Tether-backed Oobit has launched Agent Cards, a Visa-supported virtual card product that lets autonomous AI agents spend USDT directly from a company's stablecoin treasury with no fiat conversion, no human approval at point of transaction, and spend controls enforced at the transaction layer, arriving the same week MoonPay launched a Mastercard equivalent, suggesting the machine-to-machine payments category moved from concept to infrastructure in a single week.
The product is specific enough to evaluate precisely. Each AI agent is issued a single dedicated virtual card, creating a clean identity and audit trail per deployment. Finance teams configure per-agent spend limits, merchant category restrictions, per-transaction caps, and overall budget ceilings before any agent touches real money. Those controls are enforced server-side at the transaction layer with no override path for the agent itself. The cards run on Visa rails covering 150 million merchants across more than 100 countries, funded directly from a company's Oobit USDT treasury. There is no fiat on-ramp, no foreign exchange conversion, and no banking delay between a workflow trigger and payment execution. Oobit says the cards are compatible with agent frameworks from OpenAI, Claude, AutoGen, and LangChain, and businesses must pass a know-your-business compliance check before onboarding. Access is currently limited to a founding group of companies, with expansion to additional KYB-verified businesses through the end of June 2026.
The use cases Oobit describes are deliberately mundane, which is the point. An agent renewing a SaaS subscription, topping up an advertising budget, or spinning up cloud infrastructure at 3am because a workflow triggered it is not a dramatic vision. It is a description of operational tasks that already happen in every company running automated workflows, but which currently require either pre-authorised credentials embedded in scripts or human approval queues that break the automation. Agent Cards propose a third option: a controlled financial identity for software, with the same per-category and per-merchant restrictions that corporate procurement teams already apply to human employee cards. That framing, of an agent as a controlled economic actor with its own spend record rather than a process with embedded credentials, is the conceptual shift the product is introducing. The infrastructure behind it is not novel. Virtual corporate cards with spend controls have existed in B2B fintech for years, through Brex, Ramp, Pleo, and similar providers. What Oobit has done is design the product identity around the agent rather than the human employee, and fund it through a stablecoin treasury rather than a bank account.
The Tether backing matters more than it would for a typical fintech launch. Tether is the largest stablecoin issuer with over $140 billion in circulation, and its relationship with Oobit means Agent Cards are funded from Tether's treasury infrastructure directly. That removes the counterparty and conversion risk that would otherwise sit between a company's USDT holdings and the Visa settlement network. It also signals that Tether is actively building infrastructure around USDT's utility as a machine-to-machine payment primitive rather than simply issuing a dollar-pegged token and waiting for adoption. From Tether's perspective, an AI agent economy that runs its operational payments through USDT is a durable use case that sits entirely outside the retail and remittance markets where most stablecoin adoption has historically been concentrated.
The regulatory picture is the dimension that any serious enterprise evaluation has to address first. AI agents initiating real-world spending through consumer-grade card networks creates liability questions that existing compliance frameworks have not resolved. Who is the legal party responsible for an unauthorised or erroneous transaction initiated by an autonomous agent? The business that configured the card and passed KYB onboarding is the obvious answer, but the enforcement mechanism for that responsibility when the agent acts outside its intended scope, or when the configured controls fail to catch an edge case, is not established in most jurisdictions. Oobit's architecture mitigates some of this risk through the transaction-layer enforcement of spend controls, but it does not eliminate the fundamental question of what happens when an agent makes a mistake with real money at scale. That question becomes more acute as the agent frameworks integrated with Agent Cards, OpenAI's operator model, Anthropic's Claude tool use, and LangChain's agent toolkits, continue to expand the range of actions agents can autonomously initiate.
For founders building B2B SaaS products, internal automation tools, or agentic workflows for enterprise clients, the practical implication of Agent Cards is that the payments layer for AI agents now has a deployable implementation rather than a theoretical roadmap. A workflow that previously required human approval at every payment step can now be designed with a controlled financial identity for each agent, with the same visibility and audit infrastructure that corporate finance teams already expect. The operational cost reduction for businesses running high-frequency automated procurement, subscription management, or cloud resource provisioning is real and immediate. The risk management overhead required to implement it responsibly, KYB verification, spend policy configuration, monitoring for anomalous transactions, and incident response for agent errors, is also real and should not be underestimated by teams building on the product early.
The deeper pattern that Agent Cards represent is stablecoins arriving at the enterprise infrastructure layer through a route that bypasses the consumer and retail conversations that have dominated public discourse about crypto adoption. Whether stablecoins become the default machine-to-machine payment primitive before banks can respond with equivalent programmable infrastructure is a question about speed of execution rather than technical feasibility. The bank-issued virtual card networks already exist. The programmable spend controls already exist. What banks have not done is design any of it around the agent as the primary identity rather than a human employee. Oobit and MoonPay both shipped products built on that reframing in the same week. The category is no longer speculative.
Also read: llama.cpp Now Supports Multi-Token Prediction in Beta and the Implications for Local AI Tooling Are Bigger Than the PR Suggests • diVine Wants to Be the Short-Form Video Platform That Says No to AI and That Bet Is More Interesting Than It Sounds • IBM's MAMMAL Is a Quiet Demonstration That Biomedical AI Is Moving Beyond Single-Purpose Models
