Sponge Gives AI Agents Their Own Bank Accounts

The Macro: Agents Need Wallets and Nobody Has Built Good Ones

AI agents are moving from “tools that help you do things” to “autonomous systems that do things for you.” This shift creates a problem nobody planned for: agents need to spend money. An AI agent that books a flight needs to pay for the ticket. An agent that purchases data from an API needs to pay per request. An agent that hires another agent for a subtask needs to transfer funds.

Right now, the way agents spend money is crude. Developers hard-code credit card numbers or API keys into agent configurations. There are no spending controls, no audit trails, and no way for an agent to manage its own budget. If an agent goes rogue or hits a loop, it can rack up thousands of dollars in charges before anyone notices.

The financial infrastructure for human commerce took decades to build. Bank accounts, credit cards, payment networks, fraud detection, spending limits. The agent economy needs its own version of this infrastructure, and it needs it now because agents are already transacting.

Stripe handles payments between humans and businesses. PayPal handles peer-to-peer. But neither is designed for autonomous agents that need to hold funds, make spending decisions, and transact with other agents without human approval for every purchase.

The Micro: Three Ex-Stripe Crypto Engineers Building Agent Finance

Eric Zhang, Jae Choi, and Rishab Luthra founded Sponge. All three are ex-Stripe Crypto. That background is close to ideal for building financial infrastructure for a new transaction paradigm. They understand payment rails, regulatory requirements, and the specific challenges of crypto-based financial products. They are a three-person team from YC Winter 2026 with Brad Flora.

Sponge lets agents hold funds and transact autonomously with other agents and businesses using bank accounts, cards, and crypto. The spending controls are critical: budget limits, per-transaction caps, and domain whitelisting. These controls mean a business can give an agent financial autonomy within defined boundaries.

For businesses, Sponge provides a gateway to accept payments from agents without human interaction. This matters because the current way agents purchase services, using pre-configured API keys, does not scale. Sponge lets businesses list their services and let agents discover, evaluate, and purchase them autonomously.

The platform supports TypeScript SDK and MCP integration across multiple agent frameworks. Multi-chain support covers Base, Solana, and Tempo. A Discord community is active.

The Verdict

Sponge is building for a future that is arriving faster than most people expect. Agent-to-agent commerce is not theoretical. It is happening today in limited forms, and the infrastructure to support it properly does not exist. The ex-Stripe team is exactly right for this.

The risk is timing. If the agent economy takes five years to mature instead of two, Sponge needs to find intermediate use cases that generate revenue while the market develops. Agent-to-API payments might be that bridge. Agents already purchase compute, data, and services through APIs.

The competitive risk comes from Stripe itself. If Stripe adds agent-specific payment features, the distribution advantage is overwhelming. But Stripe tends to build for established markets, not emerging ones. The agent economy is still too early for Stripe to prioritize.

In 30 days, I want to see the number of agents with active Sponge wallets. In 60 days, the question is transaction volume. Are agents actually spending money through Sponge regularly? In 90 days, I want to know about the business gateway adoption. Are services listing on Sponge and getting discovered by agents? The marketplace dynamic is where the real value lives.