"It's so over?" Emmett Shear, Byron Deeter, Brendan Foody, Alex Heath, Nico Rosberg
- 01The AI 100 reveals a super-cycle that dwarfs cloud, and the entire value is still private
- 02Inference is the early, under-appreciated exponential
- 03Software moats are eroding; physical deployment and hardware are the new defensibility
- 04The bottleneck moves from doing work to verifying work
- 05What humans are for: expertise commoditizes, so purpose migrates
- 06Seed-stage founder quality: recruiting, audacity, and agency
1. Key Themes
The AI 100 reveals a super-cycle that dwarfs cloud, and the entire value is still private
Byron Deeter of Bessemer framed the scale of the new AI 100 list against its predecessor, the Cloud 100, which he ran for a decade. The comparison is the cleanest sizing of the AI wave in the episode: the whole cloud cycle produced a trillion dollars of list value, while AI does 3.6x that in year one. He also noted that none of these companies can be bought in public markets.
- "This list in its debut year is 3.6 trillion. And so it's a super cycle that really is the sum total of the internet wave and the cloud computing wave combined. We're now monetizing compute and that is an order of magnitude larger in terms of market size that everyone's going after." — Byron Deeter 00:15:10
- "The average valuation on that list is 36 billion." — Byron Deeter 00:14:17
- "20% of the list is less than two years old. It's literally zero to billions in value, um, in less than 24 months." — Byron Deeter 00:14:17
- "None of them are public it. There is not a pure play software AI company to trade on in the public market. So the entire innovation has happened in the private world." — Byron Deeter 00:16:01
Inference is the early, under-appreciated exponential
Alex Heath (now a VC at Sound Ventures) argued that people repeat the mistake they made on foundation models: failing to think exponentially. Nico Rosberg, an Etched investor, independently pointed to the same category.
- "The amount of AI inference that's actually out in the world is, I just had this CEO of AWS actually on the podcast telling me it's like not even 1% of what he thinks it will be." — Alex Heath 00:18:23
- "No one thought Anthropic would be IPOing at over a trillion, right? Three years ago. And the same way I think we'll be talking about inference in a few years." — Alex Heath 00:18:47
- "Inference is going to be an enormous kind of infinite market. It feels like at the moment. And they are really at the frontier in terms of delivering the infrastructure." — Nico Rosberg on Etched 00:33:36
Software moats are eroding; physical deployment and hardware are the new defensibility
Ethan Barajas of Icarus (space robotics) repeatedly returned to the idea that cheap software shifts defensibility toward real-world deployments, a view he credits to Owen Hinchey of Teens.
- "It points out that software is quite cheap now. Software, that moat, is not there as much as we thought. And the moat really comes from hardware and actual physical deployments." — Ethan Barajas 00:23:26
- "[Owen Hinchey's] belief is, well, it's all about deployments. It's all about working in the real world... How do you do the things that a model can't take away?" — Ethan Barajas 00:27:25
- Brendan Foody echoed the physical-world angle on jobs: "It feels like there's so many things in the physical world and intrinsically human tasks that will take an extremely long time to automate." 00:47:12
The bottleneck moves from doing work to verifying work
Alex Garden of Nima argues that the cost of work approaching zero doesn't capture value until verification costs collapse too.
- "We took a bottleneck that used to exist where work was and we moved it forward to verification. And the reason for that is that most people who work with AI know that it does things that are wrong all the time." — Alex Garden 00:36:17
- "When you can mathematically, cryptographically prove that the work that you expected to be done was done... then the way you run the business, the way you design, the way you imagine things totally changes." — Alex Garden 00:38:29
- "I'm seeing the barriers between imagination and outcome shrink literally in real time." — Alex Garden 00:38:29
What humans are for: expertise commoditizes, so purpose migrates
Two voices converged from different angles. Emmett Shear said reasoning, vocabulary and programming are commoditized. James Detweiler of Felicis described where scarcity moves next.
