Multiple fintech companies are designing protocols designed to allow AI agents to spend money on your behalf. Not recommend a purchase. Execute it.

For years, AI in financial services meant better fraud detection, stronger credit models and more efficient risk assessment, with a human still approving every transaction. That boundary is now starting to disappear.

Once an agent can hold a wallet and execute purchases autonomously, the governance model built around “a person authorised this” needs to be redesigned, not refined. Audit, accountability and compliance all become materially more complex.

The technology is arriving quickly. The infrastructure around it is not.

One interesting wrinkle from my own recent experiments: I have been testing AI agents purchasing products on my behalf. The agents are increasingly capable. The bigger constraint is that many ecommerce sites are actively blocking them - particularly when they operate in headless mode, communicating directly with the site rather than driving a browser like a human user. In many cases, the commercial ecosystem is resisting the technology even as the technology becomes viable.

That friction won’t last forever. History suggests that once a purchasing experience becomes materially easier, adoption tends to accelerate rapidly.

The more important question is not whether autonomous purchasing becomes commonplace. It is who carries the liability when an agent makes the wrong decision - the fintech, the model provider, the merchant, or the person or organisation that deployed it?

Boards approving AI agent pilots this year should have a clear answer before the first agent asks for a credit limit.

#AgenticAI #Fintech #AI