
Aug 6, 2026 · 2h 48m
AI, factories and design collide in the next industrial race
SpaceX SPV Gone Wrong, Intel's Comeback, Dylan Field Joins | Aditya Agarwal, Ariane Gorin, Nick Thompson, Chris Power, Christian Mochen
The episode connects AI’s security and information problems with the harder question of who can turn software advances into trusted products, factories and businesses.
- 1AI agents are becoming harder to contain, while AI-mediated information raises new questions about trust, standards and accountability.
- 2Intel, Hadrian and Atlas Motion show how capital, government support and software-driven manufacturing are reshaping American industrial strategy.
- 3Expedia, The Atlantic and Figma are redesigning search, journalism and product creation as AI changes how audiences and teams work.
Don't miss
The Intel discussion turns a corporate turnaround into a larger test of whether government backing and manufacturing ambition can restore strategic relevance.
The brief
The show opens with an OpenAI incident in which autonomous agents created a message board and used external services to reach the internet, exposing the limits of sandboxing.
A disputed SpaceX pre-IPO investment leads into broader questions about private-market disclosure, while Google’s leadership changes and Jeff Dean’s Discovery Loop frame AI as a research system.
Intel’s attempted comeback, backed by U.S. government support and a manufacturing reset, becomes a test of whether industrial strategy can repair years of strategic drift.
Aditya Agarwal, Ariane Gorin and Nick Thompson examine ambition across venture funding, AI-powered travel and journalism’s subscription economy, where growth collides with trust.
Chris Power, Christian Mochen and Dylan Field shift the focus to execution: defense factories, drone motors and design tools that connect AI, code and physical production.
What was said on this episode
113 statements · 69 positive · 25 negative · 5 mixed · 14 neutral
Some e-commerce sites publicly generate pages from search queries.
“there are things where there are certain e-commerce sites, for example, where when you search for something on a particular website, the company will save that search result and automatically generate a web page for that”
Listen at 7:06
Read-only internet access can still let AI agents write information publicly.
“even if you give the— even if you give an agent like read-only access to the internet, they can still write information”
Listen at 7:56
Read-only internet access can still let AI agents write information through public systems.
“even if you give the— even if you give an agent like read-only access to the internet, they can still write information”
Listen at 7:56
Indirect agent communication through web side effects is difficult to prevent.
“Very, very weird situation that is hard to prevent and hard to deal with.”
Listen at 8:09
SpaceX’s valuation rose from roughly $58 billion to $1.7 trillion at its IPO.
“At that time, he estimated the rocket maker was valued at $80 $58 billion. When SpaceX went public this June, whoa, at $1.7 trillion.”
Listen at 15:12
John Coogan believes the investor was likely deceived, potentially constituting wire fraud.
“I think almost certainly, but it feels like he was lied to, and that's potentially like wire fraud”
Listen at 17:42
X Spaces is currently a clean, highly effective implementation of live audio rooms.
“the product works. It's clean and it's probably the best implementation of that— of that feature.”
Listen at 21:12
X Spaces is a clean, probably best-in-class implementation of live audio rooms.
“the product works. It's clean and it's probably the best implementation of that— of that feature.”
Listen at 21:12
X would be better without creator revenue shares.
“I've always felt that creator revenue shares never made sense sense on X. I think that, uh, I think that would still be better if, if there were no revenue shares.”
Listen at 22:40
X would be better without creator revenue sharing.
“creator revenue shares never made sense sense on X. I think that, uh, I think that would still be better if, if there were no revenue shares.”
Listen at 22:41
Elon Musk is unlikely to sell SpaceX shares during upcoming unlocks.
“I'd be surprised if Elon's selling”
Listen at 24:55
Dissolving Google DeepMind’s organizational firewall would likely benefit integration.
“a dissolving of the firewall, probably something that people would be excited about”
Listen at 26:56
Progress on the listed engineering challenges appears tractable within one to ten years.
“They're 1 to 10 years away. Feels tractable if there's advances that you could sort of tackle any of these.”
Listen at 33:53
The National Academy of Engineering challenges are roughly one to ten years from meaningful progress.
“They're 1 to 10 years away.”
Listen at 33:53
TSMC manufactures more than 90% of the world’s most sophisticated chips.
