🟠 YC Demo Day, Paul Graham Joins, Will AWS Buy TPUs From Google? | Harj Taggar, Paul Graham & Jessica Livingston, Richard Wang, Philip Ho, Ali Attar, Kurush Dubash & More

Summary

TBPN Podcast Summary

Episode: YC Demo Day, Paul Graham Joins, Will AWS Buy TPUs From Google?

Date: December 3, 2025
Hosts: John Coogan & Jordi Hays
Special Guests: Jessica Livingston & Paul Graham (YC), Harj Taggar (YC), founders from Absurd, Clad Labs, Lightberry, Dome, Source, Matorial, Crunched, Sava, SF Tensor, Locus, Icarus, and more.

---

Overview

This episode recaps the Winter 2025 YC Demo Day, explores the evolving AI infrastructure wars (AWS vs. Google vs. Nvidia), examines the economic and cultural forces shaping Silicon Valley, and features exclusive interviews with Y Combinator founders, including a surprise drop-in from YC co-founders Jessica Livingston and Paul Graham. The show rides a high-energy, high-context discussion on the latest trends, viral moments, industry controversies, and investment themes across AI, agentic commerce, robots, and platform infrastructure.

---

Key Themes & Discussions

1. AI Infrastructure Wars: AWS Trainium vs. Google TPU

• AWS’s Big Launch:

  • AWS has launched Trainium 3, touted as 4x faster than its predecessor and up to 50% cheaper than equivalent GPUs for AI workloads ([03:21]).
  • CEO of AWS, Matt Garman, avoided a direct answer on whether AWS will source Google TPUs, but emphasized optimizing their own stack and customer choice ([07:54]).

• Customer Choice & Amazon’s DNA:

  • Amazon’s historic motto—"we’re customer obsessed, not competitor obsessed"—guides its willingness to serve up chips (Nvidia, AMD, maybe even Google TPUs if needed) based on client needs ([10:45]).
  • While some see the Trainium launch as an anti-Nvidia move, real-world feedback is mixed: some startups thrive (e.g., Descartes), others, like Anthropic, reportedly shift to TPU ([07:19]).
  • Notable quote:

    “It’s funny to mock anyone for something related to their semiconductor supply chain… AWS is customer obsessed.” —John ([10:45])

• Market Dynamics & Alliances:

  • Nvidia’s dominance is under siege, with hyperscalers forming “anti-Nvidia” alliances to push down GPU margins ([15:09]).
  • Amazon’s role remains unique—neither in the Google complex nor OpenAI complex ([16:00]).

2. YC Demo Day 2025: Trends and Standouts

• Faster Revenue Growth, Larger Deals:

  • “The dollar value contracts that startups can close in the first few months… are bigger than ever. That’s all AI.” —Harj Taggar ([48:02])
  • Two tracks for go-to-market:
    • Land one huge contract (often defense/gov/fortune 500 via AI leapfrog)
    • Or aggregate startups (“Modern Stripe/AWS model”)

• Emergence of Full-Stack, AI-Native Startups:

  • Companies increasingly not just building “agent” tools, but vertically integrating—becoming full-stack providers (e.g., AI-native insurance brokers, trust companies) ([51:59])
  • This echoes the “full-stack startup” concept from Balaji/TLAG a decade ago, now newly possible thanks to AI ([53:32])
  • “You didn't even have the option of selling to a big customer until you sold to startups… Now with AI, you can.” —Harj Taggar ([50:43])

• Themes Across Batches:

  • Prevailing shift: From agent infra → vertical applications → AI-native firms.
  • Notable: Startups focus on “unsexy” industries (insurance, law, trusts) that big labs won’t touch ([55:20])

• YC Demo Day Coverage – Featured Startups:

  • See “Founder Interviews & Notable Startups” below.

3. Silicon Valley Cultural & Economic Forces

• Startup “Rage Bait” & Virality:

  • Discussion around the strategy and limits of rage bait (e.g., Clad Labs’ gambling-in-IDE) as a growth tactic ([60:02])
  • Paul Graham perspective: “These scammers… don’t make the giant companies. They don’t have a long-term focus… they're not earnestly doing engineering” ([152:56])

• Earnest Hacker vs. "Deals Guy" Era:

  • Recurring debate: Are “earnest hackers” being displaced by business-oriented “deals guys”?
  • PG rejoinder: “If anything, YC drifted too far away from funding earnest hackers… I would still bet on earnest hackers.” ([157:33])
  • “If the YC partners are themselves hackers, you can sniff out a faker like that.” ([158:26])

• Bubble Watch & AI Valuations:

  • “Is it a bubble?”
  • Paul: “AI is very highly priced, but it might not be overpriced. It’s definitely real, not hype.” ([160:25])

4. The State and Future of YC & Startups

• Demo Day Vibes:
  • “It’s buzzing. The energy is just kind of like what I remember in the early days.” —Jessica Livingston ([148:53])
• Startup Union Effect:
  • YC’s alumni network operates akin to a union, raising standards for founder-friendliness ([163:41])
• On The Future:
  • “There’s a secular trend of more people starting startups. The limit is what people want. And people’s wants… are infinite.” —Paul Graham ([171:59])

5. AI Progress, AGI Debates & Economic Diffusion

• Dwarkesh Patel’s “AI Progress” Essay ([29:32] – [37:43]):
  • AGI’s timeline is not just scale, but skills baked in; economic “diffusion lag” is “cope for missing capabilities.”
  • Example: Even with advanced models, human curation and emotional intelligence (e.g., podcast clip selection) still outperforms AI ([32:01]).
  • “Models are nowhere near as capable as human knowledge workers.” ([37:43])

---

Notable Quotes & Moments

• On the AI Hardware Market:

  “Nvidia has an insane amount of power right now. … They ramped up revenue from $27B in 2023 to $130B in 2025… one of the greatest revenue ramps at scale in history.” —John ([13:28])

• On Earnest Hacking:

  "Isaac Newton was an earnest hacker. … This is what wins." —Paul Graham ([158:42])

• On Rage Bait:

  “That sort of technique sounds like something popular with a bit of a scammer… they don’t have a long-term focus… skip those companies.” —Paul Graham ([152:56])

• On Trust Infrastructure and Stripe Atlas Parallels:

  “When the infrastructure gets better, usage goes up. … The safe is an incredible invention.” —Nimit Maru, Sava ([126:26])

• On the Next Wave of Robotics:

  "We literally have a humanoid robot upstairs right now, emceeing the entire event for demo day… fully autonomous." —Ali, Lightberry ([79:35])

• YC Founders’ Legacy:

  "We never realized [the alumni network] would be so important… It’s like taking over Silicon Valley." —Paul Graham ([163:22])

