What is Ant Group's Anvita and how does it handle crypto payments?

Anvita is Ant Group's agent coordination framework that gives AI agents their own crypto wallets, enabling them to hold assets, execute trades, and settle payments in stablecoins without human approval at each step. Unlike custodial approaches, agents have independent financial identity — they're not just proxies for a user account. It runs on Ant's production infrastructure and can route through Alipay rails for fiat settlement.

How does Coinbase's x402 protocol work for AI agent transactions?

x402 is an HTTP protocol extension where a server responds with a 402 (Payment Required) status, the AI agent pays in USDC on Base, and receives the resource — all within roughly two seconds. It's framework-agnostic, so any agent that can make HTTP requests and sign transactions can use it. The design prioritizes minimalism over lock-in, letting heterogeneous agents from different vendors share the same payment infrastructure.

Why did Mastercard pay $1.8 billion for BVNK?

BVNK is a stablecoin B2B payments infrastructure company, and Mastercard acquired it to hedge against a future where AI agents default to crypto-native settlement rails instead of card networks. If agents settle peer-to-peer in stablecoins on-chain in seconds, the Mastercard network risks being bypassed entirely. The acquisition gives Mastercard stablecoin plumbing, though integrating it with legacy card infrastructure is a multi-year engineering challenge.

Blockchain

مدفوعات وكلاء الذكاء الاصطناعي: Anvita وx402 ومن يملك الشبكة

By Anurag Verma•April 25, 2026

AI Agent Payments: Anvita, x402, and Who Owns the Rail

Ant Group just shipped Anvita, an agent framework that lets AI systems hold crypto wallets, execute trades, and settle payments in real time without a human approving each step. This is not a demo. It's running on Ant's production infrastructure, processing stablecoin transactions through autonomous agents at scale.

The same week, Coinbase's x402 protocol hit developer preview, Visa published its Trusted Agent Protocol whitepaper, and Mastercard closed its $1.8 billion acquisition of BVNK, a crypto-native B2B payments company. Four moves in one news cycle. None of them compatible with each other.

That's the actual story. Not any single product, but the fact that the race to own AI agent payment infrastructure is happening right now, in parallel, with no coordination and radically different bets on what agents actually need.

What Ant Group's Anvita Actually Does

Anvita isn't an AI model. It's an agent coordination layer that wraps identity, custody, and payment execution into a single programmable surface. An Anvita agent can:

Hold a crypto wallet natively (not a custodial account on behalf of a user)
Execute trades on-chain when certain conditions are met
Pay other agents or services in stablecoins without polling a human for approval
Route through Alipay's settlement rails when fiat offramps are needed

The key architectural decision: agents have their own financial identity. They're not proxies for a user account. This matters enormously for agent-to-agent crypto transactions, where you'd otherwise need bilateral authentication handshakes before every payment.

Ant Group has $1.4 trillion in annual payment volume through Alipay. Putting that distribution muscle behind an agent framework that speaks native crypto is not a small bet.

Coinbase x402: The Protocol-First Approach

Coinbase's x402 is a different kind of bet. Instead of building a full agent runtime, it's a payment protocol, an HTTP extension that lets any server demand crypto payment before serving a response. An AI agent hits an endpoint, gets a 402 (Payment Required) status, pays in USDC, and gets the resource. The whole thing resolves in under two seconds on Base.

The elegance here is architectural minimalism. x402 doesn't care what agent framework you're using. It works the same way whether you're running Anthropic's Claude, an AutoGPT fork, or a custom Rust binary. Any agent that can make HTTP requests and sign transactions can use it.

This maps well onto how the agentic economy actually works in 2026, heterogeneous agents from different vendors calling shared infrastructure. Base hit $1B in daily DEX volume partly because it didn't try to own every layer. x402 follows the same logic.

The open question: can a protocol win without Coinbase's distribution pushing it into enterprise procurement cycles? Card networks have 50 years of that muscle. Coinbase has a few million developers and a growing institutional book.

