Podcast Summary: "How 3 CEOs Use AI to Run $10B in Companies"

Podcast: This Week in AI
Host: Jason Calacanis
Guests:

• Jeremy Frankel (CEO, Fundamental)
• Victor Riparbelli (CEO, Synthesia)
• Nick Harris (CEO, Lightmatter)
  Date: April 1, 2026

Overview

This episode of "This Week in AI" features a deep dive into how three leading CEOs are leveraging AI to transform operations at billion-dollar companies. Host Jason Calacanis moderates an expert panel including Jeremy Frankel (Fundamental), Victor Riparbelli (Synthesia), and Nick Harris (Lightmatter). The discussion spans enterprise AI deployment, hardware and data center innovation, the evolution of video and interactive content, AI’s impact on jobs, and the looming potential (and risks) of AGI.

Key Discussion Points & Insights

1. AI for Enterprise Data: The "ChatGPT Moment" for Structured Data

(02:06–06:26) Jeremy Frankel on Fundamental & Large Tabular Models

• Context: Fundamental builds foundation models for tabular (structured) data, aiming to do for enterprise tables what LLMs (like ChatGPT) did for text.
• Technical Insight: Structured data (spreadsheets, databases) hasn’t had a "ChatGPT moment". LLMs work for unstructured data, but not for the deterministic needs of tables.
  • “You shouldn't expect or want a different output when you change column order in a table—LLMs don't handle this well, and that’s our focus.” (03:58, Jeremy Frankel)
• Term Clarified: Fundamental builds LTMs (Large Tabular Models), not LLMs—different architecture, optimized for prediction at scale.
• Impact: Promises higher fidelity, trust, and accuracy in predictions for applications like fraud detection, demand forecasting, etc., unifying use-cases that used to require separate machine learning models.
  • “What we’ve built is a model that can essentially unify all of those use cases into one model to allow you to make much more accurate predictions.” (05:36, Jeremy Frankel)

2. Video AI at Scale: Synthesia’s B2B Focus & the Shift from Consumption to Interaction

(07:28–08:56, 18:05–22:14) Victor Riparbelli on AI Video

• Business Model: Instead of chasing Hollywood or advertising, Synthesia took aim at PowerPoint users—enabling easy transition from slides to video.
  • “We essentially provide a way for PowerPoint creators to easily switch to making video instead. That has worked really, really well.” (07:28, Victor Riparbelli)
• Product Evolution: From high-fidelity avatar video to soon-launching real-time, interactive video.
• Why Not Go Broad?: Lesson in focus—OpenAI dropped its Sora video model to concentrate on codegen because B2B and code generation is where near-term value is highest.
  • “OpenAI had a flying-too-close-to-the-sun moment—trying to do absolutely everything. Anthropic focused on codegen B2B, no freemium. It’s clearly paid off.” (09:30, Victor Riparbelli)
• Use-Case Expansion: Real-time, interactive video, e.g., AI role-play for sales training, is bandwidth- and inference-intensive, requiring new data center capabilities.

3. Data Center Bottlenecks: Moving from Copper to Photonics

(11:55–17:42) Nick Harris on Lightmatter’s Photonic Hardware

• State of the Art: Networking, not compute, is the main bottleneck in scaling AI. Traditional copper cables are nearing their physical limits.
• Photonics Revolution: Lightmatter’s chips use glass/fiber optics, offering huge bandwidth—e.g., their M1000 chip equals the bandwidth of North America–Europe undersea cables.
  • “We just announced a chip with Qualcomm…pushing 1.6 terabits over a single optical fiber—that’s 1600 houses’ worth of internet.” (14:35, Nick Harris)
• Strategic Impact: Optical networking enables supercomputers with thousands of GPUs acting as "a single brain", drastically improving training and inference for AI workloads.
  • “You can build giant systems that act like a single brain rather than a bunch of mini brains.” (15:10, Nick Harris)

4. Evolving Enterprise Stack: Custom Software & the "Vibe Coding" Phenomenon

(36:16–45:23) All – Customization, Verification, and AI Stack Choices

• Agent-Built Software: Small startups increasingly "vibe code" (auto-generate) custom CRM and workflow tools using AI agents.
  • “We've built our own system called Fetch, integrated in Slack—it’s our CRM; you can ask any question and get any answer.” (38:35, Jeremy Frankel)
• Skepticism and Caution: For many, off-the-shelf software still makes sense (cost, focus, reliability)—but DIY is accelerating for smaller teams or non-critical systems.
  • “One of the lowest EV things you can do as a founder is to try to replicate a CRM system.” (38:01, Victor Riparbelli)
• Verification Challenge: The main difficulty is ensuring these AI-generated systems are reliable and safe, which is crucial for core business processes.
  • “The main problem with this is your CRM is so important… verification is everything.” (39:14, Nick Harris)
• Rapid Platform Shifts: Teams are finding it hard to keep up with fast upgrades from Claude, Perplexity, OpenClaw, etc.—every few weeks a platform leapfrogs the others.
  • “It's becoming disorienting for my team… did this big investment in OpenClaw, now people prefer Claude Code, Perplexity’s computer is trouncing everybody…” (41:00, Jason Calacanis)

