Agentcard gives AI agents their own credit cards

An AI agent can now order you DoorDash, renew your Vercel Pro plan, buy Kendrick Lamar tickets on Ticketmaster, and send your mom $50 on Cash App — all without a human touching a checkout form. The product enabling this is Agentcard, a virtual-card service that gives AI agents their own prepaid Visa cards with scoped spending limits.

Agentcard launched in private beta in March 2026 and has been quietly accumulating users who are giving their agents financial autonomy. The company describes itself as a “financial infrastructure layer for AI agents.” It issues one-time or persistent virtual cards that agents can use at any online merchant that accepts Visa. The cards come with programmable budgets, per-transaction caps, and real-time spend tracking. Setup takes “1 min,” according to the product page, and involves a few API calls or a natural-language command.

The product is simple. The implications are not.

The agent becomes an economic actor

For the past two years, AI agents have been trapped in a read-only paradigm. They can browse the web, summarize documents, write code, and generate images. But when the task requires spending money — buying API credits, provisioning cloud infrastructure, ordering supplies — the agent hits a wall. It has to hand off to a human, who manually fills out a payment form. That handoff breaks the autonomy loop.

Agentcard closes that loop. An agent that can pay its own bills is no longer a tool. It is a counterparty. It can enter into transactions, manage a budget, and reorder inventory without supervision. The company’s website lists over 20 supported merchants including DoorDash, OpenAI, Airbnb, Amazon, Stripe, Uber, Instacart, and Shopify. The agent does not need to navigate a checkout form. It sends a payment instruction, and Agentcard processes it.

This is the infrastructure play that several users on X have called out. Uttam, an early adopter, wrote in March: “THIS IS THE INFRASTRUCTURE PLAY NOBODY IS TALKING ABOUT. YOUR AI AGENT GETS ITS OWN VISA CARD. IT BUYS WHAT IT NEEDS. YOU SET THE LIMITS. NO BANK MEETINGS. NO PAPERWORK.”

The security model is the interesting part

Giving an AI agent a credit card sounds reckless. Agentcard addresses this with a layered security model. Each card has scoped spending limits — a monthly cap, a per-transaction cap, and merchant restrictions. The agent can only spend what the human authorizes. The card is prepaid, not a line of credit, so the blast radius is bounded by the balance loaded onto it.

Bill, another early user, described running three agents with separate budgets: “the one handling ad spend literally cannot touch the API billing card. this is the kind of guardrail stuff I’ve been waiting for.”

The product also claims to reduce “token waste” — the computational cost of an agent navigating a checkout form, parsing error messages, and retrying failed payment steps. By handling payments through a direct API, Agentcard eliminates the need for the agent to interact with a human-designed checkout UI at all.

What this means for the AI economy

Agentcard is a small company with a simple product. But it signals a structural shift in how money moves through the AI stack. Today, almost all AI spending flows through human-initiated transactions: a developer signs up for OpenAI, a company pays its AWS bill, a consumer subscribes to Claude Pro. The agent is the user of the service, but the human is the payer.

In a world where agents have their own payment instruments, the unit of economic activity changes. An agent can provision its own compute, buy its own API credits, and manage its own subscriptions. This has consequences for SaaS pricing, for cloud billing, for fraud detection, and for financial regulation.

Stripe and other payment processors are already building for this. Stripe issued a report in 2025 on “agent commerce” predicting that autonomous payments would become a major category. Agentcard is the first product to deliver on that prediction with a dedicated card product rather than a general-purpose API.

The product is not without limits. It currently works only with supported merchants — the agent cannot spend at an arbitrary online store. The pricing model is pay-as-you-go: “buy a card, top up whenever you need,” with no subscriptions or hidden fees. The company has not disclosed transaction volume, user count, or whether it has raised venture funding.

The open question: fraud and liability

The biggest unknown is fraud. If an agent is compromised and spends its entire balance on something the human did not authorize, who is liable? Agentcard’s terms of service are not public, but the product’s security model relies on the human setting appropriate limits and monitoring spending. In practice, that means the human is still the backstop.

This is fine for low-stakes use cases — ordering DoorDash, buying API credits, renewing a subscription. It is less fine for high-value transactions. An agent with a $10,000 limit that gets prompt-injected into buying gift cards is a real scenario. Agentcard’s scoped spending model limits the damage, but it does not prevent it.

The company has not published a security audit or a red-team evaluation of its card-issuance flow. For a product that touches payment infrastructure, that is a gap. Early adopters appear comfortable with the risk. Wider enterprise adoption will require more.

What to watch

Agentcard is not the only company building in this space. Several startups are working on “agent wallets” — crypto-native or fiat-based — that give agents financial agency. But Agentcard is the first to ship a Visa-integrated product that works with existing merchants. That matters.

The product is in private beta as of June 2026. The company plans to add on-chain payments, DeFi integrations, and wallet-native spending in future releases. For now, it is a simple API that turns an agent from a read-only assistant into a paying customer. That is a bigger change than it sounds.