TBPN Podcast Summary

Episode: Google’s AI Breakthrough in Cancer, Protein Powders Exposed | Marc Benioff, Eiso Kant, Dante Vaisbort, Alice Bentinck, Eric Seufert, Pim de Witte
Date: October 16, 2025
Hosts: John Coogan, Jordy Hayes
Guests: Marc Benioff (Salesforce), Eiso Kant (Poolside), Dante Vaisbort (Albacore), Alice Bentinck (Entrepreneurs First), Eric Seufert (Mobile Dev Memo), Pim de Witte (Agentic)

Overview

This lively episode covers major breakthroughs in biotech and AI, specifically Google’s use of an LLM to generate experimentally-validated cancer hypotheses, dramatic shifts in AI infrastructure and enterprise strategy, and the industry-wide race to apply foundation models to real-world applications. The show also dives into the protein powder lead scandal, the state of AI advertising, and hosts several vibrant guest interviews, including Salesforce CEO Marc Benioff and emerging founders from defense and AI gaming.

Key Discussion Points & Insights

1. Google’s AI Breakthrough in Cancer Research

Segment: [00:00–16:36]

Key News: Google, with Yale, used their new 27B-parameter C2S (Cell2Sentence) model, based on Gemma, to generate novel hypotheses about cancer cell behavior, experimentally validated in living cells.
Metaphor: Coogan uses "Call of Duty" to explain: immune cells are like players, tumors are enemies, and drugs can be like popping a UAV to reveal “cold” (invisible) tumors to the immune system ([02:31]).
Significance: It’s the first time an LLM has directly hypothesized something new and had that hypothesis validated in vitro—a landmark but early-stage result.
Perspective: Comparison to past AI milestones (e.g., AlphaFold) and caution: does this mean “ChatGPT moment” for bio, or is it more like the 2005 DARPA Grand Challenge—a breakthrough that still needs years to commercialize?
Scaling: Scaling laws apply! As these models grow (from 4B → 27B → trillion+ parameters), they will generate more such scientific insights; the gating factor may be data, regulation, and FDA cycle time, not compute ([10:42], [14:56]).

Quotes:

“Now, just to put it in perspective… AI got one point on the board and humanity has tens of thousands of these discoveries.” – John Coogan [04:29]
“The bear case for a fast takeoff is the drug approval cycle—not the model. It could take 20 years before you see these discoveries in patients.” – John Coogan [12:39]
“I think this is just good comms. It’s good PR. Every foundation lab should be competing in that.” – John Coogan [07:11]
“It’s a vindication of Rune’s concept: text is the universal interface—even for bio.” – John Coogan [08:00]

2. Protein Powders Lead Scandal and Food Safety

Segment: [30:38–35:41]

News: Consumer Reports found high levels of lead in top-selling protein supplements, especially “Huel Black.” While none exceeded US legal limits, it’s a reminder many processed foods (incl. “healthy” ones like dark chocolate) can contain lead due to supply chain complexity.
Context: Previously, microplastics were the health concern du jour; now attention returns to lead, with “certified” vs “legal” levels discussed.
Industry Reaction: Hosts emphasize that many such problems originate upstream in the supply chain and can impact major food-tech companies (Soylent example).

Quotes:

“You can do these for almost every category of food and get some pretty shocking results.” – Jordy Hayes [33:28]
“Lead is like, ideally, you want it to be zero… but you can legally have quite a significant amount in food in America.” – Jordy Hayes [32:41]

3. The State of AI Agents & Enterprise Competition (Salesforce, Poolside, etc.)

Segments: [118:36–132:23] (Poolside/Isso Kant)

Guest: Eiso Kant, Poolside – building foundation models and massive (250+ MW, soon GW-scale) data centers; focus on vertical integration from energy → compute → intelligence.
Go-to-Market: Poolside started with defense/government vertical due to strict deployment requirements; now expanding into broader enterprise.
Infra Arms Race: Compute availability is now the main bottleneck for scaling models—getting 40,000 H100s is nearly impossible unless you co-develop with infra companies like Coreweave.
Verticalization: Poolside’s approach is to own the stack and reduce lead time from when you need compute to when you get it, including novel “prefab” data center construction.

