Skip to main content

Oobit's Agent Cards: How Tether Just Handed Every AI Bot a Visa Card

· 13 min read
Dora Noda
Dora Noda
Software Engineer

On April 30, 2026, a Tether-backed payments startup did something no Fortune 500 bank, no incumbent fintech, and no Silicon Valley unicorn has yet shipped to production: it issued corporate Visa cards directly to autonomous AI agents.

Oobit's Agent Cards launch is short on flash and long on consequence. Each AI agent — your customer-support bot, your ad-buying optimizer, your DevOps incident responder — gets its own virtual Visa card, funded directly from a USDT treasury, with spend policies that the agent itself cannot override. No fiat conversion. No human in every approval loop. Just a card, a pile of stablecoins, and a server-side rulebook that decides what the model is allowed to buy.

It is, on first read, a small product launch. On second read, it is the first salvo in a category war over who issues the corporate card of the agent economy.

The 48-Hour Window That Validated a Category​

Forty-eight hours after Oobit shipped Agent Cards, MoonPay announced its own answer: MoonAgents Card, a virtual Mastercard debit product that lets users and AI agents spend stablecoins anywhere Mastercard is accepted, built with Monavate and self-custodial wallet provider Exodus.

Two products. Two networks. Two stablecoin philosophies. Forty-eight hours.

That cadence matters. When two well-capitalized teams ship competing products in the same week, they are not reacting to each other — they are both reacting to a third signal that says the market is about to compress. In this case, the third signal is the convergence of three trend lines that have been crawling toward each other since 2024:

  1. Agentic AI moving into production. McKinsey's 2025 State of AI survey found that 23% of organizations are now scaling at least one agentic AI system in their enterprise, with another 39% actively experimenting. That is the demand floor.
  2. Stablecoin infrastructure crossing into mainstream rails. Tether's USDT alone now sits at roughly $186 billion in circulation; the total stablecoin market cap has crossed $317 billion. That is the funding floor.
  3. Card networks opening their pipes to agent traffic. Visa Intelligent Commerce Connect and Mastercard Agent Pay both went live earlier in 2026 as protocol-agnostic on-ramps for AI-initiated transactions. That is the rail.

When demand, funding, and rails arrive in the same quarter, the products that connect them ship in clusters. The Oobit-MoonPay one-two is the cluster.

What Oobit Actually Built​

Strip away the press release and the architecture is brutally simple — which is the point.

  • One agent, one card. Each AI agent gets its own dedicated virtual Visa card number. No shared corporate card. No reissued physical plastic. Just a tokenized PAN bound to a single non-human principal.
  • USDT-native funding. Cards draw directly from a company's USDT treasury at Tether. There is no fiat on-ramp, no pre-funded prepaid balance, no nightly settlement to a partner bank. The dollar your agent spends is the same dollar that was on-chain a millisecond ago.
  • Server-side spend policies. Every card carries a policy bundle — category whitelists, per-transaction caps, daily ceilings, merchant blocklists — enforced at Oobit's authorization server, not in the agent's prompt. If GPT-5 or Claude tries to charge a forbidden category, the network declines the auth before the request ever touches the merchant.
  • Real-time human-readable logs. Every approved or declined transaction emits a structured log that finance teams can read without parsing JSON. No "what did the bot just do" forensics.
  • Framework-agnostic API. Oobit advertises native compatibility with OpenAI, Claude, AutoGen, and LangChain. That phrasing matters: it says the company built a standard agent-callable surface rather than vendor-locking to a single framework — the same UX bet Stripe and Brex made on developer-first APIs in the 2015–2018 fintech wave.

Onboarding opened to a "founding group" of businesses on launch day, with KYB-verified expansion rolling through June 30. Tether backs the issuing infrastructure.

The product is not novel because it does something nobody else can do. Crossmint, Lobster.cash, and Privy already issue agent-scoped cards. The product is novel because of who is behind it and what the funding source is.

Why USDT-Native Changes the Economics​

Most "stablecoin debit cards" on the market today are accounting fictions. The user holds USDC or USDT in a wallet; the card issuer converts it to fiat at the moment of authorization through a partner bank; the merchant settles in dollars. The stablecoin is a funding source, not a settlement instrument.

