Most of what gets called AI in finance is a copilot. It suggests, recommends and then waits for a person to click approve. An agent is different. It reads the invoice, checks it against the PO and the contract, applies your policy, posts to the ERP, and escalates only when judgment is genuinely required.
Luis Solano of Huron and Imran Gohar of AppZen work through where that line sits: which finance work an agent can complete on its own, which needs a human checkpoint, and which should stay with people. They cover how teams earn enough trust in agent decisions to widen the scope, and what changes for the humans once the repetitive work moves.
In this session:
- Why the AI-assisted metric is a trap, and what AI-completed reveals instead
- The verifiability spectrum: where agents can run unsupervised and where they still need a person
- How a proof of concept builds the organizational trust that lets agents take on more
- What the people on a hybrid finance team actually spend their time on, and why those roles grow in value
- Which finance functions get rebuilt first, and what that looks like over the next 18 months