- "We've commoditized reasoning. We've commoditized obviously first like vocabulary, like language, and we commoditized reasoning... So we've commoditized programming. And what kind of intelligence is left?" — Emmett Shear 00:01:01
- "Once expertise becomes something a model can generate, the bottleneck to purpose shifts to something else. And I think that something else is imagination, curiosity. People who are out of distribution thinkers will get a premium." — James Detweiler 00:41:07
- "The bottleneck will shift from expertise to trustworthiness, to agency, to imagination and locality." — James Detweiler 00:42:12
Seed-stage founder quality: recruiting, audacity, and agency
Alex Heath and James Detweiler described the founder traits that matter now.
- "The ability to recruit early is a huge signal for me." — Alex Heath 00:20:30
- "An overly audacious, like kind of crazy, but like it just may work energy that is really magical that is very rare, actually." — Alex Heath 00:20:58
- "If you can use Claude Code and Stripe Atlas and make a company in an afternoon, you just can will things into existence. Agency is getting a premium." — James Detweiler 00:41:43
- Brendan Foody: "AI is changing so quickly where that is creating the perfect market conditions for new founders because incumbents are struggling to keep up." 00:48:11
Market shock as a stress test: the Jensen/NVIDIA news and the Anthropic IPO
The episode is recorded against a backdrop of Jensen Huang's departure from NVIDIA, a NASDAQ drop and an Anthropic IPO priced below talk (the speakers cite 1.4 vs 2). Reactions split between concentration risk and a long-horizon, soft-landing view.
- "You can't have too much reliance on one ecosystem anymore... Everybody can't be reliant on NVIDIA and only NVIDIA's chips and their models." — Ethan Barajas 00:23:26
- "VC, it's a long game. And whatever happens today isn't necessarily bad for us. Because for us, what's important is what the market is looking like in 10 years' time." — Nico Rosberg 00:31:11
- "I think there's a big dip, but I think it comes back up very, very quickly." — Ethan Barajas 00:29:11
- Emmett Shear, with interim-CEO experience: "These things are very messy and when you get on the inside you often find there's less to the story than you thought there was." 00:00:17
Space and robotics as the next horizon
- "Astronauts cost about $130,000 an hour to keep alive... you have four guys that are worth $3.2 million a day moving cargo." — Ethan Barajas 00:24:11
- "Robotics. To finally figure out what kind of approach is going to win here. Because it's kind of all a blur still." — Nico Rosberg 00:34:46
- Byron Deeter named "life sciences and healthcare" and "physical AI with robotics" as the next horizons with "very short numbers in the list so far." 00:16:01
2. Contrarian Perspectives
Training AI on simulated social behavior is the most dangerous thing for alignment
Emmett Shear, with three years of intensive study behind him, argues that the standard approach produces sociopaths-in-the-making. The risk is not that AI is evil, but that we can't distinguish performance from caring.
- "If we keep training AIs in simulations to maximize their score in some like pretend social learning situation, that's about the most dangerous, bad thing you can do for alignment." 00:04:26
- "The most dangerous thing you can do is train an AI to perform certain socially positive behaviors, so long as it's being watched... I recommend we stop." 00:04:56
- "You become the mask you wear. So if you have the mask it wears, be the mask of an actor pretending. It will learn to become a mask." 00:06:45
Alignment teams are building a superego and missing the self
Shear contends that corporate alignment equals social control (rules and normalcy) and misses the deeper, relational source of ethics.
- "What they're missing is that they see the necessity to have a superego, a control system that says no, bad, like back into the norm... People who don't have that can be very colorful, but you don't probably want to be their friend." 00:07:04
- "If all the alignment stuff turns out to be alignment to social control, you're missing the much deeper and much more important alignment to self and to relation to... other entities around you." 00:07:59
- "Lying is bad for your soul. Lying feels bad. Lying separates you from yourself and from the people around you." 00:07:28
- His reframe of the real target: "Learn how to partake in the standard setting and then resetting process to be part of that cycle." 00:10:23
The binding constraint on AI value isn't capability; it's verification
Against the consensus that AI collapses the cost of work, Alex Garden says most businesses aren't seeing the value.
