Twenty-eight finance and data leaders, one evening, one question: are AI agents the real deal - and how do you actually build them for a real business?
A 9-minute walkthrough of everything the evening covered - from defining AI agents to the data foundation that makes them work in practice.
The evening was built around real questions from the room - here are the four themes we worked through together.
We opened by cutting through the hype with a working definition. An AI agent is an intelligent application that can perceive inputs - like reading your general ledger or bank statement files. It can reason about what to do by breaking goals into smaller steps, act the way a human would on a computer, and remember what it's already done to apply as stored, learned knowledge.
Are they overhyped? The room's conclusion: no - the capabilities are real. But they cannot work from day one without putting the right systems and data in place first.
Four reasons agents matter specifically for finance:
Most Gen AI deployments stall here. The honest picture:
The answer isn't a better prompt - it's an architecture. Think about building agents the way you'd build a team:
The more context an agent has, the more dependable it becomes - and the less likely it is to go wrong.
"Prompts generate language. Agents generate systems. And real business processes are made up of systems."- Strategic Intelligence Forum, March 2026
These were the closing points from the evening - the ideas the room said they'd be taking back to their desks.
The opportunity is real. But agents don't work because of clever prompts alone. They work when they're built on the right structure, with the right architecture behind them. Foundation first, agent second.
Business needs consistency, accountability, and control. That means clear logic, proper checks, and human judgement at the points that matter most. Gen AI is a component - not the whole system.
Just as you would with any new hire: define the role, give them the right skills, connect them to the right tools, and make sure they have the right information to do their job well. Same logic, different medium.
If the data is messy, fragmented, or poorly governed, the agent's outputs will be too. Agents are only ever as good as the data they work with. You cannot shortcut this step.
Learn the fundamentals. Build the platform. Design the workflows carefully. Test them alongside your existing processes. Trust is not built by assumption - it is built through results, consistency, and time.
Feedback collected on the night.
"Well structured and a level of depth that wouldn't alienate non-technical attendees."
"The concept of skills as markdown files was a standout moment for me - a very useful framework to have."
"The session gave me a new lens on deploying agents into finance processes - and the architecture-first approach is exactly the right one."
"There was a lot of important information shared - and it gave me a much clearer path forward on where to start."
Two pieces published after the event that captured what the room was thinking.
"The evening reframed how I think about deploying AI agents - not as a technical challenge, but as an onboarding problem. The same things that make a new hire effective are the same things that make an agent effective."Read the article on LinkedIn
"[Paste Rebel Walters' LinkedIn post text - Rishi to provide from the post she shared]"View the post on LinkedIn