Visa's Trusted Agent Protocol: Betting on Authentication

Visa's Trusted Agent Protocol takes a different angle entirely. Rather than building payment primitives, Visa is trying to own agent identity. The pitch: before any agent transacts on behalf of a user, it should be verified against a Visa-attested credential.

This is smart if you believe the bottleneck is trust, not settlement. Enterprise buyers are going to ask "how do I know this agent is authorized to spend money?" before they ask "how fast does it settle?" Visa has spent 50 years answering that question for cards. They're applying the same frame to agents.

The problem is latency. Card authorization infrastructure, even modernized, adds 200-800ms per transaction. For a single human purchase, that's invisible. For an AI agent making 400 micro-decisions per minute in an automated supply chain, it's a structural ceiling. The Visa Trusted Agent Protocol, as currently specced, doesn't solve the throughput problem that agent commerce creates by design.

Mastercard's $1.8B BVNK Acquisition Signals the Real Anxiety

The most telling move was Mastercard paying $1.8 billion for BVNK. BVNK is a stablecoin infrastructure company, it helps businesses issue, move, and settle with stablecoins across multiple chains. It's not an AI company. It's not an agent framework.

Mastercard bought BVNK because they watched the Anvita and x402 announcements and realized their core product, the payment rail, is at risk of being bypassed entirely. If agents can settle peer-to-peer in USDC on Base or Solana in two seconds with no intermediary, what does the Mastercard network add?

The acquisition is a hedge. BVNK gives Mastercard the stablecoin plumbing to stay relevant in a world where agents default to crypto-native settlement. But integrating a crypto-native B2B stack into a legacy card network without creating an authentication Frankenstein is a multi-year engineering problem. The clock is running.

Why Crypto-Native Rails Have a Structural Head Start

Here's the uncomfortable truth for card networks: the requirements of AI agent stablecoin payments were never on their product roadmap.

Agents need:

Programmable conditions, pay only if X is true, verified on-chain
Sub-second finality, not 3-5 business days, not even 200ms, but block-time settlement
Machine-readable receipts, transaction data that downstream agents can verify without parsing a PDF
No human-in-the-loop authorization, agents can't wait for a 2FA push notification
Composability, payments that trigger smart contract logic atomically

Card networks were built for human-initiated, human-verified, human-disputed transactions. Every one of those five requirements is either absent or architecturally hostile to how Visa and Mastercard process payments today.

Blockchain rails check all five by default. Stablecoins on L2s settle in seconds. Smart contracts encode payment conditions natively. On-chain receipts are immutable and machine-readable. There's no cardholder authentication loop because the agent's private key is the authentication.

This is the same pattern I wrote about when analyzing how stablecoin bridge architecture has matured, the primitives that seemed experimental two years ago are now the load-bearing infrastructure for real financial volume.

The Authentication Standard Is the Real Prize

Whoever sets the agent authentication standard wins the decade. Not the payment network. Not the settlement layer. The identity and authorization layer.

Right now there are at least four incompatible approaches:

Anvita's wallet-native identity (agent = wallet address)
x402's protocol-level payment attestation
Visa's centralized credential registry
Emerging W3C DID-based proposals from the open-source agent community

This is going to get messy before it gets clean. Enterprise buyers will demand interoperability. Regulators, especially in the EU and Singapore, will push for standardized agent identity frameworks as AI autonomy expands into financial decisions. The companies that get ahead of that regulatory ask with an open, auditable standard will have the same advantage that EMV chip specifications gave card networks in the 2010s.

The irony is that blockchain already has a credible answer here: on-chain identity, verifiable credentials anchored to wallet addresses, and smart contract-enforced authorization policies. The infrastructure exists. The question is whether enterprises trust it enough to stake financial operations on it, and whether the crypto-native players can package it in a way that doesn't require a six-month security audit before procurement.

The agentic economy isn't coming. Anvita is live. x402 is in developer hands. The rails are being poured right now, and the concrete sets fast.