AI agents increasingly need to make purchases, but today’s payment systems assume a human is involved. The Linux Foundation’s x402 initiative, backed by over 40 organizations including Visa, Mastercard, and Stripe, aims to create internet-native payments for software. Casper Network has upgraded its csprUSD stablecoin for x402 micropayments. Whether blockchain or traditional rails win, the infrastructure for machine-to-machine commerce is being built now.
For the past two years, the AI industry has focused on making models smarter. Every new release promises better reasoning, longer context windows, or more capable autonomous agents. Increasingly, people assume that software won’t just answer questions; it will also complete tasks for us. This raises a concern that has received far less attention: what happens when those agents need to make a purchase?
A travel assistant might need to pay for access to a premium flight database. A coding agent could rent computing power for a few minutes to run a complex workload. A research assistant may need to unlock a paywalled paper or buy access to a proprietary API. None of these transactions require a human decision at the moment. They just need to happen quickly enough for the software to keep working.
Today’s payment systems were not designed with this in mind.
Most online commerce still assumes that a person is involved somewhere in the process. Credit cards need names, billing addresses, and authentication. Subscription services expect someone to consciously sign up and then decide whether to renew each month. Even digital wallets generally require a human to approve a transaction before money moves. These assumptions create issues when the customer is software making thousands of low-value purchasing decisions every hour.
The economics are tricky too. Charging $20 a month for API access makes sense when the customer is a business or an individual. It makes much less sense when an AI agent only needs a single request that costs a small fraction of a cent. Charging by the request seems like an obvious solution, but existing payment systems were never built for transactions that small, frequent, or automated. This challenge is beginning to draw serious attention outside the crypto industry.
Earlier this month, the Linux Foundation announced the launch of the x402 Foundation, an initiative supported by over 40 organizations, including Visa, Mastercard, American Express, Google, AWS, Stripe, Coinbase, Circle, and Cloudflare. The goal is to create a common standard for internet-native payments that software can initiate automatically, using the long-unused HTTP 402 “Payment Required” status code as the basis for machine-to-machine commerce.
The list of participants is notable for bringing together companies that rarely come together on emerging technologies. Payment networks, cloud providers, blockchain firms, and internet infrastructure companies are all recognizing the same problem. If AI agents become important players in the digital economy, they need a way to exchange value as naturally as they exchange information.
For blockchain networks, this presents a different chance than what they sought in the last decade. Much of crypto’s early ambition focused on replacing current financial systems or convincing consumers to pay with digital assets. Adoption has been slower than many anticipated, mainly because traditional payment systems already work reasonably well for most people. Software has different priorities.
An AI agent does not care whether a payment goes through a familiar banking interface or a blockchain network. It cares whether the payment settles reliably, whether the cost is predictable, and whether the fee is low enough that it does not outweigh the value of the transaction itself.
This shift helps explain why some blockchain projects are beginning to focus on AI infrastructure rather than consumer payments. Casper Network is one of them. Today, the network announced an upgraded version of its dollar-backed stablecoin, csprUSD, which is meant to serve as the default settlement asset for AI agents using x402 payment flows on Casper.
According to the announcement, the upgrade makes it compatible with Casper 2.0 and CEP-3009, the network’s standard for authorized transfers. This allows payment authorizations to be signed off-chain and settled on-chain as part of a standard HTTP request. The network has also reduced minimum transaction costs to make low-value machine payments economically viable, while integrating its ProofLayer infrastructure to provide proof of collateral and redemption activity.
On its own, another stablecoin launch would likely struggle to stand out in a market that already has hundreds of them. However, viewed through the lens of AI, this announcement reflects a broader shift happening across the industry. Stablecoins are increasingly being discussed not just as trading tools or payment options for people, but as operating infrastructure for software. If AI agents become regular users of online services, they will need access to stable digital assets that can move automatically, settle quickly, and integrate directly into internet protocols.
Whether blockchain ultimately becomes the leading foundation for those payments is still uncertain. The involvement of companies like Visa, Mastercard, and Stripe suggests that traditional financial systems will play a significant role alongside crypto-native systems. The eventual framework will likely combine elements from both, especially if developers prioritize compatibility over ideological preferences. The underlying trend itself is becoming harder to ignore.
The conversation around AI has mainly focused on intelligence, reasoning, and productivity. Yet these abilities only become commercially useful when software can engage in the economy without constant human oversight. Buying data, paying for computers, accessing digital services, and compensating other software will likely become routine for autonomous systems.
If that happens, the challenge will not only be building smarter AI. It will be creating an internet where software can transact as easily as it communicates. This may become one of the more important infrastructure debates in the next phase of AI, even if it attracts less attention than the models themselves.
Get the most important tech news in your inbox each week.
TNW newsroom and editorial staff were not involved in the creation of this content
