For two decades, the internet economy has been built around a simple assumption: there is always a human in the loop. Humans sign up for services, manage passwords, approve payments, and decide when money moves. That assumption is starting to break, and a new startup believes the shift will redefine how digital commerce works.
Sapiom, a new company founded by former Shopify payments engineering leader Ilan Zerbib, has raised $15.75 million in seed funding to build what it calls the financial access layer for AI agents. The round was led by Accel, with participation from Gradient, Array Ventures, Okta Ventures, Menlo Ventures, Anthropic, Coinbase Ventures, Formus Capital, and Operator Collective. Strategic angels from companies including OpenAI, Shopify, GitHub, Vercel, Circle, and Mercury also joined.
The pitch behind Sapiom is straightforward but ambitious: AI agents are already smart enough to do real work, but they are blocked from participating in the real economy.
Today’s AI systems can write code, design marketing campaigns, plan product launches, and coordinate complex workflows. What they cannot do is pay for the tools required to execute those plans. An agent can generate an app, but it cannot provision cloud infrastructure to run it. It can design an SMS campaign, but it cannot pay Twilio to send the messages. It can build a product that needs authentication, but it cannot subscribe to an identity service.
That gap — intelligence without access, is what Sapiom is trying to close. The company is building infrastructure that allows AI agents to securely spend money under strict policies, giving them trusted access to paid APIs, software services, data, and compute. Instead of relying on pre-built integrations or human-managed accounts, Sapiom turns spending itself into a programmable function, something software can do safely and automatically.
Read more: How Wetrocloud Is Making Artificial Intelligence Human
At the core of the idea is a simple insight: money is the universal API key. If an agent can spend in a controlled, auditable way, it can access almost any service in the digital economy.
Sapiom abstracts away the hard parts that usually block this kind of automation. Identity verification, wallets, policy enforcement, risk controls, billing, and settlement are handled behind a single integration. Each transaction generates context, who the agent is, what policy applies, which vendor is being paid, and what the outcome was , creating a feedback loop that improves control and trust as usage grows.
This matters because economic activity is quietly shifting from human-to-business to machine-to-business. AI agents are moving beyond chat interfaces and starting to act on behalf of companies. They are expected to provision infrastructure, purchase data, call paid APIs, and coordinate workflows continuously, without waiting for human approval at every step.
That shift is already visible in the rise of “vibe coding” tools, which let non-technical users build apps by describing what they want in plain language. These tools can generate working software, but turning prototypes into real products still requires manual setup of payments, APIs, and third-party services. Sapiom’s goal is to remove that friction entirely, letting agents decide what services to buy and when.
Investors believe this infrastructure gap is one of the biggest blockers to the next phase of AI adoption. Amit Kumar, a partner at Accel, said he has seen dozens of startups exploring AI payments, but most focus on consumer use cases. Sapiom’s focus on enterprise infrastructure, he argues, is what makes AI agents truly usable at scale.
“If you really think about it, every API call is a payment,” Kumar said. “Every time you send a text, spin up a server, or run inference, money is moving. There’s no clean way for agents to handle that today.”
Zerbib’s background helps explain why he is tackling this problem. Before founding Sapiom, he spent nearly five years at Shopify, leading engineering for payments and helping scale Shop Pay to more than $100 billion in gross merchandise value. Before that, he founded Earny, a consumer fintech startup acquired in 2021. That experience taught him a lesson he now sees repeating: infrastructure built for one era does not stretch cleanly into the next.
Just as plastic cards enabled in-person commerce and payment APIs unlocked online shopping, Zerbib argues that machine-driven commerce needs its own primitives. AI agents are becoming economic actors, not just tools, and the systems that govern money need to reflect that reality.
Sapiom is starting with business use cases, where companies are more willing to trust software with spending decisions under defined rules. Over time, the same infrastructure could power personal AI agents that book rides, shop online, or manage subscriptions autonomously. But Zerbib is cautious about hype. His focus is on enabling real work, not encouraging more consumption.
The shift to machine-native commerce is already underway. The missing piece is not smarter models, but access. Sapiom is betting that once AI agents can safely participate in the economy, their usefulness — and their impact — will expand far faster than intelligence alone ever could.
