Back by popular demand — investor and corporate finance specialist Matthew Costello returns to show how far AI-assisted financial modelling has moved in just six months.
Matt walks through his current AI setup and workflow, including how he structures “agent” systems (his own agent, “Giorg5,” delegates to sub-agents for research, admin, and building) rather than relying on one-off prompts. He demonstrates two real, live financial models built this way — a project finance/carbon sequestration model and a three-statement model for a beverage startup raising capital — each complete with an audit trail, source documentation, and an AI-generated “teach me” walkthrough explaining how the model works.
Throughout, the conversation returns to a central theme: the model may be AI-assisted, but the accountability isn’t. Matt and Danielle discuss “verification debt,” why modellers still need to be able to hand-build and explain a model to a client or board, and where human judgement remains essential even as AI takes on more of the heavy lifting.
We cover:
- The shift from “typing formulas” to designing and directing models
- Models vs. harnesses vs. skills vs. agents — a practical breakdown of the current AI toolkit
- Structuring a project as a folder/knowledge base an AI agent can work from, instead of a single prompt
- Two full model build demos, each with audit reports and auto-generated documentation
- Why accountability, professional judgement, and the ability to explain a model still sit with the modeller