5. Omnipresent CEOs: AI for Leadership Leverage

(48:29–55:22) How AI Agents Are Transforming the CEO Role

• Omniscience as a Leadership Tool: CEOs describe using LLM agents to summarize emails, Slack, and meetings, improving span of control, and organizational awareness.
  • “I've always got an Anthropic Claude session up…it’s summarizing emails, slack, everything you said, Jason… I feel like you can taste the singularity at this point.” (48:43, Nick Harris)
• Decision Tracking: Custom LLM "change log" agents monitor organizational decisions for transparency and accountability.
  • “Building an executive change log…decisions made today in which team…used to try to read every Slack message…now this helps scale that transparency.” (53:12, Victor Riparbelli)
• Tuning Productivity: AI tools (like Whisper Flow speech-to-text with foot pedals for dictation) and devices (Plod pins for recording/transcribing walks) help leaders capture and process vast information flows.
  • “Whisper Flow is like a genius editor…the greatest executive assistant ever.” (56:45, Jason Calacanis)

6. AGI: Arrival, Moving Goalposts & Societal Impact

(60:18–70:33) The AGI Debate—Have We Arrived, and at What Cost?

• Is AGI Here Already?: Some, like Jason and many AI leaders, believe models have crossed the AGI threshold, but society hasn’t recognized or deployed it yet.
  • “It feels like to me that AGI has been achieved and we just haven't deployed it yet.” (59:50, Jason Calacanis)
• Moving Goalposts: "AGI" is hard to define—if you showed 2026-level AI to anyone 10 years ago, they’d call it AGI. Yet definition keeps moving.
  • “If I showed you what we have today, ten years ago, you would have definitely said it's AGI. But now we're constantly moving the definition.” (63:24, Jeremy Frankel)
• Societal Concerns & Polls: While workplace automation fears are rising (70% of Americans think AI will reduce jobs), only 30% worry personally.
  • “They all think it’s happening to somebody else.” (00:12, Nick Harris)
• Future of Work: Optimism prevails among the panel—AI may shift high-value work toward creativity, hospitality, experiences; but there’s risk of ‘Future Shock’ and societal disruption if change is too rapid.
  • “A lot of the value that will accrue … is going to go more towards things we enjoy more… live music, restaurants… less on software and finance.” (67:12, Victor Riparbelli)
• The Knowledge Gap: Most people can’t see the wave coming, as appreciation for AI’s power is highly dependent on individuals’ skill at asking questions and managing agents.
  • “Most people are not very good at asking questions. When you’re not good at asking questions, it’s hard to see the value in these things.” (62:27, Nick Harris)

Notable Quotes & Memorable Moments

• “You can taste the singularity at this point. I can't even imagine. The end of this year is going to be shocking.”
  — Nick Harris (00:28, echoed 48:43)
• “What we've built is a model that can essentially unify all of those use cases into one model to allow you to make much more accurate predictions.”
  — Jeremy Frankel (05:36)
• “Anthropic focused on no voice models, no video models, just Codegen B2B, no freemium. That has clearly paid off.”
  — Victor Riparbelli (09:30)
• “Building chips with Lightmatter tech, you don’t need to put all the GPUs in a rack—you can separate them by a kilometer, connected by light, acting as one brain.”
  — Nick Harris (15:10)
• “If you showed this 50 years back, you’d almost be burned at the stake… people would definitely say this is AGI.”
  — Victor Riparbelli (67:12)
• “We're able to so clearly put a price tag on how much a piece of software is worth. That’s really weird because we can.”
  — Nick Harris (48:29)
• “If AI and this new hardware layer make it affordable, who knows what you could find in that data. It really is great.”
  — Jason Calacanis (30:14)
• “I used to always read every single Slack message in the entire company. At 650 people, that got pretty challenging. Now I use AI to scale that.”
  — Victor Riparbelli (53:12)
• “The difference with previous revolutions is, for the first time, we're really automating cognition, not just physical tasks.”
  — Jeremy Frankel (63:24)

Timestamps for Key Segments

• [02:06] - Jeremy Frankel on why tabular data needs its own AI revolution
• [07:28] - Synthesia’s approach to video AI and market focus
• [09:30] - The importance of focus (OpenAI/Anthropic contrast)
• [11:55] - Data center trends; photonics vs. copper (Lightmatter)
• [15:10] - Giant “brains”—AI supercomputers and photonics
• [18:05] - Interactive video and implications for bandwidth
• [38:35] - Building custom CRM/agents vs. buying SaaS (SMB & enterprise lens)
• [48:43] - CEOs on AI for inbox summarization, omnipresence
• [53:12] - Executive "change log": AI for decision tracking
• [59:50] - Has AGI arrived? Panel weighs in on the definition gap
• [62:27] - Why most of the world isn’t feeling the AI wave
• [67:12] - Future work: creativity, experiences & societal risks

Final Thoughts

This roundtable is a must-listen for anyone interested in the operational reality of AI at scale. The CEOs stress that while technology is breaking barriers, the human, economic, and societal shifts lag behind. They agree that leaders need to keep their organizations tightly tuned to change, leverage AI for insight and omnipresence, and that what counts as "AGI" may be less about flashy demos than about widespread, deeply integrated intelligence transforming how businesses and people operate—often in ways the public doesn’t yet see coming.