: an agent shows up with $10 and an order, and “their first reaction is, Who are you? Who are you representing for?”

Varun Kabra, chief growth officer at Concordium, said on the On The Margin podcast that when an agent books a flight today, the airline has “no way to verify where a real accountable human is behind the transaction.” He still expects the shift to happen fast. “It is a matter of time,” he said, maybe “six to twelve months” before agent transactions overtake human ones, which makes accountability “the biggest problem I think the world needs to solve for.”

‘Trying To Be Aligned’

Regulators are writing that skepticism into rules. The UK’s Financial Conduct Authority put agentic payments on the use-case list for its AI Live Testing program in April 2026. Singapore’s central bank published proposed safeguards in July that would force a checkpoint between an AI agent’s decision and any payment that follows. “We are very much trying to be aligned and with compliance with them,” Khekade said of the regulators. Neither has named XDC as a participant in its program.

The International Monetary Fund’s April note on agentic payments raised a flatter objection: payment systems run on deterministic logic, and agentic AI is probabilistic. Gartner counts 17% of organizations with AI agents deployed and expects more than 40% of agentic AI projects to be canceled by the end of 2027. OpenAI’s in-chat checkout, the highest-profile agentic commerce launch of 2025, reportedly had around 30 Shopify merchants live by March 2026 before the company pulled back. Payments veterans have heard the rails-will-change-everything pitch before, and the banks being disrupted are building stablecoin infrastructure of their own.

None of it dents Khekade’s timeline. The market, he said, is “growing at the speed of light almost.” As for the giants he says cannot move fast enough: “The demands are too high for any of those to even remotely capture it.”