Alchemy and Visa launch AgentCard: AI agents get their own payment tools
Alchemy, in collaboration with Visa, has introduced the AgentCard solution — a service that equips artificial intelligence with full-fledged financial tools. Now, AI agents can independently make online purchases on behalf of the user, without their direct involvement.
The integration with Visa's Intelligent Commerce infrastructure provides neural networks with a comprehensive set of data: a virtual Visa card, an email address, a phone number, and a cryptocurrency wallet. Developers only need to connect a single API, and access to paying for goods and services is configured in under a minute.
Agents based on OpenAI or Anthropic models can now book tickets, order groceries, renew subscriptions, and perform other routine tasks. The user does not need to intervene in the checkout process — the AI handles all interactions with the merchant.
The system provides flexible restrictions: spending limits, bans on store categories, and customizable budgets. By default, payments are processed via Visa tokens, allowing users to retain bank bonuses and credit lines. If the merchant accepts digital assets, the service automatically switches to the crypto wallet.
"Every technological shift creates new economic participants. AI agents are the next stage, and they need access to the global economy," noted Alchemy CEO Nikhil Viswanathan. Visa representatives emphasize that their infrastructure ensures the security and scalability of such transactions.
The AgentCard protocol independently selects the optimal payment method based on merchant support. This makes the process as seamless as possible for the end user.
Previously, MetaMask announced a wallet for the era of autonomous AI, and Coinbase launched a service that connects an AI agent to a user's account for trading and payments within set limits. Estonia also announced the creation of digital IDs for AI agents.
Expert opinion: This move by Alchemy and Visa marks a transition from the concept of AI assistants to full-fledged economic entities. If previously neural networks only helped in decision-making, they are now becoming independent participants in transactions. However, the key challenge remains security: how to ensure that the agent does not exceed its authority or become a victim of manipulation? The solution with limits and category restrictions is a good first step, but the industry will need to develop more reliable control mechanisms.