---

Noteworthy Segments & Timestamps

• AI infrastructure wars and AWS/Google/Nvidia triangle – ([01:20] - [19:13])
• Dwarkesh Patel’s AI progress and diffusion (deep AGI discussion) – ([29:32] - [37:43])
• Clip discussion – why AI can't replace human curation yet – ([32:01] - [34:40])
• YC Demo Day trends/Harj Taggar interview – ([46:18] - [58:56])
• Clad Labs (“Rage Bait IDE”) founder interview – ([59:44] - [65:48])
• Absurd (AI-native ad agency) founder interview – ([66:41] - [77:57])
• Lightberry (social brains for robots) founder interview – ([78:27] - [88:59])
• Dome (prediction markets API) founder interview – ([90:01] - [97:44])
• Early summary and YC community discussion – ([148:22] - [166:26])
• Paul Graham & Jessica Livingston (surprise appearance): deep-dive reflections on YC, trends, advice – ([147:41] - [172:43])

---

Founder Interviews & Notable Startups

• Clad Labs: Building a viral, ad-fueled coding IDE—integrates “brain rot” sites/ads at user demand, betting on subsidizing dev tools with contextual affiliate links ([59:55] - [65:48]).
• Absurd: AI-native creative agency (“the new full-stack”), producing viral ads using frontier video gen models, charging $30K+/ad, growing towards retainer/checkpoint deals ([66:52] - [77:57]).
• Lightberry: “Operating system for robots”—voice-triggered software for humanoid and non-humanoid robots with manufacturers like Unitree; focus: easy B2B deployment ([78:27] - [88:59]).
• Dome: API/unified interface for prediction markets (Kalshi, Polymarket, etc.), focused on both institutional and developer tools—arbitrage and liquidity unification ([90:01] - [97:44]).
• Source: "Tinder for jobs": consumer job-app tool, viral through TikTok, uses browser agents to auto-fill applications ([98:52] - [104:39]).
• Matorial: Integration layer for agentic AI (Gmail, SAP, Salesforce, etc.), focused on security/access control and multi-agent compatibility ([105:23] - [112:08]).
• Crunched: Excel copilot for finance power users (banking, PE, consulting)—detects errors and works within complex templates ([112:31] - [119:31]).
• Sava: AI-powered trust company (estate/fiduciary), modernizing expensive/inaccessible legal vehicles ([121:15] - [127:04]).
• SF Tensor: Infrastructure for AI training—abstracts away where/what clouds, orchestrates across NEOs/hyperscalers, supports multi-arch (TPU, AMD, GPU) ([128:47] - [134:34]).
• Icarus: 20-foot, solar-powered, long-flight drones for defense and stratospheric applications ([136:22] - [143:10]).
• Locus: Payment infra for agentic commerce—"pay per use" agent payments, inspired by X402 ([144:35] - [147:24]).

— Many others covered throughout, especially during live founder call-ins between [59:19] and [143:14].

---

Paul Graham & Jessica Livingston: Deep Dives & Wisdom

Key Takeaways:

• AI Hype vs. Reality: “AI is highly priced, but maybe not overpriced. It’s definitely real.” —PG ([160:25])
• “Earnest Hacker” Endures: “I would still bet on earnest hackers.” ([157:33])
• Rage Bait Is No Recipe for Enduring Success: “Scammers… don’t make the giant companies. They don’t have a long-term focus.” ([152:56])
• Founders Should “Just Make Something Really Good”: “You don’t need a coalition, just that small, intense fire.” ([153:58])
• On the Future of Startups: “People’s wants are infinite… There’s infinite demand for good stuff you could make.” ([172:04])
• Simple Recipe: “Build stuff and talk to users.” ([169:49])
• On Pivots & Growth: “All that matters is growth rate, not the absolute numbers.” ([168:08])
• YC Alumni as a “Union”: Yes, but “we want everyone to play by the rules… The big wins don’t come from breaking the rules.” ([163:45])

---

Memorable & Fun Bits

• Paul Graham called “rage bait” the word of the year, but not a long-term strategy. ([153:22])
• Sim racing rigs debated: AMD Ryzen X3D vs. Intel ([105:03])
• San Francisco vibes improving: “It gets better every time we come back.” —Jessica Livingston ([155:23])
• Big names and animal-based startup puns—a running joke.
• The show’s chat had surprise guest drop-ins (Gary Tan, Ali, Jana).
• Hosts’ rapid-fire, irreverent tone (“Ham Solo”, “swine-themed names”, goat sound cues, etc.).

---

Final Thoughts

This episode highlights the extraordinary energy, technical ambition, and occasionally zany cultural currents running through today's technology ecosystem—especially at the intersection of generative AI, deep compute, verticalization, and the relentless hunt for new business models. The hosts and their guests surface the pulse points of YC’s world: full-stack AI-native startups, platform infrastructure playbooks, and the enduring gospel of “earnest hackers who build things people want.” With wisdom from OGs like Paul Graham and Jessica Livingston and front-line founder reality checks, the TBPN team turns a firehose of Silicon Valley action into a surprisingly hopeful and entertaining broadcast.

---

Essential Timestamps

| Segment | Start (MM:SS) | End (MM:SS) | |-------------------------------|---------------|--------------| | AI Hardware Wars/Trainium | 01:20 | 19:13 | | Dwarkesh Patel AI Progress | 29:32 | 37:43 | | Clad Labs Interview (Rage Bait)| 59:44 | 65:48 | | Absurd (AI Agency) | 66:41 | 77:57 | | Lightberry (Robots) | 78:27 | 88:59 | | Dome API for Prediction Mkts | 90:01 | 97:44 | | YC/Paul Graham Deep Dives | 147:41 | 172:43 |

For quick founder overviews, scan [59:44]-[143:14], where most pitches occur.

---

The Tone

The episode is high-velocity, irreverent, and opinionated—mixing Backchannel-worthy analysis, bold metaphors, inside jokes, and a direct, founder-to-founder candor. If you want the unfiltered pulse of Silicon Valley 2025, this is it.

---

End of summary.
(For questions or particular founder deep dives with quotes, ask for timestamps or segments!)

Transcript

[John]

You're watching TVPN. We are live from the TBPN Ultradome. It's Wednesday, December 3rd, 2025. You probably thought we were at YC Demo Day in San Francisco. We gotta go to New York City tomorrow. We're interviewing Jim Cramer, a bunch of other folks today, actually.

[Jordi]

Yeah, traveling today.

[John]

We are traveling today. So we couldn't, we unfortunately couldn't be in San Francisco at the palace of Party Rounds. But we still have a ton of YC Demo Day content lined up for you folks. We got Harsh Dagar coming on at 11:45. Then we got Clad Labs, the chat makers of Chad Ide, the Rage Bait. Yeah, the company that sparked they. By their own, by their own definition, they call themselves the Brainrot Ide. We're getting to the bottom of that story and then we're talking to probably 10 or 20 other founders going to be asking them how they're building their businesses, what they're building, what they're seeing.

[Guest]

They're.

[John]

It's always a fun time to check in with the good folks over at YC. And of course we will be telling you about ramp.com, time is money save, both easy use, corporate cards, bill pay accounting and a whole lot more. Aye, aye, ay, ay, ay. And I will also be telling you about Fall, the general media platform for developers develop and fine tune models with serverless GPUs and on demand clusters.

So today I wrote about Will AWS buy TPU's from Google?

