TBPN Episode Summary: "Mapping Neo Labs, Unlocking LLM Growth, Evan Spiegel Live in the Ultradome"
Date: February 18, 2026
Hosts: John Coogan & Jordi Hays
Special Guests: Tyler Cosgrove, Travis Brashears, Blake Dodge, Freddy deBoer, Sohel Prasad, Evan Spiegel
Episode Overview
This jam-packed live episode of TBPN covers the dynamic landscape of AI "labs" (from legacy giants to cutting-edge neolabs), practical improvements for consumer-facing LLM adoption, seismic changes in private tech investing, and a candid interview with Snap’s Evan Spiegel. The hosts, John and Jordi, blend deep-dive analysis with irreverent humor, drawing connecting lines between exploding AI categories, business models, and the sociopolitical impact of technology.
Key Segments & Insights
1. Mapping the World of AI Labs (00:41–25:55)
Guest: Tyler Cosgrove
-
Market Map of All Companies:
Tyler describes his project mapping every company with a Wikipedia article using Quen 3 embedding, visualizing clusters in 2D space based on article embeddings.- “On Wikipedia...there’s all sorts of cool things you can do.” — Tyler, 01:23
- Fun aside: The 2D map’s coincidental resemblance to the United States sparks banter.
-
Defining the "Neo Lab" Explosion
Tyler walks through the taxonomy of AI labs:- Trad Labs: OpenAI, DeepMind, Anthropic, Xai, Mistral—large, frontier pre-training labs.
- “They did a transformer-based pre-training run. They're not just fine-tuning.” — Jordi, 06:15
- Sovereign Labs: Non-US leaders (e.g., Mistral in Europe, Cohere in Canada), with mention of China’s Quen, DeepSeek, Kimi Unitree.
- Legacy Labs: Entrenched research orgs (Microsoft Research, Bell Labs, FAIR).
- Neo Labs: Recent, research-first labs, often spun out from trad labs (e.g., Prime Intellect, Thinking Machines, SSI).
- “At the core...still trying to find these new novel approaches. It’s research.” — Tyler, 09:19
- SaaS & Neo-SaaS Labs: Focus on enterprise LLMs; differentiated by depth of customization for business use cases.
- Post Labs: Labs building on frontier models, providing evaluation and tooling (Meter, Pangram).
- Safety Labs/Open Source Labs: Goodfire, Eleuther—focus on interpretability and open-source releases.
- Consumer Labs: Focused on direct impact; e.g., Eureka (Karpathy), Humans.
- Visual, Auditory, Kinetic, Robotics Labs: Spanning multimodal (Midjourney, Meta), voice (11 Labs), robotics (Figure, Unitree), and simulation.
- Neo-Trad Lab, Recursive Labs, Math Lab, Wet Lab, Dark Lab: Specialized subcategories reflecting single-moonshot research, recursive research, mathematical problem solving, biotech, defense/government, and simulations.
- “The lines are blurry, but these distinctions help make sense of the explosion in new research orgs.” — Tyler, 13:19
- Trad Labs: OpenAI, DeepMind, Anthropic, Xai, Mistral—large, frontier pre-training labs.
-
Funding and Market Growth:
The hosts estimate over $200 billion invested in "neo labs" alone, underscoring the sheer pace and capital intensity of the sector.- “It’s evolving so fast...these things are coming out every day.” — Tyler, 24:55
2. Exploding Consumer LLM Adoption: Five Obvious Fixes (25:55–39:45)
Hosts: John & Jordi
Jordi outlines five urgent, practical ways LLM chat interfaces should improve for mass adoption:
-
Cache Popular Responses for Instant Speed:
LLMs should cache responses to frequently asked questions to eliminate unnecessary GPU use and latency.- “It doesn’t need to light the GPUs on fire...Cache those results, give them to the user instantly.” — Jordi, 25:55
-
Model Routing, Not Model Selection:
Hide confusing model names. Instead, smart routing should interpret user intent and pick the right model. -
Integration of Ads: Emphasizes the necessity and historical inevitability of ads to drive product scale—debated in light of recent Perplexity pullback on ads.
- “Not only are ads the best way to deliver high quality products...they just make products better top to bottom.” — Jordi, 33:05
-
Advanced Agentic Coding:
Power users will want the ability to trigger advanced agentic, code-writing actions seamlessly, akin to OpenClaw or Codex. -
Better UI/UX Polishing:
A/B test everything—BUT claim taste, not A/B tests, to keep your job!- “If you run the AB test, don’t tell your boss. Tell them it was taste.” — Jordi, 39:45
3. Private Company Exposure: The New Retail Gold Rush (41:10–44:07, 125:10–136:22)
Guest: Sohel Prasad, Destiny
- Destiny offers retail investors real-time exposure to a basket of top private tech startups via a closed-end fund (DXYZ on NYSE).
- “Most people in the world that use all these tech companies—they have no way of owning that.” — Sohel, 125:30
- Destined to invest in large late-stage companies (e.g., SpaceX, Anthropic), but also up-and-coming unicorns, using both secondary markets and primary rounds.
