EP 089 · 55 MIN ANTHONY POMPLIANO WORKS FOR THE AI NOW

How an Investor in 300+ Companies Picks What to Build With AI

November 3, 2026 55 min Rory Flynn & Drew Brucker with Anthony Pompliano

The short version

Anthony Pompliano, chairman and CEO of Silvia, an AI CFO for independent investors, explains how Silvia is moving off frontier models: it launched as a ChatGPT wrapper, switched to Claude, then built its own harness, data pipelines, fine-tuned models and coding agent, which he says cuts token costs 97%. He also covers running 10 employees alongside dozens of agents, why speed and focus are his moats, and admitting AI is smarter than him.

Key takeaways

  1. 01

    Start from your own problem: Silvia began as Pompliano's month-by-month wealth spreadsheet that he wanted turned into a dashboard; the AI came later.

  2. 02

    Cut model calls before switching models: Silvia killed a feature nobody missed and stopped hitting the model on page refreshes.

  3. 03

    Pompliano's four-layer stack: a harness with a custom model router, proprietary data pipelines (indexed tax codes), fine-tuned models, and a coding agent for questions like Monte Carlo simulations.

  4. 04

    Put an agent in the support queue: Silvia's Slack agent Toby investigates a ticket, proposes fixes and can open a PR for human approval in under two minutes.

  5. 05

    Speed and focus are the moats: a 10-person team shipped four features in four days, and a vertical stack can out-answer general-purpose labs in its niche.

  6. 06

    Iterate, don't spec: Pompliano built his personal command center by mind-dumping a direction, using the rough build, and fixing it screenshot by screenshot.

  7. 07

    To tell real AI users from performers, ask what they've built that they're proudest of, not whether they've swapped Google for a chatbot.

Receipts · said on the record

“When we process tokens on, let's say, Claude, versus if we do it on our hardware and software, it's a 97% cost savings. It's like a no-brainer.”
Anthony Pompliano Moving Silvia off frontier models Play from 8:23
“the process of ticket being submitted to Toby being able to actually identify issues and potentially write the code, less than two minutes.”
Anthony Pompliano Silvia's support agent in Slack Play from 11:30
“the single biggest moat I personally think, speed. It is impossible to keep up with a company that is iterating quickly and continuing to improve their product.”
Anthony Pompliano What keeps the big labs from eating Silvia's lunch Play from 15:23
“There's something addictive about it when you get in there because you know it's yours and you know it all like comes from you.”
Rory Flynn Building the Fast Hours website from transcripts Play from 31:09
“people are working harder than ever, even though the technology is supposed to give you more free time.”
Anthony Pompliano The paradox of AI and free time Play from 40:13
“With AI, the equivalent to that was when I realized it's smarter than me. And I'm cool with that.”
Anthony Pompliano His zero-to-one moment with AI Play from 51:49

Chapters

  1. 0:00 Meet Anthony Pompliano
  2. 2:47 The Personal Problem Behind Silvia AI
  3. 4:46 The $550,000 AI Bill That Forced a Rethink
  4. 6:17 Inside Silvia's Custom AI Stack
  5. 9:56 10 People, Dozens of AI Agents
  6. 10:48 Toby: AI Support With Human Approval
  7. 13:23 Competing When Big AI Labs Copy Your Features
  8. 17:12 Building AI Products People Want to Keep Using
  9. 21:13 The AI Command Center Pomp Built for Himself
  10. 22:11 Building With AI Through Fast Iteration
  11. 23:29 Shipping Speed as an AI Investing Signal
  12. 25:16 Should You Build It or Keep Paying for Software?
  13. 26:38 The AI Projects Pomp Makes Time For
  14. 27:06 Turning Podcast Episodes Into Articles and Clips
  15. 32:24 Pomp's Late-Night AI Building Habit
  16. 33:16 Using Multiple AI Agents to Find Opportunities
  17. 36:28 Protecting Family Time When You Love Building
  18. 40:06 Why AI Saves Time and Creates More Work
  19. 40:36 Training an AI to Filter YouTube Comments
  20. 41:34 How Pomp Spots an AI Idea Worth Building
  21. 43:12 Spotting Real AI Adoption
  22. 47:43 When Expertise Gets in the Way of New Ideas
  23. 48:39 Asking AI What to Do Next
  24. 50:31 Giving AI the Context to Be More Creative
  25. 51:19 How AI Changed Pomp's Curiosity and Delegation
  26. 53:20 The Fast Hours Sign-Off

Questions this episode answers

Who is Anthony Pompliano?

Anthony Pompliano is chairman and CEO of Silvia, an AI CFO for independent investors, and hosts the Pomp Podcast. On Fast Hours Ep. 89 he explains how Silvia built its own AI stack and runs 10 employees alongside dozens of agents.

How did Silvia cut its AI costs by 97%?

Pompliano says Silvia first trimmed needless model calls, then shifted work from frontier models like Claude onto its own stack: a harness with a model router, indexed data pipelines, fine-tuned models and a custom coding agent. He puts the saving per token at 97%, with more accurate, faster answers.

What is Toby, Silvia's support agent?

Toby is an agent in Silvia's support Slack channel. Pompliano says it investigates and recreates each ticket, identifies fixes and can write the code, then opens a PR or sends it to a human engineer, all in under two minutes.

What does Anthony Pompliano look for in an AI product?

Pompliano values shipping velocity over any feature list, and asks what motivated the builder: people solving a genuinely painful problem of their own are the ones who see it through.

Did Fast Hours talk about building a podcast website with AI?

Yes. Pompliano built pomppodcast.com in two or three hours on a weekend, turning each episode's topics into articles with embedded clips. Drew and Rory had launched the Fast Hours site two weeks earlier with a very similar approach.

Show notes

Anthony Pompliano chooses AI projects around problems he needs solved, friction worth fixing, and builds he enjoys.

After investing in 300+ companies, Pomp still builds software because something in his own day annoys him. He joins Drew Brucker and Rory Flynn to explain when an AI idea earns his time, when paying for existing software makes more sense, and why a weekend experiment can be valuable just for the practice. The hosts recognize the habit of turning every free evening into another build (their podcast website gives them away).

Anthony Pompliano covers AI project selection, build versus buy decisions, and rapid product iteration. Silvia, ChatGPT, Claude, custom AI models, and Slack support agents supply concrete examples. Personal AI command centers, podcast repurposing, YouTube comment filtering, and creative context connect AI development to everyday work.

Every word Read the full transcript

If you liked this one