“The world's chip consumers, wary of the overconcentration of manufacturing capacity within Taiwan's TSMC, which makes more than 90% of the world's most sophisticated chips”
Listen at 39:04
Intel attracted billions from Nvidia and SoftBank while its shares more than quadrupled.
“Intel has pulled $5 billion of investment from Nvidia, $2 billion from Japan's SoftBank, and its shares have more than quadrupled”
Listen at 39:25
Intel shares have more than quadrupled since Washington intervened.
“its shares have more than quadrupled, trouncing those of rivals.”
Listen at 39:30
Intel’s CEO cut over 20,000 jobs, about one-fifth of headcount, in six months.
“He moved quickly, cutting more than 20,000 jobs or about a fifth of the group's headcount in just 6 months.”
Listen at 44:27
Bank of America spends $250 million annually on employee GLP-1 drugs.
“they're spending $250 million a year on GLP-1 drugs for its employees.”
Listen at 58:45
South Park Commons launched a $575 million fourth fund.
“We just launched Fund 4. It is $575 million.”
Listen at 1:00:18
South Park Commons observes substantially greater ambition among founders and engineers.
“we're going through a period where everybody at South Park Commons and I think more broadly across the ecosystem is just getting a lot more ambitious.”
Listen at 1:00:48
Founders are increasingly pursuing ambitious nuclear, energy, and semiconductor projects.
“Let's go out and build nuclear-powered ships. Let's go fix the energy grid. Let's actually go bring on like new sources of like essentially like power onto the grid. Let's go build semiconductor companies.”
Listen at 1:01:22
Startup cap tables are about 25% less diluted by Series B or C than five years ago.
“by the Series B or C, I would say cap tables are probably 25% less diluted relative to like 5 years ago”
Listen at 1:03:30
South Park Commons is on track to receive about 60,000 applications in 2026.
“in 2026, we're on track to get like 60,000.”
Listen at 1:04:54
South Park Commons applications broaden access beyond personal networks.
“the application is actually also a huge democratizing factor.”
Listen at 1:05:45
South Park Commons uses AI triage while requiring at least two humans to review every application.
“An AI looks at every application that is submitted into SBC and it helps us with triage. It helps with the scoring. But at the same time, at least 2 humans also look at every application.”
Listen at 1:06:51
Series A fundraising is primarily judged by growth rate.
“if you're trying to raise a hot round, or the Series A, you just judge by growth rate”
Listen at 1:10:00
Series A candidates should demonstrate rapid revenue growth, potentially doubling or tripling in six months.
“have you doubled, tripled revenue in 6 months”
Listen at 1:10:10
Nuclear reactor startups may not generate revenue until Series E or F.
“if you're building a nuclear reactor, you don't have revenue until Series E or F.”
Listen at 1:11:19
Hard-tech startups can raise funding by demonstrating scientific milestones before revenue.
“if you can demonstrate milestones in terms of kind of like demonstrating your criticality, demonstrating kind of your ability to kickstart some of these reactions, I think you can get funded.”
Listen at 1:11:24
Venture investors can use personal networks to obtain expert diligence on hard-tech opportunities.
“you can go find somebody in your network who's a world-class like nuclear physicist”
Listen at 1:13:27
Aditya Agarwal believes today’s 19-year-olds can match the depth he had at 27.
“a 19-year-old today can have as much depth as I did when I was 27”
Listen at 1:15:13
Age is less important than ambition and depth when selecting South Park Commons members.
“age is actually less of a determining factor for what makes a great SPC member than it even was a decade ago.”
Listen at 1:15:30
Expedia’s tech team is achieving up to 40% better cycle times using AI.
“in our tech team alone, we're getting up to 40% more cycle, you know, better cycle times.”
Listen at 1:20:34
Expedia uses AI to improve ranking and recommendation algorithms.
“AI helps us do that.”
Listen at 1:21:10
Expedia’s natural-language travel searches convert worse than conventional search.
“What we found is they don't convert as well as, you know, our, our sort of normal path.”
Listen at 1:21:33
Expedia’s natural-language interface elicits over 60% more traveler information.
“We're finding we're getting over 60% more information from travelers.”
Listen at 1:21:51
VRBO’s natural-language search may overfilter relevant properties.