Oobit's Agent Cards push the stablecoin one layer deeper into the stack. Funding draws from the company's USDT treasury directly — no fiat round-trip, no FX spread on cross-border auths, no settlement latency from waiting on the partner bank's overnight cycle. For an agent making thousands of small purchases across borders (think: an ad-buying agent rotating across regional ad networks, or a procurement agent purchasing SaaS licenses in 40 countries), the cumulative spread savings show up in the P&L.

This is also the strategic crux of the deal. Tether is the largest stablecoin issuer on earth and has spent two years watching Circle build distribution moats — Stripe's stablecoin pilots, Anchorage's M0 modular issuance stack, and the rest of the USDC commerce funnel. USDT, despite leading market cap, is structurally weaker in the "embedded into Western fintech" category that defines new agent commerce volume.

Backing Oobit gives Tether something Circle cannot easily replicate: a USDT-native commerce rail that bypasses the Stripe-Bridge-Anchorage stack entirely. If agent commerce is the next growth surface for stablecoins, Oobit Agent Cards become the funnel that pulls USDT into that surface before USDC can stake a claim.

Server-Side Enforcement Is the Unsexy Breakthrough​

Of all the decisions Oobit made, the one that will age best is enforcing spend policy server-side rather than in the agent's prompt or middleware.

Every team that has shipped a non-trivial LLM agent has learned the same lesson: the model will, eventually, do the thing you told it not to do. Prompt-based guardrails leak. Middleware checks can be bypassed by a sufficiently determined function call. Tool-use restrictions don't survive a model upgrade. The only durable answer is to push policy out of the model's blast radius entirely.

Oobit's design treats the agent the way a corporate IT shop treats an intern: trusted to act, untrusted to be the security boundary. The card network — not the agent — is the policy enforcement point. If the agent hallucinates a $50,000 cloud GPU rental at 3 a.m., the auth fails at the network and the policy never even reaches the merchant.

This is the same architectural pattern that defined cloud security's evolution from monolithic servers to microservices with isolated trust boundaries — and the same pattern Coinbase's Agentic Wallet adopted by separating "wallet as callable service" from agent reasoning. The industry is converging on a consensus that should not have been controversial but somehow was: agents must not be the security boundary.

Visa vs. Mastercard, Round Two​

Watch the Oobit-MoonPay split closely, because it is the 1980s Visa-vs-Mastercard segmentation playing out again at agent-economy speed.

DimensionOobit Agent CardsMoonPay MoonAgents Card
NetworkVisaMastercard
Stablecoin modelUSDT-onlyMulti-stablecoin (USDC, USDT, future KRW)
CustomerKYB-verified businessesRetail and institutional
Card designAgent-onlyHuman + agent shared
Wallet modelTreasury-fundedSelf-custodial (Exodus partnership)
Geographic launchUS-led, Q2 2026 expansionUK and LATAM today; US/EU coming

These are not minor product differences. They are the contours of two different bets on what agent commerce will become. Oobit is building the enterprise agent expense card — the equivalent of a Brex card, but for bots. MoonPay is building the consumer agent payment card — the equivalent of an Apple Cash, but the spender is an LLM acting on your behalf.

Both might be right. The 1980s ended with Visa and Mastercard both winning, segmenting the credit card market into overlapping but distinguishable lanes. The 2020s agent-card market may end the same way — and the early movers who define the segment shape will collect the toll.

The Competitive Map: Five Bets on the Same Question​

Strip back further and the broader landscape clarifies. As of May 2026 there are at least five distinct architectural bets on "where does the agent corporate card category land":

  1. Crossmint Agent Cards — multi-network virtual cards (Visa + Mastercard) plus stablecoin wallets across 40+ chains, sold as a developer API. Launched 2025, built for Solana-native agent platforms first, now broadening.
  2. Lobster.cash (Crossmint + Mastercard + Circle + Visa + Basis Theory + Stytch) — an open payment standard that lets AI agents transact through consumers' existing Mastercard cards. The "agent-payment WalletConnect" play.
  3. Mastercard Agent Pay — network-level agent identity and Agentic Tokens, expanded recently into Hong Kong as part of an international agentic-commerce push.
  4. Visa Intelligent Commerce Connect — protocol-agnostic, supporting MPP, ACP, UCP, and Trusted Agent Protocol. Visa's positioning is to be the network that authenticates, authorizes, and tokenizes agent payments regardless of which agent stack initiates them.
  5. Coinbase Agentic Wallet — wallet-as-callable-service, custody-only, no card issuance. Adjacent rather than competitive.