In the front page of the Wall Street Journal's Business and Finance section.

They'Re singing the Trainium chips praises. Amazon chips. Amazon's chips pose risk to Nvidia. The whole week we've been talking to people.

[Guest]

Does that clickbait?

[John]

I don't know. We're gonna find out. We'll see. It certainly doesn't seem good to have more competition in the, in the market. And take him. Came on the show yesterday to talk about how Nvidia was strong and, and really was not going to face significant headwinds from the TPU threat. Of course, Dylan Patel over at semianalysis wrote a 10,000 word piece all about how the TPUV7 was pretty good and Anthropic. Was going to be buying some and they were also going to be leasing some and they maybe had through midgate. And that sparked a lot of backlash from Nvidia bulls. And also folks who are really tied to amd, they're upset about it. There's a lot of losers if Google winds up winning with tpu. And so the losers came out to fight, apparently. But let's read through. Let's just get the facts down from Amazon's Trainium 3 launch. We of course had the CEO of AWS on the show yesterday and I asked him about this question. Will Amazon be bought buying tpu? I think that's an interesting question. But first let's see what Amazon's actually planning with their own AI.

[Jordi]

To be clear, he didn't. No cliffhanger here. He did not say yes or no. He just kind of.

[John]

I think I, I think you can read between the tea leaves and understand how the decision will be made, even though the decision has not been made yet. But we'll go through that. So Amazon.com is the latest big tech company to muscle in on Nvidia's turf. Give me a sound cue from the fall sun.

[Jordi]

How about this?

[Guest]

There we go.

[John]

That's right. On Tuesday, Amazon Web Services announced the public launch of its Trainium 3 custom AI chip, which it says is four times as fast as its previous generation of artificial intelligence chips. 4x speed up. That's actually very significant. That's great. The company said Trainium 3, produced by AWS's Annapurna Labs. Fascinating company. Acquired a decade ago for around 350 billion or 350 million. So it's pretty small acquisition actually. 350 million in AI, you never know. But back then you start a custom silicon company, you could barely clear nine figures on the way out the door. But in a partner labs has been working on custom silicon for Amazon for a long time. They actually do have a custom CPU at AWS to accelerate CPU based workloads. Then for the last few years they've been working on GPUs or ASICs for accelerated workloads. And so this custom chip design business, Annapurna Labs, can reduce the cost of training and operating AI models by up to 50% compared with systems that use equivalent GPUs. The chips are meant to provide a stronger backbone of computing power for software developers like Dean Leiters. Leiters, the co founder and co and chief executive officer of the startup Descartes, who we had on the show. And Descartes.

Is valued now at $3.1 billion. Let's go. So if you don't remember, Descartes came on and Dean was doing live AI video generation while he was doing the interview with us. It was really crazy.

[Jordi]

Yeah, he basically, yeah, it was real time. He looked like he was in a video game, but it was happening with little to no delay. Really, really cool demo.

[John]

Yeah. Before we move on, let me tell you about Restream 1 livestream 30 plus destinations. If you want a multi stream, go to restream.com so he said his company had a breakthrough enabled by Trainium3 chip. By the Trainium3 chip. After trying out several other competitor chips including Nvidia's processors, dozens of programmers and AI researchers from his San Francisco based company had been trying four months to train a version of Descartes flagship AI power video generation application known as Lucy that would be able to render footage in real time without bugs or hiccups. AWS gave Descartes early access to training 3. After meeting with the startup and being impressed with the founders, the company was two weeks into a marathon coding session in a rented house in Silicon Valley which I think he took us on a tour of while he was in wizard land and AI generated Sci fi world. It was very fun that a few of his employees were celebrating wildly behind him. Wait, well that's like a reference. I think that's a reference to the actual call that I'm referring to.

[Guest]

Weird.

[John]

This is very weird. Reading the journal. Yeah, I've experienced this. The moment that I saw it worked, I saw four people just start jumping up and down, said Dean. The next question was how fast can we get it to market and start changing industries with it? The launch of Trainium 3 is the latest broadside against Nvidia, which dominates the GPU market. A flurry of deals in recent months and have caught the attention of investors, indicating that more AI firms are seeking to diversify their suppliers by buying chips and other hardware from companies other than Nvidia. So Meta Platforms is in talk with Google to buy billions of dollars worth of advanced AI processors known as TPUs. And OpenAI has struck deals with rival Nvidia, rival AMD as well as Broadcom. And so very exciting that Descartes got good results out of the Trainium chip. That's awesome. Obviously I'm sure everyone over at Amazon has been working very hard on. At the same time we've heard that Anthropic maybe didn't have that great of an experience with Trainium and that's why maybe they're moving over to TPU a.

[Jordi]

Little bit more even though Amazon remains a major shareholder.

[John]

Exactly, yeah, Anthropic. And so my question is, will AWS buy TPU from Google? I asked Matt Garmin that question.

[Jordi]

You didn't ask me that question. Yes, I said they will be mocked.

[John]

They would be mocked.

[Jordi]

They would be mocked.

[John]

Which is ridiculous. And we'll get to why that's ridiculous. I mean, first off, it's funny to mock anyone for something related to their semiconductor supply chain and what they rack in their massive data centers. AWS is a massive business where I.

[Jordi]

Just say, please, my arch rival, can I please get some chips for my data center? To compete with your data center?

[John]

Okay, well, let's actually go to what Matt Garman, the CEO of AWS, said on TBVN yesterday, because I asked him, Will you be buying TPUs? And he said, hey, look, we' very excited about Trainium and I think it has. And we think it has enormous potential and we absolutely think there's a benefit to optimizing every layer of that stack. And so people were joking on the timeline, oh, there's this new Trainium chip. And somebody was like, all five people using Trainium are ecstatic that there's this new news. But probably Ballistic here says, Amazon's so bad at hype. Trainium is used by 500 million people through Bedrock, but their marketing team just can't. AWS is undervalued, blah, blah, blah. And he's obviously a bull on the stock. But what's interesting is that, like, it is. It is.

[Jordi]

I met some of their GTM staff today. Let's just say you'll have years to accumulate stock at cheap prices.

[John]

Everybody. And so. And so, like, yes, there obviously is value. Even if Trainium winds up being for a particular niche, like, maybe it's for real time video. Like, maybe that's the. Maybe that's what it gets really good at. It could get really good at diffusion. It could get really good. It doesn't need to just be like, your ASA can be honed and honed.

[Jordi]

And honed to the thing with real time video. That's interesting. Something that Descartes is focused on is working with live streamers, specifically on Twitch. Amazon owns Twitch.

[John]

Oh, that'd be cool.

[Jordi]

That makes that kind of partnership more interesting.

[John]

I like that.