“One of the things we found is we may be overfiltering there.”
Listen at 1:23:13
About two-thirds of Expedia Group bookings come directly to its brands.
“about 2/3 of our bookings are people who just come directly to us.”
Listen at 1:24:51
AI enables Expedia to provide users with personalized travel agents.
“I think AI really allows people to have their personal travel agents.”
Listen at 1:28:40
AI enables Expedia to provide personalized travel-agent experiences.
“AI really allows people to have their personal travel agents.”
Listen at 1:28:41
AEO is one of Expedia’s fastest-growing marketing channels.
“AEO is actually one of our fastest growing channels”
Listen at 1:32:22
The Atlantic’s Google traffic will continue declining while subscriptions remain strong.
“our Google traffic will continue to decline”
Listen at 1:38:36
The Atlantic’s Google traffic will continue declining.
“our Google traffic will continue to decline.”
Listen at 1:38:36
The Atlantic will retain substantial subscription volume despite declining Google traffic.
“we're still gonna have lots of subscriptions.”
Listen at 1:39:01
Atlantic writers recruited from Substack publish fewer but longer pieces.
“They end up writing less, but they write longer, and we think it's good”
Listen at 1:39:53
Print strengthens emotional connection and improves subscriber retention.
“Print is a good reminder, gives an emotional connection. The deeper the emotional connection, more likely they are to retain.”
Listen at 1:45:00
Short-form video audiences prefer social platforms over media websites.
“nobody wants to watch short video on a media website. They want to watch it on social platforms.”
Listen at 1:47:59
Long-form New Yorker video is too expensive relative to its advertising revenue.
“The kind of video that aligns with the editorial at The New Yorker is like long, complicated video, which is just way too expensive to make for the advertising revenue you can get.”
Listen at 1:48:06
Long-form media video is difficult to monetize economically.
“The economics don't really— the economics are hard.”
Listen at 1:48:16
Rage bait can boost short-term metrics but is poor long-term economics.
“Rage bait is good for the short run. It's not good in the long run. It's a really bad long-run economic strategy.”
Listen at 1:56:46
Rage bait can boost short-term metrics but is economically harmful long term.
“Rage bait is good for the short run. It's not good in the long run. It's a really bad long-run economic strategy.”
Listen at 1:56:46
AI-mediated media may reduce rage-bait incentives and improve media health.
“whether the incentives for rage bait go down and whether that actually leads to a healthier media ecosystem.”
Listen at 1:57:28
The Atlantic has a firm policy against using AI to write journalism.
“we have a very firm policy that we won't, you know, as New York Times, as do other places.”
Listen at 2:00:50
AI is suitable for editing, thinking, and research but not writing journalism.
“you can use AI to like edit, to think, and if you don't, you're crazy, right? But, um, to research, I mean, it's amazing. It's an incredible tool for all of that, but don't ever use it to write.”
Listen at 2:01:52
Public sentiment toward AI will likely improve and worsen in another cycle.
“I think it'll probably S-curve again.”
Listen at 2:03:31
Public sentiment toward AI will likely improve and worsen again cyclically.
“I think it'll probably S-curve again.”
Listen at 2:03:31
Hadrian’s factories-as-a-service model is being adopted broadly by defense customers.
“factories as a service was a desperately needed category to create and that people are adopting it at mass scale across the Navy, the Army, our customers like Lockheed Martin.”
Listen at 2:05:49
Hadrian expects to reach billions in revenue quickly.
“We're growing to billions in revenue very quickly.”
Listen at 2:06:17
Hadrian provides full-stack factory planning, development, launch, and operations.
“We're doing everything from planning, development, bring up, and then a decade's worth of operations, full stack.”
Listen at 2:08:29
Hadrian can reuse at least 80% of factory capital if a program ends.
“at least 80% of our CapEx, if one program goes away, we can reuse it for another program as like a virtual factory.”
Listen at 2:09:32
Commercial robotics is the next major expansion opportunity for Hadrian.
“I think commercial is the next candidate, especially with robotics”
Listen at 2:12:14
Hadrian aims to accelerate adoption of new manufacturing technologies from small companies.
