AI and Money: Why Legacy Financial Systems Fail for AI Agents

AI builders are rapidly innovating to expand what AI agents can do for consumers and businesses. But to fully realize the potential of the agentic era, intelligent AI systems need the ability to handle money.

Unfortunately, today's financial systems only serve humans. By design, they fight AI agents and resist intelligent automation. This situation creates fundamental barriers to the development of the agent economy. To work with money, AI systems need new financial institutions that are designed from the ground up for AI.

Let’s walk through 5 key challenges that current financial systems pose for AI solutions.

1. Identity and Verification

For AI agents to use financial services, they need new identity and verification protocols and schemes to ensure secure and reliable transactions. Existing approaches will not work for an agentic internet. Agent builders will need to solve problems such as verifying that an agent in a transaction is the agent it claims to be, identifying who owns the agent, and, potentially, who is using the agent. Traditional financial institutions are not prepared to lead the creation and adoption of these new approaches.

In addition, agent builders can’t merely use simplistic API wrappers on existing payment networks or blockchain solutions. Identity and verification are required for fraud and risk management, see below, as well as regulatory compliance requirements such as Anti Money Laundering (AML) and Combating the Financing of Terroism (CFT). Agent builders need to be careful not to ignore these requirements. Developers take on significant legal risks when they choose to use simple payment APIs, fail to partner with regulated financial institutions, and skip compliance requirements. (Think jail time.)

2. Legacy Payment Networks

Legacy payment systems such as card networks, ACH, and wires are not well suited to AI agents. First developed in the 1970s, they’re slow, expensive, and designed to resist intelligent automation. They require complex layers of intermediaries that should be unnecessary in a world of autonomous automated systems powered by AI. They are entirely the wrong approach for the AI agent economy, where agents will autonomously transact with each other in a wide range of contexts.

The good news is that regulated stablecoins, such as USDC, are ready for primetime and ideally suited for agentic payments and money transfers. The bad news is that very few existing financial institutions have the capacity to reliably and securely use stablecoins, much less utilize them for AI-powered financial transactions.

3. Fraud and Risk Management

Fraudsters play where money moves. Payments seem simple on the surface: enter a card and hit submit. But behind the scenes, financial institutions maintain complex systems and procedures to prevent fraud and manage risk. Today’s systems were not built for an agentic world.

It’s inevitable that as agents become participants in the economy, criminals will find new ways to commit fraud. To successfully integrate agents into financial transactions, we’ll need new approaches to fraud and risk management. These new strategies will fight malicious AI with a combination of AI and humans in the loop. They won’t be built by legacy financial institutions that move at a glacial pace.

4. Agentic Interfaces and Integrations

The web UI and APIs offered by existing financial institutions are not well suited to the needs of AI-based systems. Agentic workflows demand different kinds of interfaces, such as the ability to hand off end-user conversations to conversational financial agents. Moreover, the AI industry is innovating rapidly with approaches for discovering and accessing intelligent services, and traditional financial institutions will not keep up with the change. It’s time to rethink the integration surface for financial services and adapt it to the unique needs of AI agents.

5. Monitoring and Reporting Systems

To monitor agent behavior and understand their interactions with financial services, we need new reporting tools, dashboards, and ways to present information to humans and monitoring agents. Current financial infrastructure does not provide the telemetry and controls that will be required for agentic systems.

Final Thoughts

The agentic internet is happening. It’s creating new opportunities and challenges that existing financial systems can’t handle. These systems should be rebuilt from the ground up to empower intelligent AI systems that will require access to a broad array of financial products to increase prosperity for all. We’re psyched to be building a new AI-first, regulated financial institution that will accomplish these goals.