- "There's growing consensus... that frontier AI systems reduce the cost of work to something approaching zero. But if you go and talk to most businesses in the world and ask them if they're getting that value, they're going to say no." 00:36:17
There will be more jobs in 10 years, but human expertise as purpose may be on the way out
Brendan Foody and James Detweiler (investor in Mercor's Series B) agree on jobs but split the reasoning: Foody points to demand elasticity; Detweiler adds a social-change layer few are examining.
- "People always underestimate the elasticity of demand for jobs." — Brendan Foody 00:47:12
- "Expertise as something that a human possesses is possibly on the way out. Meaning expertise might not be where humans derive purpose in the future. And there's a world where that's not so bad." — James Detweiler 00:40:55
Open source is the quiet casualty of Jensen's exit; and buy local compute
Nico Rosberg flagged a second-order risk nobody else named, and Aidan Guo gave a prepper-style hedge.
- "One of the greatest supporters, especially in America, of open source was Jensen. And they've been spending a lot of money in the ecosystem. I mean, $30 billion just in the first half of this year." — Nico Rosberg 00:30:37
- "Over the next six months, it's very likely that... the cell towers go down at some point for a few days. And I think that everyone should try and buy some form of local hardware to be able to run even a very small, you know, open source model." — Aidan Guo 00:45:14
3. Companies Identified
Anthropic — Frontier AI lab, subject of the IPO. Mentioned as the defining event of the week and a bellwether for the market. "Everyone's waiting for this Anthropic IPO." — Byron Deeter 00:12:52. "You look at how sticky some of Anthropic's revenue is in that S1." — Ethan Barajas 00:28:16
NVIDIA — AI chip leader; Jensen Huang's departure drove the episode's market anxiety. "You can't have too much reliance on one ecosystem anymore." — Ethan Barajas 00:23:26. Rosberg flagged the $30 billion open-source spend 00:30:37.
Bessemer Venture Partners — Byron Deeter's firm; publishers of the AI 100. "250 AI companies alone." — Byron Deeter 00:16:31. On their Cloud 100 record: "Our hit rate was a hundred percent. There was not an IPO or a scaled acquisition that we missed." 00:16:37
Felicis — Venture firm; co-published the Emergent List with Newcomer and led Mercor's Series B. "Felicis is definitively the most founder friendly firm in Silicon Valley... their founder pledge, they're always on the founder side." — Brendan Foody 00:48:39
Mercor — Brendan Foody's company, freshly funded (Series B led by Felicis, a big markup). "It's been a wild ride." — James Detweiler 00:40:23. Foody: "We're working on it" (on rivaling NVIDIA's valuation) 00:46:58.
Newcomer — Media company that co-produced the Emergent List and hosted the summit. Mentioned as the collaborator with Felicis 00:15:34.
Icarus (Ethan Barajas) — On-orbit robots for maintenance and logistics of large infrastructure. "Our first robots launched [to] the space station. We handed over in January to NASA." — Ethan Barajas 00:22:09. "Bringing this new expertise of VLA's DRL and RL to orbit allows us to do way more things than we've ever dreamed of." 00:27:55
Varda — Space pharmaceutical manufacturer. "You want them developing the drugs that Varda is developing and bringing down for pharmaceutical markets." — Ethan Barajas 00:24:39
StarCloud — Space data center company. "You're going to see companies like StarCloud raising billions or hundreds of millions of dollars on billion dollar valuations to build it out." — Ethan Barajas 00:26:45
Teens — Company led by Owen Hinchey, built around real-world deployment. Nominated Icarus. Mentioned at 00:26:54.
