Wall Street just told you how AI agents will pay their bills—and banks are already worried about who picks up the tab when things go wrong.
The Summary
- Goldman Sachs says AI agents are shifting from conversation to action, moving from answering questions to booking travel and making purchases
- The bank expects agents to monetize through ads and premium subscriptions—the same business models that built Web2
- Financial institutions are raising concerns about fraud and scam risks as agents gain purchasing power
- Meta's Muse launch brought the monetization question into focus, but the fraud liability question remains unanswered
The Signal
Goldman Sachs is calling this "the beginning of a paradigm shift" from conversational AI to action-oriented agents. Eric Sheridan, who leads Goldman's tech research group, says the mass market will look like Web2 all over again: ads and subscriptions. That's the revenue model that built Google, Facebook, and Netflix. It's also the model that turned users into products and attention into currency.
The timing is notable. Meta just launched Muse, an agent designed to shop and book on your behalf. OpenAI, Google, and Anthropic are all racing toward similar capabilities. The technology works. The question Goldman is answering is whether the business model works.
"The mass market for AI agents will be monetized with advertising and subscriptions, much like the way the web operates today."
But while Goldman maps the revenue side, banks are flagging the liability side. Financial institutions are worried about agents making purchases that turn out to be scams or fraudulent transactions. Who eats the loss when your AI agent books a fake hotel or wires money to a sophisticated phishing scheme? The agent? The platform? The user? The bank?
This isn't hypothetical hand-wringing. Banks process the chargebacks. They field the fraud claims. They know what happened when e-commerce took off without robust identity verification. They're seeing the same pattern: new technology enabling transactions faster than fraud prevention can scale.
Key tensions:
- Revenue clarity (ads + subscriptions) vs. liability ambiguity (who pays for agent mistakes?)
- Agent capability moving faster than fraud infrastructure
- Web2 business models being grafted onto Web4 agent behavior
The monetization path Goldman describes assumes platforms control the agents and capture the revenue. That's the centralized model. But if agents can transact autonomously, the fraud surface area explodes. Every agent becomes a potential attack vector. Every purchase becomes a trust problem.
The Implication
If you're building agents, Goldman just told you the business model Wall Street expects to see. If you're a bank, you're about to get hit with a wave of "my AI bought this and I didn't mean it" disputes. The smart play for platforms is to solve fraud and liability before regulators force a solution. The smarter play for builders is to consider whether Web3 identity and on-chain transaction logs could make agents more auditable and fraud-resistant than the ad-supported model Goldman is predicting.
Watch for the first major lawsuit where an AI agent's purchase goes wrong and everyone points fingers. That case will set the precedent for how agent commerce actually works.