Sapiom has secured a $15 million seed round to build what many agentic-AI developers say they need most: a secure financial layer that lets autonomous software agents purchase and access third-party tools on their own. The startup, founded by former Shopify payments engineering lead Ilan Zerbib, aims to turn procurement, authentication, and micro-payments for APIs, data, and compute into behind-the-scenes plumbing.
Why AI Agents Need a Wallet for Secure Autonomy
Agentic systems increasingly orchestrate complex workflows across dozens of SaaS and infrastructure providers—from sending an SMS to charging a card to spinning up a GPU. Each step typically requires a human to open an account, verify identity, manage credentials, and handle billing. That manual glue doesn’t scale when agents operate continuously and make thousands of tiny decisions a day.
Research on software sprawl underscores the problem: Okta’s Businesses at Work report shows large enterprises now use over 200 apps on average, a web of vendors that complicates access control and cost management. In that environment, agents need a wallet, spend limits, and policy controls—not a shared API key and a corporate credit card.
How Sapiom’s Financial Layer Works for AI Agents
At a high level, Sapiom issues a payment and identity “envelope” for each agent. When the agent decides it needs a capability—say SMS messaging—it triggers a delegated authorization flow, obtains ephemeral credentials to a vendor like Twilio, and executes a metered purchase. Sapiom settles the micro-transaction and records a detailed audit trail tied to the agent, the task, and the vendor.
For developers using prompt-to-code or “vibe coding” platforms, the experience stays simple. Instead of manually signing up for providers, storing API keys, and wiring billing, the platform passes through usage fees to the builder while Sapiom handles provisioning, authentication, and payments in the background. The result: prototypes with SMS, email, or payments can graduate to production without a maze of procurement tickets.
The same pattern applies to data subscriptions, model inference, vector databases, and compute. An agent can purchase a dataset on demand, choose a model endpoint based on cost and latency, or burst to additional GPUs—while Sapiom enforces spend caps, business rules, and enterprise policies.
Backers Bet On Enterprise-Grade Autonomy
The round is led by Accel, with participation from Okta Ventures, Gradient Ventures, Array Ventures, Menlo Ventures, Anthropic, and Coinbase Ventures. Investors are wagering that the winning approach in agent payments won’t be consumer cards on file, but enterprise-grade controls: per-agent wallets, budget policies, vendor allowlists, and compliance baked in from day one.
Zerbib’s background at Shopify’s payments arm gives Sapiom hard-won experience in risk, reconciliation, and payouts—areas that become critical when thousands of micro-purchases occur across multiple vendors every hour. It’s a familiar playbook in fintech, repurposed for autonomous software buyers.
Early Use Cases and Real Costs for Autonomous Agents
Consider a customer-support agent that drafts replies, sends a confirmation text, checks billing status, and issues a refund. Today, that flow touches an LLM API, an SMS gateway, a billing system, and a payments processor. Costs are granular: an SMS in the U.S. is typically under one cent, LLM calls can be fractions of a cent per thousand tokens, and card processing fees vary by transaction. Multiply that across millions of events, and robust micro-billing and vendor selection meaningfully impact margins.
Sapiom can route purchases to preferred vendors, switch endpoints dynamically based on price-performance, and centralize invoices by agent and task. For finance teams, this offers the visibility of a virtual spend card per agent, backed by real-time metering and export to ERP or spend-management systems.
Security, Compliance, and Guardrails for Agent Spending
Letting software buy things raises obvious governance questions. Sapiom’s pitch centers on fine-grained policies (who can buy what, under which conditions), KYC/AML and vendor risk alignment, SOC 2-grade controls, and immutable audit logs. Ephemeral credentials reduce the blast radius of leaked keys, while anomaly detection can flag spend spikes, suspicious vendors, or off-policy purchases.
Enterprises will also demand clean integration with identity providers, procurement catalogs, and finance stacks—think Okta for access, ServiceNow or SAP Ariba for approvals, and NetSuite, Ramp, or Brex for accounting and controls. The goal is to treat autonomous purchases as first-class, reviewable business events rather than shadow IT.
From B2B Foundations to Consumer Agents and Services
While Sapiom is targeting B2B workflows, the same rails could eventually support personal agents that order rides, restock supplies, or subscribe to news and data on the user’s behalf. The barrier isn’t only technology; it’s consumer trust, consent UX, dispute handling, and safeguards that prevent overspend or malicious behavior.
For now, the immediate opportunity sits inside companies that want agents to take on real work without introducing new attack surfaces or budget chaos. If Sapiom can make autonomous purchases as controlled and auditable as swiping a corporate card—only at machine speed—it could become a crucial layer in the emerging agent economy.