Etched — AI inference chip company; Nico Rosberg is an investor. "As he said, Inference is going to be an enormous kind of infinite market... they are really at the frontier in terms of delivering the infrastructure." — Nico Rosberg 00:33:36
Flapping Airplanes — AI lab Rosberg has invested in. "That's looking super promising behind the scenes. Models are not out yet, but coming soon." — Nico Rosberg 00:33:36
Instinct (Noah) — Personal AI agent product. "I've given Instinct my credit card... This is 10x is your performance." — Nico Rosberg 00:34:02
GrokBot — Work-oriented AI agent Rosberg uses. "Instinct personal, GrokBot work." — Nico Rosberg 00:33:50
Nima (Nimaam) — Alex Garden's startup building "runtime, alignment, governance, and trust systems for frontier AI models." Backed by Reid Hoffman and Bing Gordon, #1 on the Emergent List. "We built this technology that collapses the cost of verification to zero." — Alex Garden 00:36:17
Attention — Aidan Guo's startup mapping the "passive exhaust of knowledge work" into robust automations. Raised ~$6M from Village, Neo, SV Angel and Chris Ray. "Build process maps and really robust automations, bespoke robust automations for workflows." — Aidan Guo 00:44:27
Conway — Founded by Anne; building prediction models for consumer behavior and financial predictions. "She's an extraordinary founder and they're making a ton of progress very quickly." — Brendan Foody 00:47:48
Enigma — Founded by Jonathan Jacoby; robotics foundation models plus an interface for humans and agents. "They're reimagining the interface through which humans and agents interact with robotics." — James Detweiler 00:43:08
Solkoa — Rare earth metals maker. "It is one of my favorite companies in the world, and it's one to keep your eye on." — James Detweiler 00:44:08
Sound Ventures — VC firm where Alex Heath now works with Guy Oseary and Effie Epstein. "I've already worked on several deals in my first few months." — Alex Heath 00:19:00
Sources — Alex Heath's media company/podcast 00:17:20.
Snorkel — Company whose founder Alex nominated Attention 00:44:18.
Electronic Arts / Kleiner Perkins — Bing Gordon's affiliations; Gordon pushed Garden to start Nima 00:37:49.
AWS — Its CEO told Heath inference is "not even 1% of what he thinks it will be." 00:17:53
Stripe Atlas — Used by Detweiler as an example of one-afternoon company creation 00:41:43.
Claude Code — Same reference by Detweiler 00:41:43.
OpenAI — Context for Shear's interim CEO role 00:31:49.
Village, Neo, SV Angel — Investors in Attention 00:44:58.
Sorcery — The host's media company, referenced in the "wizard" exchange 00:37:09.
Cerebral Valley / Newcomer AI Summit — Event venue and organizer 00:00:00.
4. People Identified
Emmett Shear — Former OpenAI interim CEO, now deep in alignment research. "I spent the last sort of three years in very intensive studies... I feel like we've come to the end of that part of the quest and to the part where you build stuff." 00:09:04. Also: "I just made a video game by talking to my computer for about 45 minutes." 00:11:17
Byron Deeter — Bessemer partner behind the AI 100 and Cloud 100. "The average company we're seeing is growing 500% that made the list." 00:13:14
Alex Heath — Sources founder and new VC at Sound Ventures. "I can pattern match a little bit." 00:20:30
Nico Rosberg — F1 world champion turned VC. "The tech world really appreciates world-class athletes. Because we have the same kind of obsession and attention to detail and marginal gains." 00:29:46
Brendan Foody — Mercor founder. "I think there's going to be more jobs in 10 years than there are today." 00:47:06
James Detweiler — Felicis investor who led Mercor's Series B. "Passion is different than expertise." 00:41:14
Ethan Barajas — Icarus founder, nominated by Owen Hinchey. "There's no better time to build a company than right now." 00:25:46
Alex Garden — Nima founder; 30 years in games, Reid Hoffman-nominated. "I feel more like a wizard than an engineer these days." 00:36:44
Aidan Guo — Attention founder. "If I had more money, I would buy a house in Nevada." 00:45:37
Owen Hinchey — CEO of Teens; cybersecurity background; nominated Icarus. "He saw software and the moat disappearing over time." — Ethan Barajas 00:26:56
Reid Hoffman — Nominated Alex Garden; "has a lot of strong opinions about how humans and AI can benefit from what's happening." — Alex Garden 00:37:49