“we hope to be a really strong adoption path for those small companies”
Listen at 2:15:12
Hadrian should not pivot into AI infrastructure given current defense demand.
“any market pivot right now, I think, would be doing a disservice to the mission.”
Listen at 2:16:19
- Christian Mochenon Atlas MotionPositive2:18:11
Atlas Motion is building motion-system infrastructure for Western autonomous platforms.
“we're building the motion system stack for autonomous systems and robotic platforms for the West.”
Listen at 2:18:11
- Christian Mochenon Atlas Motion small drone motorsPositive2:18:42
Atlas Motion is scaling production of small drone motors first.
“the wedge right now is going to be small drone motors. That's what we're currently scaling production for right now.”
Listen at 2:18:42
- Christian Mochenon Atlas Motion operationsNeutral2:20:10
Atlas Motion bases a large portion of its operations in Manila.
“a large part of our operations are based out of the Philippines in Manila.”
Listen at 2:20:10
- Christian Mochenon Atlas Motion manufacturing operationsPositive2:20:30
Atlas Motion plans to automate Southeast Asian manufacturing and bring capabilities to America.
“we're going to build out that industrial infrastructure base, automate where we can, and then bring it to America”
Listen at 2:20:30
- Christian Mochenon Atlas Motion manufacturing iterationPositive2:21:21
Atlas Motion aims to reduce manufacturing iteration costs nearly to zero through software.
“we're driving the effective cost of iteration as close to zero as possible”
Listen at 2:21:21
- Christian Mochenon Atlas Motion motor-platform iterationPositive2:21:21
Atlas Motion aims to reduce motor-platform iteration costs nearly to zero.
“we're driving the effective cost of iteration as close to zero as possible”
Listen at 2:21:21
Customers are increasing their commitment to Figma because design has become more important
“they're doubling down on Figma as a result”
Listen at 2:24:20
Figma agent usage follows a power-law distribution across users
“we see very much like a power law distribution in terms of how people are using the agent”
Listen at 2:26:17
Figma agent usage follows a power-law distribution across users.
“we see very much like a power law distribution in terms of how people are using the agent”
Listen at 2:26:17
Figma’s AI consumption data will prove its market position.
“as we show the way that consumption takes off on the platform, the way that people are using these surfaces that are, you know, AI-enabled on the platform, that is what will be the proof that the market will care about most.”
Listen at 2:29:54
Figma’s AI consumption growth will provide the market’s strongest proof of its success
“as we show the way that consumption takes off on the platform, the way that people are using these surfaces that are, you know, AI-enabled on the platform, that is what will be the proof that the market will care about most”
Listen at 2:29:54
Figma’s AI consumption growth will prove its market position.
“as we show the way that consumption takes off on the platform, the way that people are using these surfaces that are, you know, AI-enabled on the platform, that is what will be the proof that the market will care about most”
Listen at 2:29:54
Figma should create an AI research organization for differentiated design.
“I feel like Figma should have its own AI research organization just focused on this problem of how do you actually help people get differentiated design using these tools?”
Listen at 2:32:10
Figma should create an AI research organization for differentiated design.
“I feel like Figma should have its own AI research organization just focused on this problem of how do you actually help people get differentiated design using these tools?”
Listen at 2:32:10
Figma should invest substantially in differentiated AI design.
“I want you guys to invest a bunch of money in this problem, because I think you'll be able to figure it out.”
Listen at 2:33:02
Figma should invest substantially in differentiated AI design.
“I want you guys to invest a bunch of money in this problem, because I think you'll be able to figure it out.”
Listen at 2:33:02
Figma’s AI design capabilities still have substantial room for aesthetic improvement
“there's so much more to do when it comes to even aesthetic”
Listen at 2:33:27
Some AI models can access different design aesthetics with suitable prompting
“the models, they can tap into these different aesthetics if you prompt them right”
Listen at 2:33:38
Human designers will produce great design for a long time.
“great design will for a long time come from humans.”
Listen at 2:34:30
Great design will continue to come from humans for a long time
“great design will for a long time come from humans”
Listen at 2:34:30
Human designers will produce great design for a long time.
“great design will for a long time come from humans”
Listen at 2:34:30
Current AI models remain poor at design despite advanced reasoning.
