Last week, a Tether-backed startup called Oobit launched something that sounds like science fiction but is actually live infrastructure: Visa-supported corporate expense cards designed specifically for AI agents. Not for the humans managing AI systems. For the AI agents themselves.
These “Agent Cards” let AI bots make purchases autonomously using USDT balances, with built-in corporate controls — per-transaction limits, category-level spending caps, and automatic audit trails. Every transaction, approved or declined, gets logged with a human-readable reason.
This isn’t a concept demo. It’s a working product, accepted wherever Visa works online. And it marks a tangible shift in how businesses will need to think about AI — not just as a tool that completes tasks, but as an entity that spends money.
How Agent Cards Actually Work
The mechanics are relatively straightforward, which is part of what makes them significant. Each AI agent gets its own dedicated virtual Visa card, funded from a USDT (Tether stablecoin) balance. No fiat conversion is needed for transactions — the card operates on crypto rails but spends like a traditional corporate card at any Visa-accepting merchant online.
The control layer is where the practical value lives. Businesses can set:
- Per-transaction spending limits — cap what any single purchase can cost
- Per-merchant caps — limit how much an agent can spend at a specific vendor
- Category-level restrictions — block spending in categories that don’t match the agent’s function
- One card per agent — each AI agent has its own card and spending profile, preventing cross-contamination of budgets
Every transaction generates an automatic log entry with a human-readable explanation of what happened and why. Approved purchases include the reason. Declined transactions explain what rule was triggered. This creates a built-in audit trail without requiring manual reconciliation.
Oobit also offers payment processing integration, allowing AI agents to manage recurring tasks like subscription billing, vendor payouts, and service payments connected to platforms like Stripe. The system is designed to work with existing finance workflows rather than requiring a complete overhaul.
Why This Matters Now
The significance isn’t the card itself — it’s what the card represents. We’re watching the infrastructure layer form for a new category: agentic commerce.
Until now, AI agents have operated in a mostly task-based model. They write emails, generate reports, analyze data, schedule meetings. They do things. But they don’t buy things. The moment an AI agent can autonomously decide to purchase a service, renew a subscription, or pay a vendor, the relationship between business and AI changes fundamentally.
This matters for several concrete reasons:
Operational efficiency at scale. If you’re running multiple AI agents handling different business functions — marketing, procurement, customer support — each one may need to make purchases. A marketing agent might need to buy stock images. A procurement agent might need to place small orders. Currently, these transactions require human intervention. Agent Cards remove that bottleneck.
Speed in automated workflows. Many AI-driven workflows stall when they hit a payment step that requires human approval. If an AI agent is managing ad spend, replenishing inventory, or purchasing cloud resources, inserting a human approval step at every transaction defeats the purpose of automation. Controlled autonomous spending solves this.
Financial identity for AI. This is the conceptual leap that most businesses haven’t made yet. When an AI agent has its own card, it has its own financial identity — its own spending history, its own limits, its own audit trail. This creates accountability at the agent level rather than the human level, which is essential for any business running multiple autonomous systems.
The Governance Problem You Need to Solve First
Here’s where the enthusiasm needs to be tempered with practical caution. Giving AI agents financial autonomy introduces governance challenges that most businesses aren’t prepared for.
Liability. If an AI agent makes a bad purchase — overspends, buys from a fraudulent vendor, or misinterprets its instructions — who’s responsible? The business owner? The developer who configured the agent? The platform? Current legal frameworks don’t have clean answers for AI-initiated financial transactions.
Drift. AI agents can behave unpredictably over time, especially as they interact with changing external environments. An agent that’s well-calibrated today might make increasingly questionable spending decisions as conditions change. Without active monitoring, small drift can compound into significant financial exposure.
Control architecture. The controls Oobit offers — per-transaction limits, category restrictions, audit logging — are a solid starting point. But businesses need to think about what happens beyond the individual transaction. Who reviews the aggregate spending patterns? How often are limits recalibrated? What triggers a human review?
Compliance. Depending on your industry and jurisdiction, autonomous AI spending may create regulatory questions. Financial reporting standards, anti-money laundering requirements, and procurement regulations weren’t designed with AI agents in mind. Businesses that move fast on agentic commerce without consulting compliance could create problems that are expensive to unwind.
The practical recommendation: set up your governance framework before you deploy the cards. Define spending limits, review cycles, escalation triggers, and compliance checks in advance. The technology is ready. The question is whether your controls are.
The Broader Agentic Commerce Landscape
Oobit isn’t operating in a vacuum. Multiple companies are building infrastructure for AI agents to transact financially.
Coinbase has been developing capabilities for AI agents to interact with crypto wallets and make payments. Their approach focuses on the underlying blockchain infrastructure, allowing agents to transact in multiple cryptocurrencies with programmable controls.
OKX is exploring similar territory, building tools that allow AI bots to execute trades and manage financial operations with defined parameters.
Stripe has been quietly expanding its API capabilities to support machine-initiated transactions, recognizing that a growing share of online purchases will be triggered by AI systems rather than human users clicking buttons.
Skyfire, another startup in this space, has been building payment infrastructure specifically designed for AI-to-AI transactions — cases where one AI agent needs to pay another AI agent for a service, with no human in the loop at all.
What’s forming is a category, not just a product. The infrastructure for AI agents to participate in commerce — as buyers, as spenders, as financial actors — is being built across multiple companies and approaches. The convergence suggests this isn’t speculative. It’s happening.
For founders and SMB operators, the practical implication is that your competitors may soon have AI agents that can execute financial transactions faster and more autonomously than your current workflows allow. The advantage won’t come from having the cards — it’ll come from having the governance, the workflows, and the integration to use them effectively.
What to Watch and What to Do
If you’re running a business that uses or plans to use AI agents, here’s what deserves attention now:
Start mapping where AI agents hit payment walls. Look at your current AI-powered workflows and identify where they stop because a human needs to authorize a purchase. These are your highest-value candidates for autonomous spending.
Evaluate your risk tolerance. Autonomous AI spending isn’t equally appropriate for all business functions. Low-stakes, high-frequency transactions (cloud services, stock photos, small SaaS subscriptions) are low-risk candidates. High-value procurement or client-facing purchases probably aren’t — at least not yet.
Build monitoring before deploying autonomy. Don’t give AI agents spending power and then figure out how to track it later. Set up dashboards, alerts, and review cadences first. The audit trail is only useful if someone’s actually reading it.
Watch the regulatory landscape. Financial regulators are just starting to grapple with AI-initiated transactions. Early movers will need to adapt as rules crystallize. Stay close to your compliance advisors on this one.
Don’t over-rotate on crypto. Oobit’s USDT-based approach is one model. The broader trend — AI agents with financial autonomy — will play out across fiat rails, crypto rails, and hybrid systems. Focus on the capability, not the currency.
AI agents spending money isn’t a concept anymore. It’s infrastructure you can deploy today. The question isn’t whether this becomes normal — it’s whether your business is ready to manage it when it does.
The companies that build their governance frameworks now, while the stakes are still relatively low, will be the ones that scale agentic commerce effectively later. The ones that wait until it’s urgent will be playing catch-up with both the technology and the risks.
What’s Next?
Exploring AI agent deployment for your business? OpenVerb covers the practical side of agentic AI — governance, workflows, and real-world implementation for founders who want to move fast without breaking things.