Inside the Push for Agent-Native Payments at Coinbase Business

Before joining Coinbase, Sid Coelho-Prabhu built Savvy, one of the first robo-advisors in the Indian market. The product was designed to provide automated investment recommendations, a model inspired by the emergence of Wealthfront and Betterment in the United States. While the software was modern, the financial plumbing underneath remained tethered to legacy infrastructure. "We actually had to have a fax machine to receive daily operational alerts, and get customer documents by mail to scan and send back," Sid told Narb during their discussion. "It was that broken."

Sid's Origin Story

Sid began his career as an investment banker on Wall Street, advising Fortune 100 companies on managing large balance sheets and complex financial operations. That experience informed his perspective on how tooling for large corporations could be adapted for individual investors. Savvy allowed customers to answer a few questions and connect a bank account to start investing, but the friction of the underlying payment rails was a constant obstacle. The turning point for his career trajectory came in the air, during a long flight from India to the US, when he read the Bitcoin whitepaper.

I was dealing with all these problems with how broken payment systems were for my startup. Bitcoin felt like the next wave of technology that was going to disrupt and reinvent how payments and money work.

He joined Coinbase roughly seven or eight years ago, initially focusing on the development and scaling of Coinbase Wallet, the company's self-custody product. After nearly a decade in that role, he moved to build Coinbase Business, which launched a little over a year ago.

Money Moving Like Data

The development of Coinbase Business followed the growth of USDC, the stablecoin launched by Circle in 2018. Sid describes a stablecoin as an internet-native version of the US dollar, which he calls the lingua franca of global commerce. His argument is that while using dollars through traditional rails is slow and expensive, putting them on a blockchain allows money to move the way data does, instantly and anywhere in the world.

On Coinbase Wallet, Sid saw hundreds of thousands of people worldwide already using stablecoins for transactions. The missing piece was a package for businesses, which have operational needs that differ from individuals. Businesses need to send invoices, run batch payouts to contractors, and accept payments on their own websites. Coinbase Business consolidates these functions into a custodial operating account for trading, payments, treasury, and USDC rewards.

The Three Percent Profit Drain

The shift toward stablecoin payments addresses specific mathematical burdens in the legacy financial system, starting with card fees. A business accepting credit card payments typically pays a fee of around 3%. Sid framed this cost against the reality of thin margins.

If your profit margin is 10% or 15%, that 3% fee takes up 20% of your profits just to accept money.

Beyond the fees, merchants face the risk of chargeback fraud. This is particularly relevant for digital products such as AI inference, where a customer can consume the resource and then claim a chargeback through their bank. Stablecoin transactions are final once the funds reach the merchant wallet, removing the possibility of such disputes.

The second major source of friction is international reach. The internet economy allows customers and contractors to be located anywhere, from Azerbaijan and Nigeria to South America and Southeast Asia. Traditional card payments from these regions are frequently declined due to false high-risk flags. Stablecoins allow businesses in supported markets to accept payments from anywhere without these borders.

A Handshake for Machine Commerce

AI agents are primed to become economic participants on the internet. Narb predicted that agentic purchases will eventually outpace human ones, and Sid noted that Coinbase has been building for this shift. Erik Reppel, an engineering leader at Coinbase, created the x402 protocol, which Sid describes as a simple handshake for machines to pay each other. The protocol is an open standard based on the HTTP 402 Payment Required response.

Autonomous agents are increasingly required to perform tasks such as buying domains, sub-agent services, or paywalled research. The easiest payment instrument for a piece of software is a programmable crypto wallet. What was missing was the ability for businesses to accept these payments, which is why checkouts in Coinbase Business now support x402. In the demo store, CoinBiz Merch, Sid showed the difference between a human checkout, which involves a QR code and a mobile wallet tap, and an agent checkout.

Sid demonstrated agentic commerce in action when he opened Cursor AI and asked the agent to buy a hat from the store URL. The agent crawled the page, listed the available merchandise, and asked for confirmation. Once prompted to buy the hat, the agent inspected hidden x402 metadata in the HTML, extracted the requirement for 0.01 USDC on the Base network, and checked its own wallet balance. It then created its own wallet, funded it, and posted the transaction onchain. The entire flow ran without a visual interface, returning an order record and a transaction hash directly to the terminal.

Relentless Focus on Customer Problems

For developers building in the space, Sid pointed to the Coinbase Developer Platform (CDP). A CDP account linked to a Coinbase Business account provides the checkout APIs, wallet infrastructure, and onramps needed to build these agentic payment flows. Sid's advice for founders was one grounded in user behavior.

Especially in crypto, there's a lot of tech hype versus what customers actually need. Don't focus on your technology; focus relentlessly on solving real customer problems.

The hat purchase by the Cursor agent served as a proof of concept for this frictionless environment. The hat was never physically shipped, but the wallet the agent created and the onchain transaction it settled were real. For Sid, the success of the demo showed that the same checkout system could serve both a human with a smartphone and a software agent running inside a code editor.