Quotes:

“The true bottleneck in our industry is not chips and it’s not energy… it's getting powered shells/data centers online.” – Eiso Kant [126:20]
“We treat intelligence as a commodity. But you don’t want to be in a commodity business… You want to help customers grow revenue or cut costs.” – Eiso Kant [129:03]
“You can build the world’s most capable model, but if you can’t serve it, scale it up further, or train the next generation—you’re just cosplaying.” – Eiso Kant [127:17]

4. Salesforce, Slack, & AI in Enterprise (Marc Benioff Interview)

Segment: [132:27–156:42]

Guest: Marc Benioff, Salesforce CEO—charismatic, energetic interviewee.
AI Strategy: Salesforce's “Agentic Enterprise” will provide a platform where customers can pick from a range of models (Claude, OpenAI, Gemini, Mistral, local/vertical models, etc.), focused on compliance and data residency.
On AI Acquisitions: Salesforce continues to be extremely active—100+ companies acquired, 1% of Anthropic, large investments.
Pricing: Shift from simple seat-based pricing to sophisticated, custom agreements per client due to diversity of needs.
On Sales as a Career: Strongly pro-sales, sees interpersonal (human) connection as essential in a world of AI.

Quotes:

“Some customers want our models. Some want to build their own. Some like a certain brand. Some countries like a certain model. We have to give people choice.” – Marc Benioff [139:32]
“I think sales… listening, asking good questions, empathizing with the customer… having fun—isn’t that what sales is all about?” – Marc Benioff [153:33]
“If you are a Slack user, you need to see the Slack bot we introduced here, built on Anthropic. It’s amazing.” – Marc Benioff [137:34]

5. The Coming AI Ads Revolution (Eric Seufert Interview)

Segment: [163:30–186:55]

Guest: Eric Seufert, Mobile Dev Memo (ads/monetization expert)
Prediction: “OpenAI will run ads. It’s inevitable, ineluctable. They already hired Meta/FB ads people.”
Monetization Math: OpenAI’s ARPU is ~$11.38 (global, free+paid), well below Meta. Free-tier inference is expensive—ads offset it.
Challenge: OpenAI must not cross the “bright line” between product recommendations that feel trustworthy and those polluted by ads—user trust is fragile in conversational interfaces.
Format: Ads need NOT be anchored to chat context; can be based on purchase history, just like Instagram/Facebook. Native insertion will be via App Store integrations (“directories always lead to ads”).

Quotes:

“I wrote: ‘Obviously, OpenAI will monetize with ads.’ … They already hired people from Meta. Once you have a directory, what always follows? Ads.” – Eric Seufert [164:42], [176:44]
“Once you lose consumer trust, it’s gone forever. The answer must be objective, not just whoever paid most.” – Eric Seufert [171:38]
“Ads are a beating heart of the economy. Everyone should be excited—it's a new opportunity for businesses.” – Eric Seufert [183:34]

6. Startups & Venture: Demo Days and Capital Flow

Segment: [157:28–163:14]

Guest: Alice Bentinck, Entrepreneurs First
Demo Day: 20 new companies, with notable themes being: very young founders targeting old-line, “boring but lucrative” industries with AI.
Metrics: Investors are looking for sticky revenue, not just quick $; some companies had $100K+ revenue in under three months.
Selection: EF “finds” people, tests them through hackathons for speed of execution and ability to create fast.

Quotes:

“We want to be the CAA for founders—find the talent, not just wait for applications.” – Alice Bentinck [161:54]

7. Defense Tech and Agentic Robotics

Segments: [96:00–104:44] (Dante Vaisbort, Albacore) & [187:16–194:27] (Pim de Witte, Agentic)

Albacore: Building deployable undersea “loitering munitions” (robotic drones w/ 250lb payload), with long endurance, for military use. Focus: out-innovate larger defense contractors with agility.
- “Our drone can loiter for 30 days, 1,000 nautical miles, and carry enough explosives to do real damage.” – Dante Vaisbort [100:07]
Agentic: $133.7M raised for general foundation models trained on gaming data and game-controller interfaces for "deep spatial/temporal reasoning".
- Applications range from drone swarms for search-and-rescue to next-gen game AI, and eventually real-world robots.
- “Maybe simulation will be bigger than the physical world; there’s only one physical reality, but many simulated ones.” – Pim de Witte [191:02]

8. Miscellaneous:

Lead in Food & Nutrition: [30:38–35:41]
Trading Cards/Podcast Merch: Discussion of making TBPN rookie cards for seed stage founders ([21:12]).
Misc Technology & Market News:

Paxos stablecoin minting error ([55:07]), Mr. Beast in fintech ([29:08]), Starlink on United flights ([71:08]), and OpenAI + TikTok, Spotify, Netflix, and YouTube “anti-coalition” analysis ([179:34]).

