Skip to main content

Sapiom's $15.75M Bet: Why AI Agents Need Their Own Wallets, Identity, and Payment Rails

· 9 min read
Dora Noda
Dora Noda
Software Engineer

When a human developer needs an API, they pull out a credit card, fill in a billing form, and start making calls. When an AI agent needs the same API, it hits a wall. No identity. No wallet. No way to pay. Sapiom's $15.75M seed round, led by Accel with backing from Anthropic, Coinbase Ventures, and Okta Ventures, is a bet that this wall is the single biggest bottleneck holding back the agentic economy — and that whoever tears it down will own the financial plumbing of a $3–5 trillion market.

The Problem No One Built For​

Every major AI lab — OpenAI, Anthropic, Google — has shipped agents that can reason, plan, and execute multi-step workflows. But the moment those agents need to buy something — a Twilio SMS, an AWS Lambda invocation, a data feed from a premium API — the workflow breaks. The agent cannot open a bank account. It cannot pass KYC. It cannot sign a terms-of-service agreement.

The result is a bizarre bottleneck: agents that can write production code and orchestrate complex pipelines still need a human to swipe a credit card on their behalf.

Ilan Zerbib, Sapiom's founder and CEO, spent nearly five years as Shopify's director of engineering for payments, where he scaled Shop Pay to over $100 billion in gross merchandise value. His diagnosis is blunt: the financial system was built for humans, and retrofitting it for machines requires a purpose-built layer — not patches on existing rails.

What Sapiom Actually Does​

Sapiom calls itself the "autonomous spend API for the machine-to-business economy." In practice, it abstracts five critical functions behind a single integration that takes as few as five lines of code:

  • Identity (KYA — Know Your Agent): Each agent gets a verifiable identity that establishes who operates it, what it's authorized to do, and under what constraints. This is the agent equivalent of KYC, but designed for software principals rather than human ones.

  • Wallets: Agents receive programmable wallets with embedded spending policies — per-transaction limits, vendor whitelists, daily caps — enforced at the infrastructure level rather than the application layer.

  • Policy Enforcement: Every transaction flows through configurable risk controls. Enterprises define guardrails; Sapiom enforces them before any money moves.

  • Multi-Rail Settlement: Payments can route across fiat rails, stablecoin networks, or API credit systems depending on the vendor and use case.

  • Metering and Billing: Usage is tracked, aggregated, and reconciled automatically, turning agent spend into auditable line items on enterprise invoices.

The thesis is simple: if you make "spend" a developer primitive — as easy to call as fetch() — agents can participate in the API economy without human babysitting.

The $67 Billion Signal​

Sapiom is not building into a vacuum. The evidence that autonomous commerce is arriving faster than most predicted is piling up:

  • Black Friday 2025: AI agent traffic surged 805% year-over-year, with agents influencing $67 billion in global Cyber Week sales — roughly 20% of all orders.

  • Market projections: Analysts estimate the agentic economy will reach $3–5 trillion globally by 2030, with up to 30% of e-commerce transaction value mediated by AI agents.

  • Infrastructure readiness: 85% of financial institutions acknowledge their current systems cannot handle high-volume, autonomous agent-initiated transactions — a gap that represents both a crisis and an opportunity.

The pattern is familiar from the early internet: traffic arrives before infrastructure is ready. The companies that build the pipes capture durable value.

The Payment Rails Race: TradFi vs. Crypto​

What makes the AI agent payments landscape uniquely interesting is that two fundamentally different financial systems are racing to own it simultaneously.

The Traditional Finance Approach​

In Q4 2025, the three largest payment networks moved in near-lockstep:

  • Visa launched Intelligent Commerce, issuing 16-digit tokenized credentials that AI agents use like virtual credit cards linked to a consumer's real account. Partners include Microsoft, Shopify, Stripe, and Worldpay.

  • Mastercard debuted Agent Pay, completing the first live agentic payment transaction in September 2025 — an actual AI agent purchasing a product using a tokenized credential.

  • PayPal introduced the Agent Toolkit and adopted the Agentic Commerce Protocol, embedding payments directly into ChatGPT conversations starting in 2026.

These solutions inherit the existing card network infrastructure: merchant agreements, chargeback protections, fraud detection, and regulatory compliance. They work today, at scale, for consumer-facing agent commerce.

The Crypto-Native Approach​

The crypto ecosystem is building a parallel stack optimized for machine-to-machine transactions where the card network model breaks down:

  • Coinbase Agentic Wallets (February 2026) give any AI agent an autonomous crypto wallet using the x402 protocol — the same HTTP 402 "Payment Required" status code repurposed for blockchain micropayments. Over 50 million transactions have already flowed through x402, with programmable spending limits and TEE-based key isolation ensuring agents never expose private keys to their own language models.

  • Stripe's x402 integration (February 2026) enables developers to charge AI agents $0.01 USDC per API request on Base — no account, no API key, just pay-per-call. This turns any API into a vending machine for autonomous software.

  • MoonPay Agents provides non-custodial infrastructure where users complete one-time identity verification, fund a wallet, and then let AI agents trade, swap, and transfer digital assets programmatically.

