Asking Good Questions with Edward Roske
In this podcast, we explore the CFO's office's past, present, and future, examining how technology and AI are transforming finance. Join Edward Roske and referents as they discuss innovative ways to enhance the role, optimize processes, and shape the future of financial leadership.
Designed for CFOs and finance leaders, each episode provides actionable insights into leveraging technology to drive growth, improve efficiency, and address challenges.
With an engaging tone, the podcast simplifies complex concepts, covering topics like automating tasks, real-time data analysis, and strategic planning.
Key themes include the shift from traditional finance methods to innovative, technology-driven processes and real-world examples like scenario modeling and automated reporting.
The podcast also highlights future trends, such as generative AI for forecasting and advanced analytics for decision-making, promising to shape financial management. Edward Roske inspires listeners with practical advice and tools to embrace technology confidently.
Tune in for strategies to thrive in the evolving financial landscape.
Asking Good Questions with Edward Roske
The Only Thing I've Been Seeing Is More Work | Filo Gonzalez
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Before Databricks he did Azure data and AI work at Microsoft, and before that he consulted for Avianca and for First Medical here in Puerto Rico. His whole approach starts with one line: "a prompt will give you a happy path all the time." It works right up until the live demo, when somebody asks "And what if…" and it falls over, and he says that's still true even if the LLM is AGI, unless the architecture gets better. So he builds an agent whose only job is asking the what-if questions, with a human in the loop.
He also calls it a conflict of interest when a model reviews its own work, which I felt personally, because I have Claude checking Claude (so I've been asking the same guy the same question three times in different agent hats, and I host a show called Asking Good Questions). His fix is making different models argue, inside Omnigent, an open-source project a Databricks team built, where an agent named Polly does the work and one named Debby debates it.
I asked how he gets a company from a vague concept of AI to a live AI app in two weeks, and his method was less magic than I expected (which is why I believe it): once the prerequisites are in place, the customer approves a requirements document and every task gets graded 1 to 10 (a 7 or up gets chopped again before an agent sees it). (Neither of us could say what PRD stands for, so I'm calling it a tie.) He's also seen the one where you fix the formula so the AI says 100, then the customer comes back with "Actually, it was 99, not 100."
Then I asked what he'd tell high schoolers, and he didn't do the reassuring adult thing: "The only thing I've been seeing so far is more work." In his read, companies that let their experts go are finding out it was a mistake, because now somebody has to read 60,000 lines of code. (I said The 4-Hour Workweek came out about 15 years ago. It was 19, which is close enough for a podcast.) My guess is that by next fall, two weeks is on every vendor slide in this business and almost nobody hits it, because they skipped the paperwork. If I'm wrong I'll eat the paperwork in public (I once promised to eat my hat, and paper is a lot easier to eat).
Filo is speaking at the Caribbean AI Summit, October 9-10 in San Juan, on "Ship a Production AI App in Two Weeks." I'll be in the room writing down the parts I've been doing wrong (come copy off me).
I'm Edward Roske, and this is Asking Good Questions. New conversations with the people figuring out the human side of AI. If one of these questions stuck with you, subscribe and send it to someone who'd take the other side.
LINKS
Caribbean AI Summit (Oct 9-10, San Juan): https://CaribbeanSummit.AI
Filo on LinkedIn: https://www.linkedin.com/in/filomeno-gonzalez/
Filo on GitHub: https://github.com/filomeno-gonzalez
Omnigent: https://github.com/omnigent-ai/omnigent
Databricks: https://www.databricks.com
BOOKS & RESEARCH MENTIONED
The 4-Hour Workweek, Tim Ferriss (2007)
"Data Scientist: The Sexiest Job of the 21st Century," Harvard Business Review (2012)
Dexter's Laboratory (Cartoon Network, 1996), and the other Dexter (Showtime, 2006)
Omnigent (Databricks, open source, 2026), home of Polly and Debby
Also mentioned: LangChain Deep Agents, ChatGPT, Claude Code, Gemini, Groq, Codex, GPT-5.6 Sol, Cursor, Pi, MCP, RAG, Genie Agents, Lakebase, Microsoft Copilot
This episode of Asking Good Questions is brought to you by Caprus.Ai.
Check how we empower the office of the CFO to harness the transformative potential of artificial intelligence (AI) and revolutionize their operations.