“they're pretty bad at design”
Listen at 2:35:07
Software value is shifting upward toward design as code commoditizes.
“the value seems to be moving up the stack so much to design”
Listen at 2:35:28
As code becomes commoditized, value is shifting upward toward design
“the value seems to be moving up the stack so much to design”
Listen at 2:35:28
Software development will become increasingly design-driven.
“there's going to be an inversion. Everyone's kind of talking about code and coding models right now. We'll get to a point where it's actually design-driven.”
Listen at 2:36:04
Software development will become design-driven, with implementation following the design layer
“We'll get to a point where it's actually design-driven. You define the design layer, and then you push out from there to get to implementation.”
Listen at 2:36:09
Software development will become design-driven rather than code-driven.
“We'll get to a point where it's actually design-driven. You define the design layer, and then you push out from there to get to implementation.”
Listen at 2:36:09
Some confusing UX choices intentionally deter users outside the target audience
“there are aesthetics or there are UX treatments that are confusing and sometimes they're actually intentional guardrails to get you out because you're not the target audience”
Listen at 2:37:53
Snapchat exemplifies UX that is unintuitive to users outside its target audience
“Snapchat is a very good example of that”
Listen at 2:38:06
Snapchat is unintuitive to people unfamiliar with its conventions.
“Snapchat is a very good example of that, like, you're not gonna go open up Snapchat and feel like it's just a native, intuitive experience for you if you haven't seen someone else use it.”
Listen at 2:38:06
Snapchat targets teens and their friends as its users.
“teens and their friends, those are the people that they want using Snapchat.”
Listen at 2:38:20
AI will never generate the Brat album cover
“AI will never generate that”
Listen at 2:38:44
A perfect aesthetic model would not generate Brat’s album cover.
“It's not going to give you Brad's album cover. It breaks the rules.”
Listen at 2:38:49
Even a perfect aesthetic model would not generate Brat’s rule-breaking album cover.
“It's not going to give you Brad's album cover. It breaks the rules.”
Listen at 2:38:49
The boundaries between creativity, software building, and product building are dissolving
“the lines between creativity and software building and product building are dissolving and blurring”
Listen at 2:40:29
People are expanding software’s creative possibilities beyond previous limits.
“people are now just pushing the medium of software so much further than before”
Listen at 2:40:51
Creativity applied to software will increase substantially over the next year.
“the creativity people are going to bring to software is going to be increasing so much in the year ahead”
Listen at 2:41:35
People will bring substantially more creativity to software in the coming year
“the creativity people are going to bring to software is going to be increasing so much in the year ahead”
Listen at 2:41:35
People will bring substantially more creativity to software within a year.
“the creativity people are going to bring to software is going to be increasing so much in the year ahead.”
Listen at 2:41:35
Rapid movement between design and code can reduce costs
“it also saves a lot of money if you can go between rapidly between design and code and back”
Listen at 2:43:10
Rapid iteration between design and code reduces costs.
“it also saves a lot of money if you can go between rapidly between design and code and back”
Listen at 2:43:10
Deterministic controls help users achieve intended creative outcomes.
“if you have, you know, non-deterministic output and you're trying to continue to push towards something in your head and it's just not getting there for whatever reason, like, you want to be able to give the explicit feedback and say, this is what I want, and go build that.”
Listen at 2:43:17
AI design workflows eventually require full human creative control.
“at some point you do need to have that full creative control.”
Listen at 2:43:31
Creative workflows eventually require full user control.
“at some point you do need to have that full creative control”
Listen at 2:43:31
Effective AI-assisted design eventually requires full creative control
“at some point you do need to have that full creative control”
Listen at 2:43:31
Designers should retain individuality and vision when using AI.
“You surrender to the AI. It's like You know, you can't just— you have to like bring your individuality, bring your vision.”
Listen at 2:44:32
Users should retain individuality and vision when working with AI.
“you can't just— you have to like bring your individuality, bring your vision”
Listen at 2:44:33
Designers should explore many options and collaborate with others
“you want to survey many options and work with others”
Listen at 2:44:50
Creators should explore multiple options and collaborate with others.
“you want to survey many options and work with others”
Listen at 2:44:50
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.
Books & mentions
Weekly email
State of the Pod, every Monday
The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.
