Emil Chetty

About Emil Chetty

Emil Chetty leads data and AI product at Wise, the London-founded company for moving and managing money across borders. By his figures, Wise moved over $243 billion in cross-border value last year for nearly 19 million active customers, through more than 90 local banks and payment providers and under more than 80 licenses.

In an AI Engineer talk, he explains how Wise is designing agents that pay suppliers and reconcile results for small businesses without asking the customer to approve every step.

Notable Positions

On Approval Fatigue

Agents that ask permission for every lookup turn the user into the API. Customers want the job handled, not supervised.

On Authorization

A coding agent asks whether an action looks safe; a payment must answer whether the customer authorized this payment to this recipient on these terms. Evals and harnesses can't answer that, so Wise uses an immutable job the customer approves once. See Human-in-the-Loop.

On Enforcement

Access is granted deterministically from the overlap of the confirmed job, what Wise permits agents to do and the customer's current rights. Downstream payment services check every action against the job, so a prompt-injected agent can't add a new payee.

Key Quotes

• "We ultimately made the human the API."
• "What you approve is ultimately what we enforce."

Related Reading

• Agent Identity - Who an agent acts as
• Human-in-the-Loop - Where approvals belong