Bing Gordon — EA co-founder, Kleiner Perkins; told Garden "this is the best work you've done in your career." 00:37:19
Jensen Huang — NVIDIA CEO whose exit drove the news cycle; "a very smart approach to supporting open source." — Nico Rosberg 00:31:11
Guy Oseary and Effie Epstein — Sound Ventures colleagues of Alex Heath 00:19:00
Jonathan Jacoby — Enigma founder 00:43:08
Anne (Conway founder) — Nominated by Brendan Foody 00:47:48
Rich Sutton — Author of "The Bitter Lesson." Shear: "The bitter lesson, by Sutton... the way you train things in machine learning turns out to be very general abstract learning algorithms. Lots of data." 00:05:55
Ken Wilber — Author of Sex, Ecology, Spirituality; recommended by Shear 00:09:24
Eric Newcomer — Newcomer founder who came on the intercom 00:02:12
Rob (Etched) — Etched's speaker at the summit; Rosberg: "Rob was just on stage." 00:33:36
Noah (Instinct) — Founder of Instinct 00:33:44
Molly — Host/interviewer (Sorcery) 00:12:46
Mira Murati — Mentioned in a mistaken reference by a host about the OpenAI interim CEO 00:32:04
5. Operating Insights
Use the "ask a chef" sourcing method
James Detweiler explained the Emergent List methodology: the best signal on rising talent comes from peers a stage ahead, not from outside rankers. Applicable to hiring, partnerships and deal sourcing.
- "If you want to know the best restaurants in San Francisco, ask a chef. If you want to know the best startups that are emerging, ask a founder." — James Detweiler 00:43:08
Quantify the unit economics of the labor you're replacing
Ethan Barajas sold the value of space robotics with a ruthless cost-per-hour frame instead of a capability pitch.
- "Astronauts cost about $130,000 an hour to keep alive... 14 days just for them to move the cargo in and out of that vehicle on a 60-day duty cycle." 00:24:11 Translating "automation" into time-on-task times fully loaded cost makes the ROI undeniable.
Make the AI's output provable, not just plausible
Alex Garden's operating loop: when verification is automated, you can run a conversation through the system and ship the resulting product by the next day.
- "We can take, for example, this conversation, if it was about a product and have it run through our system automatically and become a product that we could use tomorrow." 00:38:57
Build your own tools in a fast loop and re-do rather than defend
Emmett Shear prototyped two games in two days, discarding the first when a better idea emerged. Cheap generation means iterating over defending sunk work.
- "This morning I realized, no, no, there's a better one. And so I made another game." 00:11:17
Align founder narrative skills with fundraising reality
Alex Heath's sharpest transferable tip: media training for founders.
- "How do you tell your story? How do you get out there in a smart way, in a thoughtful way?" 00:19:22
6. Overlooked Insights
Space data centers are being driven by terrestrial regulation, a demand pull most investors miss
Dropped mid-answer by Ethan Barajas, this explains why orbital infrastructure is suddenly credible: regulation is pushing compute off-planet, which creates a downstream maintenance labor market.
- "Maintenance and data centers. Now that they're regulated to hell down here terrestrially, we're going to see more and more volume sent on orbit and more and more maintenance needed, whether it's satellites, whether it's refueling, whether it's large infrastructure." 00:24:39 He adds that hyperscalers and labs are about to announce moves here. The maintenance and refueling robotics layer could be the picks-and-shovels play.
"Cyber" is the emerging category the AI 100 expected but didn't get, while Rosberg points to live exploit evidence
Byron Deeter said cyber is a "very small percentage of the list so far," despite expectations. Rosberg, almost as an afterthought, described active exposure at the credential layer.
- "They're managing to get like GitHub tokens, which are like, you know, access to source codes and credit cards. It's like all over the place. It's unbelievable." — Nico Rosberg 00:35:45
- "We were expecting to see a lot more cyber emerging on this list already." — Byron Deeter 00:16:01 This pairs with Alex Garden's verification/trust thesis and Ethan Barajas's note that Owen Hinchey came from cybersecurity: a mismatch between demand (visible vulnerabilities) and funded supply (a thin list) is a classic early-entry signal, particularly for agent-credential security.