Victoria Zuo, a San Francisco-based partner at the venture firm, said the clearest analogy already exists inside the payments infrastructure: the long-established difference in interchange rates between card-present and card-not-present transactions, which reflects the higher fraud risk of a transaction completed without a physical card and a human at the terminal.

“Vendors should be properly rewarded for that risk, similar to how interchange fees are different between card-present and card-not-present transactions,” Zuo said, describing a category she labels human-present and human-not-present transactions. Authentication and authorisation, she added, are the key controls for limiting chargebacks and fraudulent requests in an agentic environment.

Zuo said the gap between the technology’s capability and institutional trust in it remains wide, and that she herself reflects the ambivalence many professionals feel. Despite being an active follower of AI developments, she has not connected her bank accounts to a general-purpose AI assistant.

“It comes down to trust. I still haven’t linked my bank accounts to my ChatGPT Pro account,” she said, adding that she expects specialist vertical agents in financial services and healthcare to earn that trust before general-purpose horizontal agents do.

Her investment interest follows from that view. She said startups focused on granular permission controls for payments, the kind of architecture that would let a user instruct an agent to transact only within defined parameters, represent a category worth watching.

A second thread in Zuo’s analysis concerns how merchants and financial-services providers need to rethink distribution when the end customer is an AI agent rather than a person.

Source link