The crypto approach trades the card networks' consumer protections for programmability, instant settlement, and the ability to handle sub-cent micropayments that would be uneconomical on traditional rails.

Where Sapiom Fits​

Sapiom occupies a deliberate middle ground. Rather than choosing sides in the TradFi-vs-crypto debate, its multi-rail settlement architecture routes payments across both systems depending on what the vendor accepts and what the use case demands. An agent buying a $500/month SaaS subscription might settle on fiat rails. The same agent paying $0.003 per API call to a decentralized oracle might use stablecoins.

This rail-agnostic positioning reflects a bet that the agent economy will not be mono-rail — and that the infrastructure layer that abstracts away the payment routing decision will capture more value than any single rail.

Enterprise vs. Consumer: Two Different Wars​

A subtle but critical distinction separates Sapiom's approach from the Visa/Mastercard/crypto consumer plays.

Visa's Intelligent Commerce and Mastercard's Agent Pay are designed for consumer agents — AI that shops for humans, compares prices, and completes checkout. The agent acts as a proxy for a human spender with a human credit card.

Sapiom targets enterprise agents — AI that autonomously procures APIs, compute, data, and software tools to complete business workflows. The spender is a company, not a consumer. The procurement policies, approval workflows, and audit requirements are fundamentally different.

This distinction matters because enterprise procurement is a $12 trillion global market where the buying process is governed by policies, budgets, and compliance requirements that consumer payment rails were never designed to handle. When an enterprise AI agent needs to spin up GPU instances across three cloud providers, subscribe to a market data feed, and purchase API credits from a specialized vendor — all within a single automated workflow — it needs infrastructure that understands corporate spending controls, not consumer credit limits.

Accel partner Prayank Swaroop, who led the investment, reportedly met dozens of startups in the AI payments space before backing Sapiom. His conviction centered on Zerbib's focus on the enterprise financial layer rather than consumer payments — a less glamorous but potentially more defensible market.

The Identity Problem That Defines the Category​

Underneath the payments question lies a harder one: how do you verify what an AI agent is?

Humans have passports, driver's licenses, and social security numbers. AI agents have none of these. Yet the financial system requires identity for every transaction — for compliance, fraud prevention, and accountability.

The industry is converging on "KYA" (Know Your Agent) as the answer, but implementations vary wildly:

  • Sapiom's KYA ties each agent to its operating organization, creating verifiable context around every transaction — who built the agent, what policies govern it, which vendor it's transacting with, and what outcome it achieved.

  • Coinbase's approach uses Trusted Execution Environments (TEEs) to isolate agent keys, creating a hardware-backed identity that's cryptographically verifiable but not tied to traditional identity documents.

  • Visa's Trusted Agent Protocol works with Cloudflare's Web Bot Auth technology to identify verified AI agents at the network level, creating a card-network-compatible identity scheme.

Which model wins will likely determine which payment rails dominate. If regulators require human-traceable identity for agent transactions, TradFi rails with their existing KYC infrastructure have an advantage. If cryptographic identity verification proves sufficient, crypto-native rails can operate with less friction.

What This Means for Web3 Infrastructure​

The AI agent payments race has direct implications for blockchain infrastructure providers. Every crypto-native agent payment solution — x402, Agentic Wallets, MoonPay Agents — requires reliable blockchain node access, transaction broadcasting, and state queries. As agent transaction volumes scale from millions to billions of daily micropayments, the demand for high-availability, low-latency blockchain RPC infrastructure will grow proportionally.

The shift from human-initiated to agent-initiated transactions also changes the performance profile. Humans might send a few transactions per day. An AI agent orchestrating a workflow might send hundreds per minute. Infrastructure providers must handle burst traffic patterns that look nothing like traditional user behavior.

The Road Ahead​

Sapiom plans to launch in limited beta with select enterprise customers in coming months, with a broader rollout later in 2026. The competitive landscape will clarify quickly:

  • Q1–Q2 2026: Visa and Mastercard's agentic commerce frameworks begin processing real consumer transactions. Stripe's x402 moves from preview to production.

  • Mid-2026: Enterprise agent procurement volumes become measurable. The first wave of "agent-native" SaaS companies — built to sell to AI agents rather than humans — emerge.

  • Late 2026 and beyond: Regulatory frameworks for agent identity and autonomous transactions begin taking shape. The KYA standard either converges or fragments.

The $15.75M seed round is modest relative to the ambition. But if Sapiom's thesis is correct — that the agent economy needs a purpose-built financial layer, not a retrofit of human payment systems — the company is positioned at the most strategic chokepoint in the entire agentic stack: the moment when software reaches for its wallet.

BlockEden.xyz provides high-performance blockchain API infrastructure designed for the programmatic, high-throughput access patterns that AI agent applications demand. As autonomous agents drive the next wave of on-chain transactions, our node infrastructure ensures the reliability and speed that agentic commerce requires. Explore our API marketplace to build the agent-ready infrastructure your applications need.

Share on Twitter