So obviously there is value to saying, hey, if you go to aws, you can get Bedrock and some services that have been fine tuned specifically for Trainium. You go all the way down, you're gonna get very good performance because we have a stack from top to bottom that's very efficient. But at the same time, if you're trying to do something that's sort of like not within the Trainium ecosystem, you might have a rough go, you might wind up on a different chip. But he did say something. He said, we are going to support choice for our customers as well. And so we'll continue to offer GPUs from Nvidia as an example. And we have a very tight partnership there. So this idea of customer choice I think is important. And if you to Jeff Bezos, he said we're not competitor obsessed. This idea that Google is their arch rival, that's not in Amazon's DNA. Jeff Bezos said, we're not competitor obsessed, we're customer obsessed. We're customer obsessed. And so if the customer says, look, it's great that you acquired annapurna labs for $350 million, I'm really happy with what you've done with Trainium 3. It doesn't work for me, I'm the customer and I want you to give me an Nvidia GPU in your server or in your data center, or I want you to give me a TPU in your server. They might do that because that's actually in Amazon's DNA.

[Guest]

Yeah.

[Jordi]

And then the follow up question is, is there any world where Google sells TPU to Amazon?

[John]

Maybe, I don't know already they are partnering. Like this was another partnership that came out that Ben Thompson actually wrote about in Strategy, which you should go subscribe to. So separately there was an announcement of an AWS partnership with Google Cloud. Now they aren't buying TPU's, but what they're doing is they're enabling customers to establish private high speed links between the two companies, computing platforms in minutes instead of weeks. And so the general idea here is that Google has some amazing AI capabilities that customers are just struggling to match on AWS at this point. And the same thing's happening on Microsoft as well. Because on Azure you have access to OpenAI models that you might not have access to on AWS. And so even though your whole infrastructure might be on aws, you might be going back and forth to GCP constantly, or you might be going back and forth to AWS all the time, being like, oh, I gotta go over to aws, I gotta go, gotta go back, gotta go Azure, back to aws, back to Azure, back to aws. And so Amazon finally just said like, hey, look, we have a partnership and we're just gonna create a, like a dedicated pipe that puts these two systems together.

And so companies used to think about AI as a special piece of their application, so it would be fine to bounce around to another cloud to get the best possible results. But if the next generation of companies, I'm sure we'll talk to some of the AI focused yc Demo Day companies.

[Jordi]

Today about this, I hope there's at least one.

[John]

I hope there's at least one company that's doing something with AI. That would be a real treat.

[Jordi]

And if you're just tuning in, YC Demo Day coverage starts in 30 minutes.

[John]

Yes, so, but. So it used to be fine to bounce around. Now the next generation companies, they may be making their entire infrastructure decision based on who has the best AI products. What are you laughing at?

[Jordi]

I'm laughing because I texted Simon. They have Turbo Puffer has a booth at aws. I said, how's it going at Re Invent? And he says, I'm not there. I just make it seem like I'm there as a joke because the VCs keep going to the booth. And then our growth intern is like, oh, Simon, I don't know, I think I saw him over there.

Just continuing to mog while ARR skyrockets shout out to will the growth intern at turbopuffer holding it down at Re Invent.

[John]

That's fantastic. I love it.

So let me go back to aws. Amazon needs to fight back against this. And allowing high speed interconnect between AWS and GCP solved a piece of that. But will they go further back? On Tuesday, October 21, 2025, I wrote in the daily update in our [e-mail] about increasing competition in the AI supply chain. Here's what I said. I said, not every link in the supply chain can be completely commoditized. This is about OpenAI trying to dual source from every part of the stack. And I said, Nvidia has an insane amount of power right now. They've just ramped up full year revenue from 27 billion in 2023 to 60 billion in 2024 to 130 billion in 2025. That's like one of the greatest revenue ramps at scale in history. And then also they grew their net profit margin from 16% to 56%. That's insane. Insane. Yes, Goat. That's why Jensen Huang is on Joe Rogan. And I'm sure it's going to be a fantastic episode because he's got a lot to talk about, all the hyperscalers and OpenAI. But that creates problems, right? Because all the hyperscalers and OpenAI are now sort of incentivized to form a bit of an anti Nvidia alliance to commoditize the accelerator market and drive down those margins a bit. So 56% net profit margins on 130 billion of revenue. People are Just sitting there and they're like, there's $50 billion of profit over there. Like, that's a lot of acquisitions at Perna Labs.

[Jordi]

That's our cost.

[John]

Yeah, that's our cost. You're just eating a lot off of these plates. And so CO2, I think, has done a good job explaining the current state of the anti Nvidia alliance. They call it the Google complex, which is probably a little bit better. That consists of Google, Broadcom, Celestica, Lumentum and TTM Technologies. This coalition stands in contrast to the OpenAI complex that consists of NVID, SoftBank, Oracle, AMD, Microsoft and Core Weave. But you know who they left off the chart entirely? Amazon. Amazon doesn't fit neatly into either of this.

[Jordi]

CO2 just loves, I think, I think they just love leaving a major player off any sort of graph or chart that they make. Right. They left Google off of their top fantastic 40 AI companies. So I think that's just a little, that's just, that's just them messing around a little bit.

[John]

But there's. Yeah, I mean, I think it's accurate. I like. If you said, is, is Amazon more aligned with OpenAI or Google, you'd be like, what are you talking about? Neither. That's correct. They're not in one of the complexes. Maybe they need to be, maybe they don't. Maybe they will, you know, form their own complex outside of it. But I just think it's interesting that I agree with you that it's like it was ridiculous to consider the idea of them buying tpu. That feels so uncharacteristic. And yet they serve up plenty of competitive products within aws. And.

You go back to the early days of Amazon, you can get Amazon Basics paper towels. You can also get name brand paper towels. And that exists within the AWS stack from the databases that they have on offer. There's a lot of.

[Jordi]

You should rebrand Trainium to Amazon Basic.

[John]

Gpu, Amazon Basics Accelerator, Basic BASIC chips. Basic chips. It was on Basics chips. It would be good. Really, really hard. They're like, actually, it's like one of the greatest things ever. It's the most incredible thing that America or that humanity has ever created. It's extremely difficult to make. We taught Sand a thing.

Anyway. I just don't think Trainium 3 is the, you know, obviously everyone at AWS is like excited about it and it's a big, it's a big deal, but it's just not the backbone of their business. And in the long term they might just retreat to supporting choice for their customers. And so, you know, I keep going back to that Jeff Bezos line. We're not competitor obsessed, we're customer obsessed. And so I wouldn't be as surprised.

[Jordi]

How much do you think it hurts Amazon that they don't have a dedicated podcast guy? Like, they don't have a Sholto, they don't have a Sam, they don't have a Satya.

[John]

You know how much that hurts? Because they definitely have someone in that role. You just don't know them.

[Jordi]

That's what I'm saying.

[Guest]

Yeah, they don't have someone who.

[Jordi]

They might have. They might have the title, but they're not really in. In the driver's seat.

[Guest]

Right? Yeah.

[John]

They don't have a Rune.

[Jordi]

They don't have the rune. Right. They don't have a Sholto.