Oobit's contribution to this map is the first single-vendor, USDT-native, KYB-only stack that ships actual cards rather than protocols. It is the closest thing the agent economy has to a Brex moment: a vertically integrated product that says "you don't need to assemble five primitives — here is the card, signed and shipped, with the spend policy already attached."

The Numbers Behind the Bet​

How big is the addressable market? It depends on which research firm you trust.

  • Juniper Research's April 2026 forecast pegs agentic spend at $8 billion in 2026, climbing to $1.5 trillion globally by 2030.
  • Grand View Research estimates the agentic commerce market at $5.71 billion in 2025, growing to $7.71 billion in 2026.
  • The retail and e-commerce slice alone, according to Mordor Intelligence, is worth $60.43 billion in 2026, with a 29% CAGR through 2031.
  • McKinsey's longer horizon: agentic commerce could generate up to $1 trillion in US B2C retail revenue by 2030, with global projections reaching $3 trillion to $5 trillion.
  • The corporate card category itself was $150 billion in 2025 and is forecast to reach $280 billion by 2033.

Take the most conservative estimate — Juniper's $8B in 2026 — and the question for Oobit is not whether the pie is big enough. It is whether the company can capture share before Brex, Ramp, Mercury, Stripe, and Visa themselves ship competing agent SKUs. Coinbase CEO Brian Armstrong has said publicly that "there will be more AI agents transacting online than humans very soon"; Circle CEO Jeremy Allaire predicts "literally billions of AI agents" transacting onchain in three to five years. Both are interested parties, and both are moving.

What This Means for Infrastructure​

For anyone running developer infrastructure for blockchain applications, the agent-card stack creates a new traffic shape that does not look like DeFi.

DeFi optimized for occasional, high-value, latency-tolerant swaps. Agent commerce is the opposite: high-frequency, low-value, latency-sensitive spend authorizations, real-time balance checks, and on-chain settlement notifications. The peak hour is not Friday at 4 p.m. UTC when an arbitrageur kicks off a 50-step arbitrage; it is every minute of every day, distributed across thousands of agents each issuing a handful of small auths.

That changes the RPC pattern. It changes how teams cache stablecoin balance reads. It changes the latency budget for transaction submission. And it changes which chains matter — fast-finality stablecoin rails (Solana, Base, Tron for USDT) become disproportionately attractive relative to slower L1s.

BlockEden.xyz powers enterprise-grade RPC and indexing infrastructure for the chains where stablecoin and agent commerce volume is concentrating — including Sui, Aptos, Solana, Ethereum, and Base. If you are building agent-native applications that need predictable read latency and high-throughput transaction submission, explore our API marketplace for infrastructure built to last across the next traffic shape.

The Question for the Rest of 2026​

The rest of 2026 will answer one question above all others: does the agent-card category consolidate around a small number of vertically integrated stacks, or does it commoditize into a thin layer that any fintech can ship on top of Visa Intelligent Commerce Connect or Mastercard Agent Pay?

If consolidation wins, Oobit and MoonPay are early to a winner-take-most market and Tether's bet pays for itself many times over. If commoditization wins, the value migrates up the stack to whoever owns the agent runtime — OpenAI, Anthropic, Google, or whichever framework wins the agent OS war — and the card itself becomes an undifferentiated commodity, like the Stripe API behind ten thousand SaaS checkouts.

The smart money is on a hybrid: a small number of dominant card issuers in the enterprise tier (Oobit, MoonPay, possibly a Brex or Ramp pivot) sitting on top of network-level agent rails (Visa, Mastercard) that absorb everything else. The 1980s playbook, with stablecoins replacing fiat balances and AI agents replacing human users.

What is no longer in question is whether agents will spend money. They will. The only question left is whose card they swipe.

Sources​

Share on Twitter