Notable Quotes & Moments

“Is this the ChatGPT moment for AI in bio, or is it the 2005 DARPA Grand Challenge?” – John Coogan [12:39]
“All the roads end in ads.” – Eric Seufert [177:03]
Marc Benioff’s energy and playful braggadocio, e.g., “You’re still unforgiven! … Are you a Metallica fan or not?” [Various, 136:24–154:46]
On Sales Careers: “People want to buy from people. AI doesn’t have a soul. … That’s why sales is the most valuable skill in the world.” – Marc Benioff [153:33], [151:36]
“Once you lose trust, it’s gone forever. That’s the risk with ads in chat.” – Eric Seufert [171:38]

Timestamps for Key Segments

| Topic | Timestamp Start | |---------------------------------------------|----------------| | Google's AI Cancer Breakthrough | 00:00 | | Lead in Protein Powders/Food Safety | 30:38 | | Poolside/AI Foundation Models Infra | 118:36 | | Marc Benioff (Salesforce) Interview | 132:27 | | Entrepreneurs First Demo Day (Alice) | 157:28 | | Eric Seufert on Ads in AI | 163:30 | | Defense AI (Dante Vaisbort, Albacore) | 96:00 | | Robotics/Game AI (Pim de Witte, Agentic) | 187:16 |

The Vibe & Takeaways

Language & Tone: Energetic, witty, open to metaphor (Coogan’s Call of Duty/cancer analogy), tongue-in-cheek (TBPN trading cards; joking about AGI “choosing Salesforce and Slack” [137:21]), but still detail-oriented.
Industry Mood: Exhilarated but not delusional—celebratory around AI progress, while recognizing real-world friction in deployment (esp. regulation, data, enterprise constraints).
Recurring Themes: The gap between technical breakthroughs and real-world impact; inevitability of ads in consumer AI; necessity of vertical integration as foundation models commoditize.

Recommendation:
Listen to this episode for:

Deep insight on the intersection of AI, biotech, and enterprise
Candid, unscripted interviews with the industry’s biggest personalities
Thoughtful discussion of how scientific & business-side AI challenges are converging

Produced by:
TBPN - Technology’s Daily Show, live on X and YouTube.

TBPN Podcast Summary

Overview

Key Discussion Points & Insights

1. Google’s AI Breakthrough in Cancer Research

Segment: [00:00–16:36]

Key News: Google, with Yale, used their new 27B-parameter C2S (Cell2Sentence) model, based on Gemma, to generate novel hypotheses about cancer cell behavior, experimentally validated in living cells.
Metaphor: Coogan uses "Call of Duty" to explain: immune cells are like players, tumors are enemies, and drugs can be like popping a UAV to reveal “cold” (invisible) tumors to the immune system ([02:31]).
Significance: It’s the first time an LLM has directly hypothesized something new and had that hypothesis validated in vitro—a landmark but early-stage result.
Perspective: Comparison to past AI milestones (e.g., AlphaFold) and caution: does this mean “ChatGPT moment” for bio, or is it more like the 2005 DARPA Grand Challenge—a breakthrough that still needs years to commercialize?
Scaling: Scaling laws apply! As these models grow (from 4B → 27B → trillion+ parameters), they will generate more such scientific insights; the gating factor may be data, regulation, and FDA cycle time, not compute ([10:42], [14:56]).

Quotes:

“Now, just to put it in perspective… AI got one point on the board and humanity has tens of thousands of these discoveries.” – John Coogan [04:29]
“The bear case for a fast takeoff is the drug approval cycle—not the model. It could take 20 years before you see these discoveries in patients.” – John Coogan [12:39]
“I think this is just good comms. It’s good PR. Every foundation lab should be competing in that.” – John Coogan [07:11]
“It’s a vindication of Rune’s concept: text is the universal interface—even for bio.” – John Coogan [08:00]

2. Protein Powders Lead Scandal and Food Safety

Segment: [30:38–35:41]

News: Consumer Reports found high levels of lead in top-selling protein supplements, especially “Huel Black.” While none exceeded US legal limits, it’s a reminder many processed foods (incl. “healthy” ones like dark chocolate) can contain lead due to supply chain complexity.
Context: Previously, microplastics were the health concern du jour; now attention returns to lead, with “certified” vs “legal” levels discussed.
Industry Reaction: Hosts emphasize that many such problems originate upstream in the supply chain and can impact major food-tech companies (Soylent example).