[John]

Yeah, they should step it up. They should definitely get someone. I'd love to see it. Well, fortunately, I mean, the Semianalysis crew was over there taking pictures, sharing photos in the timeline of the Trainium 3 Ultra server. Liquid cooled with a lot of hard eyes. That's some good news from. That's a glowing endorsement from the Semi Analysis crew. And look at this very purple. I wonder if that's like intentional. I wonder if they set up the purple lighting. There's a bunch of funny things going on over at Re Invent. It's also just like it's a punishing time of the year. I guess it's like right before the holidays or something because we've just been completely torn. We obviously wanted to go to YC Demo Day. I also wanted to go to Neurips, which is going on right now. The Premier AI Conference. There's also DealBook Summit. Andrew Sorkin's doing all the greatest interviews. At the same time, there's Re Invent. I wanted to go to that crazy.

[Jordi]

Interviews coming out of DealBook. I just saw some clips this morning. You got Scott Besant just going hard. You got Alex Karp going hard. No real surprises on either of those fronts, but excited to get the update there.

[John]

Let me tell you about Cognition, the team behind the AI software engineer Devin. Crush your backlog with your personal AI engineering team.

Let's close out the Trainium coverage with the Zephyr post. Who says Google is having this kind of success with TPUs? What about Amazon's Trainium? Trainium is new and underpowered. Just 667 TFLOPS BF16. It has lots of HBM, but the bandwidth is lower than the H100TPUV6e is competitive with H100 not on HBM or bandwidth. And Ironwood is competitive with Blackwell on FLOPS bandwidth and HBM capacity. I expect Ironwood to quickly gain market share as it ramps up. As you can see from throughput slash TCO Nvidia vs. Trainium Rubin Mogs Trainium 3 harder than Blackwell vs. Trainium 2 on TCO training flops and reduces the gap by 5% on TCO MEM bandwidth. So the gap between Nvidia and Trainium is actually increasing rather than decreasing. By the way, this math was done before CPX was introduced. I won't be surprised if CPX plus Ruben is cheaper than Trainium for inference. So I do think that there's a world where.

Where there's something specialized like what's going on with Descartes. Some sort of special model.

That thrives in what Trainium is good at and they can further niche down. But we'll see. I mean maybe they come from behind and they just destroy tpu and we're all talking about Trainium next year Anyway. Let me tell you about LINEAR Meet the system for modern software development. Linear streamlines work across entire developments cycles from roadmap to release. We gotta say a little rest in peace.

[Jordi]

Rest in peace. To Claude.

[John]

San Francisco's beloved albino alligator has passed away at age 30. That's a good age. I don't know how long alligators typically live, but I'm glad looking it up.

[Jordi]

30 to 50 years for the American alligator.

[John]

A little bit short but Claude was.

[Jordi]

Of course often supporting, often reaching 70 years or more.

[John]

Yes.

[Jordi]

Anyways RIP.

There was you know, obviously people started speculating immediately. Anthropic of course was the sponsor Claude. And you know people were wondering was, was there foul play involved? Was it possible this.

Poor dinosaur, not dinosaur alligator passed the day that they, that, that it, that it got announced that they've hired IPO lawyers. Some people were speculating is it possible Claude was sacrificed to the capital markets gods in some type of ritual? But anyways.

Look at this expression he has on his face. Can we zoom in a little bit?

What a cool guy. And he will be remembered.

[John]

Yeah. Dan Primack here is talking about Xlight. I think we might have the CEO on the show soon. The Trump administration will invest $150 million into a lithography startup called Xlight. Its first chips act award chatted this morning with XLite CEO. There's a few lithography companies now. We've had some on the show. This feels like an entirely new.

It's a very Interesting tier of investment, like $150 million from the government, that feels like a Series B.

They did raise a Series B this past summer, led by Playground Global, with Playground partner and former Intel CEO Pat Gelsinger becoming X Lite's executive chairman.

[Guest]

Whoa.

[John]

And so makes sense that the government's investing in Intel. Pat Gelsinger, of course, former Intel CEO. Now he's getting involved in XLite. Marshaled 40 million of capital, went and got 150 from the government. The story continues. There's also another AI startup that wants to remake the $800 billion chip industry. This one's in the Wall Street Journal. Founded by ex Google researchers, recursive intelligence raised 35 million with backing from Sequoia to automate chip design. Obviously this is not lithography, this is the design process, but still, companies are.

[Jordi]

Dylan Patel was talking a little bit.

[John]

About this, the stack. Oh, he did? I didn't hear about that. Very cool.

[Jordi]

This is AI for. For AI chip design.

[John]

Oh, that's right, yes. AI for AI chip design. Everything we need. On a quiet residential street a few blocks from Stanford University, two former Google researchers are launching a startup they hope will remake the $800 billion chip industry. Anna, Goldie and Azalea.

Are trying to build software that can automate the design of cutting edge chips. A prospect that would allow every company to build their own chips from scratch. Working from the top floor of a suburban home, the duo recently raised 35 million to kickstart Recursive Intelligence with funding from Sequoia Capital and Strike. We got it.

[Jordi]

The recursive.

[John]

We got to add.

[Jordi]

Putting it in the name.

[John]

We got to add that to the list of. Because there's standard capital, modern capital, standard intelligence, modern intelligence, raw intelligence. Raw Intelligence was the low hanging fruit. Applied was another one.

[Jordi]

Cap Intelligence.

[John]

And then what was the other one? There's. What's Lockheed Groom's company?

[Jordi]

Physical Intelligence.

[John]

There's physical intelligence, Physical capital. So it's the matrix of like capital. What was it? Capital, intelligence and intuition or something like that, and you multiply them all out and you get the whole thing.

[Jordi]

Eventually we're going to run out. Right. There's somewhat finite.

[John]

No, there will always be more names.

[Jordi]

Startup name generator, new words.

[John]

So wow, the company. 35 million for evaluation of 750 million. That's a very low dilution. What, 5% or something like that? Pretty remarkable. Definitely.

[Jordi]

VCs.

[John]

Were MOG VCs high? Yeah, I would have assumed this would be a very capital intensive business, but I suppose if it's just software that they're developing.

Maybe they have more control here. Companies such as Amazon and Google have developed custom chips for AI and data center use. And Apple saved billions of dollars by insourcing chips for its devices, including the M series chips that have helped revitalize its MacBook laptops. Such silicon options can be cheaper, more.

[Jordi]

I had a funny moment yesterday. We got an Amazon package and it was covered with like Alienware, like Alienware branding. And I asked, I asked Sarah, I was like, did you get something from Alienware? Like what is going on? And it turned out to be an ad. But they were advertising that is powered by Intel.

[John]

Oh, interesting.

[Jordi]

Which didn't make me necessarily want to immediately buy an Alienware device.

[John]

What do you do? You put the money straight back in your pocket because you're a taxpayer, you own Intel.

[Jordi]

That's true. That's true.

[John]

You should support Intel. No, intel is undisputably great for gaming. There's no question there. The question is, are they going to be able to build a fab that competes with tsmc? It's a completely different question. I might go build an Alienware intel PC.

[Jordi]

Oh, we're going to for the office sim racing rigs.

[John]

The sim racing rigs.

[Jordi]

Get ready.

[John]

Needs to be intel inside for sure. For sure.