Industry Questions Raised:
- How can funds balance early- vs late-stage exposure?
- Will closed-end funds always trade at a premium due to investor FOMO?
- What is the impact on company relationships and secondary markets?
- “Our goal...reflect the late stage venture-backed ecosystem, but use our discretion to buy at the right time and structure.” — Sohel, 133:24
4. The Great AI Economic Debate (106:58–124:56)
Guest: Freddy deBoer
- The "Motte-and-Bailey" Problem in AI:
AI boosters hype both apocalypse and modest improvements, but don’t commit to falsifiable predictions. - Challenging the AI Doom Narrative:
Freddy discusses his wager (proposed to Scott Alexander) against the prediction of mass unemployment or economic upheaval due to AI within a three-year period.- “CEO of Anthropic...said within a couple years, 50% of all jobs are going to be destroyed. That’s the thing that bothers me...” — Freddy, 111:53
- If the “real stuff” (AGI-level disruption) happens, “you’re not gonna have to convince me...the effects will be so profound.”
- Counterpoint: The Internet and mobile did reshape society, but did not trigger the economic change many expected.
- “AI will be meaningful. Eliminate some jobs. But history regresses to the mean.” — Freddy, 118:48
5. Interview: Evan Spiegel, Co-Founder & CEO of Snap (148:12–187:02)
Host: John & Jordi
-
$1B in Direct Revenue, 25M Subscribers:
Snap reaches “next stop Hulu” scale for direct subscriptions, adding features for power users and monetizing cloud storage for memories.- “We’ve reached a billion dollar annual run rate...25 million subscribers. Next stop Hulu.” — Evan, 152:44
-
Product Velocity and Taste:
Snap’s design team is becoming engineers, blurring the traditional lines of ideation and iteration.- “Now the hard part is having a great idea. I think taste is important.” — Evan, 149:18
-
AI Deployment in Consumer Apps:
Lens Studio and Easy Lens empower both users and developers to create AR and GenAI-powered experiences.- “700 million people used generative AI lenses last quarter.” — Evan, 155:46
- Growing use of prompt-based lens creation. Authentic, casual UGC outperforms heavy production.
-
Enterprise Agentic Engineering:
Internally, Snap is rolling out agents for code debugging and sales support, embracing “agentic engineering.”- “There are team members who are essentially not writing code anymore.” — Evan, 158:57
-
Hardware (Specs) & the Death of Software Moats:
Snap spins out glasses (Specs) as a standalone company, betting on hardware as the next competitive differentiator.- “Software isn’t a moat anymore...it’s so easy to build software.” — Evan, 164:11
-
Industry, State, and Society:
Evan is optimistic about growing awareness of California’s challenges:- “Number one in homelessness, poverty, unemployment. But awareness is building, and in a democracy, that’s how you get change.” — Evan, 175:24
-
On VR as a Movie/Work Platform:
Evan laughs at the concept of watching movies in VR headsets now but is bullish on lightweight glasses for immersive content and productivity in the near future.- “You’re going to watch movies in glasses for sure...If you’re traveling or on a plane, you want your set-up.” — Evan, 184:01
Memorable Moment:
Jordi brags about being the first to watch an entire movie in Vision Pro VR:
- “You watched the whole Matrix in VR? No way.” — Evan, 182:31
(Laughter erupts as Evan and the hosts poke fun at the impractical but pioneering usage.)
Notable Quotes & Timestamps
-
On the AI Lab Map:
"The lines are blurry, but these distinctions help make sense of the explosion in new research orgs."
— Tyler Cosgrove (13:19) -
On Caching LLM Results:
“It doesn’t need to light the GPUs on fire...Cache those results, give them to the user instantly.”
— Jordi (25:55) -
On Taste & Design vs. Data:
"If you run the AB test, don’t tell your boss. Tell them it was taste."
— Jordi (39:45) -
On the AI Economy:
"It cannot simultaneously be true that we are imminently facing a replacement of all jobs...but also, 18% unemployment is extravagant for this bet."
— Freddy deBoer (112:54) -
On VR & Hardware’s Future:
“Software isn’t a moat anymore...it’s so easy to build software.”
— Evan Spiegel (164:11) -
On the California Tech Exodus:
“The folks behind the ballot measure have not acknowledged that people are leaving. They call wealth flight a myth, which is crazy.”
— Blake Dodge (99:49)
Episode Vibe
The show blends technical depth (AI, infra, LLMs), business model breakdowns (labs, funding, investing), and high-level societal commentary—all in a playful, sometimes satirical, always Silicon Valley-forward atmosphere. Humor and big ideas fly, but each segment is grounded by sharp and specific examples. The interview with Evan Spiegel brings it home with practical lessons on scaling users, deploying AI, and the importance of taste in product creation.
Listen If...
- You want to understand where the AI lab boom is headed and what it means for startups and incumbents
- You’re building or investing in consumer LLM apps
- You're following the clash of public/private tech investing, or the regulatory climate in California
- You enjoy candid, expert interviews with tech founders (Evan Spiegel!)
[Episode available on Spotify and other platforms. TBPN streams live weekdays 11–2 PT on X and YouTube.]