Quotes:

“You can do these for almost every category of food and get some pretty shocking results.” – Jordy Hayes [33:28]
“Lead is like, ideally, you want it to be zero… but you can legally have quite a significant amount in food in America.” – Jordy Hayes [32:41]

3. The State of AI Agents & Enterprise Competition (Salesforce, Poolside, etc.)

Segments: [118:36–132:23] (Poolside/Isso Kant)

Guest: Eiso Kant, Poolside – building foundation models and massive (250+ MW, soon GW-scale) data centers; focus on vertical integration from energy → compute → intelligence.
Go-to-Market: Poolside started with defense/government vertical due to strict deployment requirements; now expanding into broader enterprise.
Infra Arms Race: Compute availability is now the main bottleneck for scaling models—getting 40,000 H100s is nearly impossible unless you co-develop with infra companies like Coreweave.
Verticalization: Poolside’s approach is to own the stack and reduce lead time from when you need compute to when you get it, including novel “prefab” data center construction.

Quotes:

“The true bottleneck in our industry is not chips and it’s not energy… it's getting powered shells/data centers online.” – Eiso Kant [126:20]
“We treat intelligence as a commodity. But you don’t want to be in a commodity business… You want to help customers grow revenue or cut costs.” – Eiso Kant [129:03]
“You can build the world’s most capable model, but if you can’t serve it, scale it up further, or train the next generation—you’re just cosplaying.” – Eiso Kant [127:17]

4. Salesforce, Slack, & AI in Enterprise (Marc Benioff Interview)

Segment: [132:27–156:42]

Guest: Marc Benioff, Salesforce CEO—charismatic, energetic interviewee.
AI Strategy: Salesforce's “Agentic Enterprise” will provide a platform where customers can pick from a range of models (Claude, OpenAI, Gemini, Mistral, local/vertical models, etc.), focused on compliance and data residency.
On AI Acquisitions: Salesforce continues to be extremely active—100+ companies acquired, 1% of Anthropic, large investments.
Pricing: Shift from simple seat-based pricing to sophisticated, custom agreements per client due to diversity of needs.
On Sales as a Career: Strongly pro-sales, sees interpersonal (human) connection as essential in a world of AI.

Quotes:

“Some customers want our models. Some want to build their own. Some like a certain brand. Some countries like a certain model. We have to give people choice.” – Marc Benioff [139:32]
“I think sales… listening, asking good questions, empathizing with the customer… having fun—isn’t that what sales is all about?” – Marc Benioff [153:33]
“If you are a Slack user, you need to see the Slack bot we introduced here, built on Anthropic. It’s amazing.” – Marc Benioff [137:34]

5. The Coming AI Ads Revolution (Eric Seufert Interview)

Segment: [163:30–186:55]

Guest: Eric Seufert, Mobile Dev Memo (ads/monetization expert)
Prediction: “OpenAI will run ads. It’s inevitable, ineluctable. They already hired Meta/FB ads people.”
Monetization Math: OpenAI’s ARPU is ~$11.38 (global, free+paid), well below Meta. Free-tier inference is expensive—ads offset it.
Challenge: OpenAI must not cross the “bright line” between product recommendations that feel trustworthy and those polluted by ads—user trust is fragile in conversational interfaces.
Format: Ads need NOT be anchored to chat context; can be based on purchase history, just like Instagram/Facebook. Native insertion will be via App Store integrations (“directories always lead to ads”).

Quotes:

“I wrote: ‘Obviously, OpenAI will monetize with ads.’ … They already hired people from Meta. Once you have a directory, what always follows? Ads.” – Eric Seufert [164:42], [176:44]
“Once you lose consumer trust, it’s gone forever. The answer must be objective, not just whoever paid most.” – Eric Seufert [171:38]
“Ads are a beating heart of the economy. Everyone should be excited—it's a new opportunity for businesses.” – Eric Seufert [183:34]

6. Startups & Venture: Demo Days and Capital Flow

Segment: [157:28–163:14]

Guest: Alice Bentinck, Entrepreneurs First
Demo Day: 20 new companies, with notable themes being: very young founders targeting old-line, “boring but lucrative” industries with AI.
Metrics: Investors are looking for sticky revenue, not just quick $; some companies had $100K+ revenue in under three months.
Selection: EF “finds” people, tests them through hackathons for speed of execution and ability to create fast.