[Jordi]

This is just going to turn into a sim racing show where we watch other podcasts while sim racing and reacting to it.

[John]

Yeah, I like it. Vanta Automate compliance and security AI that powers everything from evidence collection and continuous monitoring to security reviews and vendor risk. Dwarkesh Patel has a massive essay shaken up the timeline thoughts on AI progress. He says he's moderately bearish in the short term, but explosively bullish in the long term. Well, very interesting. So he says he's confused why some people have short timelines. They say AGI is coming soon, but at the same time they're bullish on RL VR which is reinforcement learning with verifiable rewards. And so he says if we're actually close to a human like learner, this whole approach is doomed. Currently the labs are trying to bake in a bunch of skills into these models through mid training. There's an entire supply chain of companies building RL environments which teach the model how to use Excel to write financial models. For example, I think we're actually talking to an AI XL analyst for Excel power users called Crunched at 1250 YC Company. I think that these are good ideas. I'm actually very bullish on.

This model. But in the context of when Does AGI arrive? When does super intelligence arrive? I understand Dwarkesh's point. He says either these models will soon learn on the job in a self directed way making all of this pre baking pointless, or they won't. Which means AGI is not imminent. Humans don't have to go through a special training phase where they need to rehearse every single piece of software we might ever use. Barron made interesting points about this in a recent blog post. When we see frontier models improving at various benchmarks.

We should think not just of increased scale and clever ML research ideas, but billions of dollars spent paying PhDs, MDs and other experts to write questions and provide. Let's give it up for the example. Let's give it up for PhDs and reasoning targets.

[Jordi]

Let's give it up for the experts.

[John]

These precise capabilities In a way this is like a large scale reprise of the expert systems era where instead of paying experts to directly program their thinking as code, they provide numerous examples of their reasoning and process formalized and tracked, and then we distill them into models through behavioral cloning. This has updated me slightly towards longer AI timelines since we since given we need such effort to design extremely high quality human trajectories and environments for frontier systems implies that, but they still lack the critical core of learning that an actual AGI must possess. This tension seems especially vivid in robotics. In some fundamental sense, robotics is an algorithms problem, not a hardware or data problem. With very little training, a human can learn how to teleoperate current hardware to do useful work. So if we had a human like learner, robotics would in large part be solved. But the fact that we don't have such a learner makes it necessary to go out into thousands of different homes and factories and learn how to pick up dishes or fold laundry. One counter argument I've heard from the takeoff within five years crew is that we have to do this kludgerl in service of building a superhuman AI researcher. And then the million copies of automated ILIA can go figure out how to solve robust and efficient learning from experience. This gives the vibes of we're losing money on every sale, but we'll make it up in volume. This automated researcher is somehow going to figure out the algorithm for AGI, something humans have been banging their heads against for the better part of a century while not having the basic learning capabilities that children have. That seems super implausible to me. Besides.

Even if that's what you believe, it clearly doesn't describe how the labs are approaching RLVR you don't need to pre bake the consultant's skills at crafting PowerPoint slides in order to automate Iliad. So clearly the lab's actions hint at.

A worldview where these models will continue to fare poorly at generalizing and on the job learning, thus making it necessary to build in the skills that they hope will be economically valuable beforehand. I want to go to the section on economic diffusion, but first I'm going to tell you about Privy. Privy makes it easy to build on crypto rail, securely spin up light label wallets, sign transactions, and integrate on chain infrastructure all through one simple API. So you've been asking about economic diffusion. What is the rate that we're diffusing? Let's see what Dwarkesh has to say about economic diffusion. He says that economic diffusion lag is cope for missing capabilities. And so this is also seems informed by the Tyler Cow intake that AGI is here. The models are good, but it just takes time to adopt them. And I'm very sympathetic to this because when I go to the doctor's office and they hand me a piece of paper, I know that a web form is good enough. Like the capabilities of the digital form are complete. It's not that the form is lacking in something or it's not reliable enough. It's not like they're like, oh yes, like the website goes down 20% of the time and so paper makes more sense still in this case it's like, no, it's just a diffusion problem. There's just someone who runs that doctor's office is like, I like doing it the old way.

[Guest]

Right.

[John]

And that's the economic diffusion lag problem that I think is real in a lot of scenarios.

[Jordi]

But the missing capabilities thing, I mean, just to give a pretty concrete example, right now AI is great at generating text, right? It's great at kind of analyzing a piece of content and then generating text based on that. And yet we still have multiple people on the team at TVPN whose job is to find interesting moments of the show and then create captions around that and share it to X and Instagram and YouTube and other platforms.

[John]

And Dwarkesh experienced that too, where he was trying to find the most interesting pieces of a full podcast with one big Gemini prompt, and he was trying all the different models and couldn't get it to actually find the most salient and viral points.

[Jordi]

Yeah, so one of the other thing that stands out is one of the seeming missing capabilities is ability to identify humor or even something like it's Almost emotional. So Ilya and Dwarkesh talked about this, where I think Ilya was giving the example of scientists studied people who had had various brain injuries that limited their ability to experience emotion. And when they took out emotion, it took them. It can take somebody two hours to figure out which pair of socks to choose. And they were kind of, like, stunned. It's just a pair of socks. You know, what's going on in your day? Why do you need emotion in order to make that kind of decision? And so it seems like, at least in AI, a missing capability is like, okay, finding out what's an interesting moment of a podcast in Dwarkesh's case. Right. Is it something that makes the audience member feel something?

[Guest]

Right.

[John]

I mean, there's just so much to pull through. Like, I remember during the Carpathy interview, I was watching it and Tyler was watching it, and there's this moment where Carpathy says, like, the coding models are amazing and they're magical, but what they produce is slop. And it's like that word slop is so. It's like the word of the year, or maybe the word of last year. Like, it's a huge word. It has a huge amount of weight coming from him.

[Jordi]

It's crazy. That rage bait, beat out slope for.

[John]

The word of the year. Slop is probably the 2024 word of the year or something like that. But anyway, the point was, like, when. When. When I heard that, when Tyler heard that, that. That word, Karpathy calling it slop, everyone was like, whoa. And I was like, we should clip that. And we looked, and it had already been clipped by a human. Like, yeah. Someone on the timeline had also identified that it was like, that was the crazy moment that we should be, like, reacting to and taking in.

[Jordi]

And it's crazy. The other thing that's. That's notable is, like, on wap, one of the best. One of, like, the top jobs that people do on wap, or way they make their first dollar online is just like clipping for various content creators and media companies. And some of the clips that they make are so sloppy. Like, it's literally just like a random segment of the show and they're blasting it out from, like, 20 different accounts. And the fact that people are still paying humans to do that, but still, I mean, it just feels notable.

[John]

Yeah. Well, let's read Dwarkesh's take on economic diffusion. Lab Lag being Cope for Missing Capabilities.

[Jordi]

Says sometimes Copium would be a beautiful name for an AI chip, by the way.

[John]

It would. It would you got trainium.

[Jordi]

Maybe they need copium.

[John]

Sometimes people will say that the reason that AIs aren't more widely deployed across firms and already providing lots of value outside of coding is that technology takes a long time to diffuse. Dorcas thinks this is cope. He says people are using this COPE to gloss over the fact that these models just lack the capabilities necessary for broad economic value. Stephen Burns has an excellent post on this and many other points. He says new technologies take a long time to integrate into the economy. Well, ask yourself, how do highly skilled, experienced and entrepreneurial immigrant humans manage to integrate into the economy immediately? Once you've answered that question, note that AGI will be able to do those things too. Dwarkesh says if these models were actually like humans on a server, they'd diffuse incredibly quickly. In fact, they'd be so much easier to integrate and onboard than a normal human employee, they could read your entire slack and drive in minutes and immediately distill all the skills that your other AI employees have. Plus, the hiring market is very much like a lemons market where it's hard to tell who the good people are beforehand. And hiring someone bad is quite costly. There's. This is a dynamic that you wouldn't have to worry about when you just want to spin up another instance of a vetted AGI model. For these reasons, I expect it's going to be much easier to diffuse AI labor into firms than it is to hire a person. And companies hire lots of people all the time. If the capabilities were actually at AGI level, people would be willing to spend trillions of dollars a year buying tokens. Knowledge workers.

[Jordi]

Think about that. We hire someone like we hire an A or we're leveraging an AI and they've listened to every single minute of TVPN ever and watched every clip.

[John]

And right now you'd have to fine tune that into the model or whatever. You don't just get that out of the gate.

[Jordi]

Yeah, and I'm just saying, like, we do end up hiring a lot of people that are previously just listeners. But getting somebody that knows every single moment that has ever happened on the show would be super powerful. But again, there's just a missing capability set that doesn't allow agents to deliver a lot of value internally.

[John]

The reason that lab revenues are four orders of magnitude off right now is that models are just nowhere near as capable as human knowledge workers. Yeah, I agree with that. The one thing that I don't necessarily agree with here, he says, well, ask yourself this quote from Stephen Burns. How do highly skilled, experienced and entrepreneurial immigrant humans manage to integrate into the economy immediately? I mean they do sort of integrates into the economy immediately. But like the immigration flow is like a slow process. Like it doesn't just happen immediately. It's not just like, you know, the amount of immigration went from like zero to like, I don't know, a million people or something. Like it's like people move around. There is like a, there is a bit of a drag. But I understand what he's saying here and it does make sense. Anyway, let me tell you about public.com investing for those who take it seriously. They got multi asset investing and they're trusted by millions of.

The Verge is trying to get it on the action. Trying to attack David Sachs with a headline. It's like, it's like so funny that the New York Times went after David Sachs and then the Verge was like, we want to go after him too. We want to get some of the hate.

[Jordi]

Wait, wait, let us, let us cook. Let us cook.

[John]

We heard tech hates, you know, this, this art.

[Jordi]

I do think he too, while I don't agree with this journalistic approach, it is a pretty funny headline.

[John]

Yeah. Oh yeah, it's hilarious. The headline is Silicon Valley is rallying behind a guy who sucks. It's like, what does that mean? Just pure, pure qualitative, just name calling. They're just like, we don't like this guy.

[Jordi]

Pure ad hominent.

[John]

But you know, go off. If people, if people, if your fans like it, if that's what your, your audience wants. It's, it's, it's rage bait. It's gonna go hard. It already got a thousand likes on a linked article. The Verge is not putting up a thousand likes per link. So this is outperformance and it's heavily paywalled. You cannot learn how David Sack sucks without subscribing to that thing. They did a good job. You gotta pay.

[Guest]

You gotta pay.

[John]

You wanna know why he sucks? I did.

[Jordi]

You pay.

[John]

I don't know why he sucks, but.

[Jordi]

That'D be really funny if behind the Paywall is like, we're just kidding.

[John]

He's actually.

[Jordi]

We think the New York Times missed on this one.

[John]

Who knows?

[Jordi]

Paul Graham on the timeline, he says a startup told me that one of their investors didn't like that they were selling to newly founded startups and wanted them to sell to bigger companies who have more money. If investors tell you this, write them off as idiots. Selling to startups is the best thing you can do, I'm sure Many of the companies we're talking with today will be selling to other companies in the batch.

A lot of people.

Say that's bad. They try to say YC is a circular economy, but you have to ignore the hundreds of very real businesses that have been been created through YC and gone on to work with every kind of company in the world.

[Guest]

Yeah.

[John]

It certainly seems at this point startups tend to be smarter, less bureaucratic, more representative to future trends. Like, even if there's some sort of insular circular economy in the startup ecosystem, there's a pretty immense amount of pressure to actually deliver something that's valuable. Because every dollar is precious.

[Guest]

Yeah.

[Jordi]

And these are every. Yeah, they're being rational. It's not like. I'm sure there's been small instances where companies were actually, you know, had somewhat bad behavior. But in general, it's like, if I'm going to pay for the SAS tool or the beta that you're running, it has to be good.

[John]

Yeah.

[Jordi]

Has to work.

[John]

Did you see Stuart Brand? He says so there's a $1.5 billion judgment against Anthropic for including 480,000 books in training, their AIs. Five of my books are among them. Word is there might be $1,500 payout per book, according to my agent, Max Brockman. That's a good name. He said, I wrote them. I wrote to my agent Max, the following. If any payment comes to me, please send it back to Anthropic with my thanks for including my books and their eyes. The judgment website offers a way to opt out of the payment, but I found it cumbersome, so I didn't in principle, but too lazy to be high. Highly principled. I really like this. He's the co founder of the Long now foundation, which takes no sides in this forum. As a private person, I do take sides occasionally, so I thought that was a funny thing. There is secondary market fraud going on left and right. But first let me tell you about graphite.de code review for the age of AI. Graphite helps teams on GitHub ship higher quality software faster.

[Jordi]