Quotes:

“We want to be the CAA for founders—find the talent, not just wait for applications.” – Alice Bentinck [161:54]

7. Defense Tech and Agentic Robotics

Segments: [96:00–104:44] (Dante Vaisbort, Albacore) & [187:16–194:27] (Pim de Witte, Agentic)

Albacore: Building deployable undersea “loitering munitions” (robotic drones w/ 250lb payload), with long endurance, for military use. Focus: out-innovate larger defense contractors with agility.
- “Our drone can loiter for 30 days, 1,000 nautical miles, and carry enough explosives to do real damage.” – Dante Vaisbort [100:07]
Agentic: $133.7M raised for general foundation models trained on gaming data and game-controller interfaces for "deep spatial/temporal reasoning".
- Applications range from drone swarms for search-and-rescue to next-gen game AI, and eventually real-world robots.
- “Maybe simulation will be bigger than the physical world; there’s only one physical reality, but many simulated ones.” – Pim de Witte [191:02]

8. Miscellaneous:

Lead in Food & Nutrition: [30:38–35:41]
Trading Cards/Podcast Merch: Discussion of making TBPN rookie cards for seed stage founders ([21:12]).
Misc Technology & Market News:

Paxos stablecoin minting error ([55:07]), Mr. Beast in fintech ([29:08]), Starlink on United flights ([71:08]), and OpenAI + TikTok, Spotify, Netflix, and YouTube “anti-coalition” analysis ([179:34]).

Notable Quotes & Moments

“Is this the ChatGPT moment for AI in bio, or is it the 2005 DARPA Grand Challenge?” – John Coogan [12:39]
“All the roads end in ads.” – Eric Seufert [177:03]
Marc Benioff’s energy and playful braggadocio, e.g., “You’re still unforgiven! … Are you a Metallica fan or not?” [Various, 136:24–154:46]
On Sales Careers: “People want to buy from people. AI doesn’t have a soul. … That’s why sales is the most valuable skill in the world.” – Marc Benioff [153:33], [151:36]
“Once you lose trust, it’s gone forever. That’s the risk with ads in chat.” – Eric Seufert [171:38]

Timestamps for Key Segments

The Vibe & Takeaways

Language & Tone: Energetic, witty, open to metaphor (Coogan’s Call of Duty/cancer analogy), tongue-in-cheek (TBPN trading cards; joking about AGI “choosing Salesforce and Slack” [137:21]), but still detail-oriented.
Industry Mood: Exhilarated but not delusional—celebratory around AI progress, while recognizing real-world friction in deployment (esp. regulation, data, enterprise constraints).
Recurring Themes: The gap between technical breakthroughs and real-world impact; inevitability of ads in consumer AI; necessity of vertical integration as foundation models commoditize.

Recommendation:
Listen to this episode for:

Deep insight on the intersection of AI, biotech, and enterprise
Candid, unscripted interviews with the industry’s biggest personalities
Thoughtful discussion of how scientific & business-side AI challenges are converging

Produced by:
TBPN - Technology’s Daily Show, live on X and YouTube.

wavePod

Google’s AI Breakthrough in Cancer, Protein Powders Exposed | Marc Benioff, Eiso Kant, Dante Vaisbort, Alice Bentinck, Eric Seufert, Pim de Witte

Powered by Wave AI

Summary

TBPN Podcast Summary

Overview

Key Discussion Points & Insights

1. Google’s AI Breakthrough in Cancer Research

2. Protein Powders Lead Scandal and Food Safety

3. The State of AI Agents & Enterprise Competition (Salesforce, Poolside, etc.)

4. Salesforce, Slack, & AI in Enterprise (Marc Benioff Interview)

5. The Coming AI Ads Revolution (Eric Seufert Interview)

6. Startups & Venture: Demo Days and Capital Flow

7. Defense Tech and Agentic Robotics

8. Miscellaneous:

Notable Quotes & Moments

Timestamps for Key Segments

The Vibe & Takeaways

Summary

TBPN Podcast Summary

Overview

Key Discussion Points & Insights

1. Google’s AI Breakthrough in Cancer Research

2. Protein Powders Lead Scandal and Food Safety

3. The State of AI Agents & Enterprise Competition (Salesforce, Poolside, etc.)

4. Salesforce, Slack, & AI in Enterprise (Marc Benioff Interview)

5. The Coming AI Ads Revolution (Eric Seufert Interview)

6. Startups & Venture: Demo Days and Capital Flow

7. Defense Tech and Agentic Robotics

8. Miscellaneous:

Notable Quotes & Moments

Timestamps for Key Segments

The Vibe & Takeaways