Yeah. Reading through this, Matt Grimm says secondary markets are rife with fraud and bad actors. And it pains me to see these bottom feeders profiting off of Anduril's leasing retail investors through unreasonable or opaque fee structures. In this week's episode of Nonsense, Ignite vc, a fund we've never taken a meeting with or had any contact with whatsoever. Founded by Brian, who we've never met as soliciting investors. Via public Google Doc to invest in an SPV that will in turn invest in another SPV that will in turn potentially enter into a forward contract with a supposedly though unnamed early Enduril employee. A few problems here. First off, so called forward contracts are notoriously hard to settle in private companies and counterparty risk is extremely real. What about the many complicating corner cases like acquisitions where shares don't trade or marriages, divorces or deaths where ownership of the underlying shares is complicated, just generally a risky structure to close that I don't think most folks actually understand. So yeah, if you enter into a forward contract and you basically buy the right to the future value of some shares and then somebody gets again married or divorce or passes away or bankruptcy is another situation where you might not be actually able to collect even if your investment should have generated some return. Matt says Second, this deal memo includes basically no details about Anduril's performance, no revenue figures whatsoever, no product specifics. I guess that's good right if they were just floating around information that they had acquired. But anyways, continuing almost as if it's soliciting investors to invest on hype and momentum and not fundamentals. Generally I'd advise folks to be skeptical of any deal memo lacking basic details. Third, forward contracts are explicitly disallowed by Anduril's stock plan and bylaws, which means that Anduril will never consent to Team Ignite's SPV actually taking possession of these shares while we are privately held. Zero chance. And finally the memo spends most of its time talking about the structure and fees which are insane. A double layered SPV with all legal and admin costs pass through in addition to an 8% upfront fee, 3% annual fee for two years, 20% carried interest, and the craziest part, an implied price per share that is completely insane. In this case the implied pps is 115% higher than the most recent preferred raise from 9 months ago. Flattered I suppose, but also puts these investors in an almost absurd position by paying more than double the price per share of our most recent transaction as stated at the top. I don't know Brian or Team Ignite at all. Maybe they're kind of wholesome people and this is all a big misunderstanding. But if I were an investor looking at this quote opportunity quote I'd run for the hills and I believe the founder replied and said appreciate the heads up. The document reference was an internal draft prepared for discussion with an existing LP and was not intended for public circulation. It appears Someone shared it without authorization and we're looking into how that happened.

[John]

But do you see what happened? There's like seven people share a screenshot of like a direct email we got with this exact number.

[Jordi]

Okay. And the other thing is they say not soliciting investment for any andural related vehicle. Matt says, really, the draft was written by your founder and managing partner. I literally watched him edit the doc in real time and he has a screenshot, screenshot of like the founder's name in Google Docs like, you know, basically.

[John]

What a mess.

[Jordi]

Anyways, well, don't do this.

[John]

Don't do it. Instead, why don't you start a company and apply to Y Combinator, build an actual business instead of going around hustling SPVs and companies that don't want to sell shares.

[Jordi]

But we are moving on to double layer SPVs.

[John]

2 hours y combinator coverage. We have Harsh Taggar here in the Restream waiting room. Let's bring him into the TVP and ultra harsh. Thank you so much for taking the time on a busy Y Combinator demo day to come talk to us. How are you doing?

[Guest]

I'm doing good. Thanks for having me.

[John]

Fantastic. Take us through. How's the day going? What is the schedule like? And then I'd love to dig into some of the trends that you're seeing. Some of the standout companies. I'm sure we're going to be talking to a lot of them, but.

What'S the run of the show today and where are we in the course of the process of, of graduating these companies?

[Guest]

So we got started almost a couple of hours ago, 10 in the morning. And so the founders kind of investors all gather together. They get into the main room here at the YC office and then the founders start giving presentations, talking about the progress, what they've built themselves, their background, the pretty quick fire presentations, one minute each and then there's like a break in between sort of blocks of presentations where the investors can hang out and talk to some of the founders and get to meet them and obviously hopefully invest in a bunch of them. So that's kind of. We're just about approaching lunch, so it's kind of like that part of the day where people have listened to a bunch of companies, probably got a sense of some of the stuff that they're interested in. I see people right now just hanging out, doing deals. So it's kind of like a fun vibe. Like live.

[John]

It's the best party rounds. We love it. I wish we could be there. Are there any Hero metrics or stats that, that the YC team shared this year to kick off demo day. How are you sharing like the shape of YC these days?

[Guest]

Yeah, I mean we didn't go too stats heavy this time around, I think. I mean at a high level it's just the continuation of the theme we've seen this whole year, which is just like the companies during the batch are just getting faster revenue growth, assigning contracts with like big companies, in some cases even like government, defense tech. The dollar value contracts that startups can close in the first few months of their life are just bigger than anything we've ever seen. And that's all very directly from AI. So it's just.

[Jordi]

Yeah, and it's a very, it's very interesting kind of approach. You can sign one big contract and generate enough revenue to go on the stage at demo day and feel confident in your pitch and have something that's compelling, or you can go and sign up a bunch of startups to something smaller plans.

[Guest]

But you see that disseminate, there's companies that keep growing. In the SaaS world you were used to just see consistent month over month growth and now in AI world you're used to big step function growth and they might be flat for a month, but then you sign another contract and it just leapfrogs again.

[John]

Interesting. Yeah. Help me square sort of the shape of revenue with some of the YC batch that we might talk to today. Because Paul Graham was on the timeline sort of defending this idea of selling to startups. We were in complete agreement with that, that selling to startups can be so much better in a bunch of different ways. But it does feel like we're entering an era where maybe it's AI, maybe it's just the maturity of the ecosystem. Like it's also been easier than ever to sell to the government or to sell to Fortune 500. And so are both happening? Are there specific companies that are really great at one or the other? Is there any advice that you've given founders on how to, to decide between those two paths?

[Guest]

Yeah, it really depends, I think on the type of product you're building. So I think the bull case for selling to startups as your customers is like the stripe or the AWS case. And it's like you get them all early. I mean you could put gusto, rippling deal into that bucket as well. It's like if you get the startups early and you can grow with them, that is one of the most powerful business models you can have, right? The Stripe team could go on vacation for two years and they would just keep growing because the cohorts would just keep going up and to the right. I don't think they're going to do that anytime soon, but they could if they wanted. So I think if you have a product like that where you can grow with the startups and you can get in early and they will just. Those startups in the future will become your enterprise customers. That's fantastic. That's absolutely what you want to do. I just think with AI, what's new is you didn't even have the option of selling to a big customer until you sold to startups. And you'd build up like, oh, hey, we don't have an enterprise customer yet, but we got like a thousand startups and in aggregate we're processing X or like we're reliable. We're not going to shut down. I think now with AI and the fact that the incumbents can't actually build the products because the engineers that work at these bigger companies don't even believe in AI. So startups in the batch are able to go to a big company and actually get them as a customer because they're the only ones that can actually deliver the product. And I think that's just new. So we still give the advice. It's very dependent on the company and the product. And like, will you be able to scale with startups or not? But like, in general, there's just more options as a founder for how you do sales than there's ever been.

[Jordi]

Let's, let's talk about themes in the batch. Two batches ago, I've felt like a lot of the companies were. At least the ones that we talked to were like various. Like infrastructure was like infrastructure for building agents. Last batch really felt like much more applied. It was like applying AI to very specific industries and opportunities. I'm curious, I'm sure you're seeing both of those kind of types of companies, but looking at the list of guests that we have today, bunch of super exciting companies, but curious to know kind of like broad themes across the batch.

[Guest]

I mean, I think you say, right. I think what we've seen is that like maybe a year ago, just a year ago, it was like infrastructure, infrastructure to build agents, like you're saying, like laying the foundation. Then it's like vertical agents just take off, like customer support, logistics, like name any healthcare, like all these verticals and they're just like taking off. And primarily what they were doing is selling these agents to the companies. In those verticals to make their operations more efficient. I think what seems to be a theme coming out of this batch, you'll notice, is the companies are going the next step and they're not actually selling the agents to the incumbents. They're going AI native, full stack. They're just actually doing the thing. So you have Fernstone being an AI native insurance brokerage. They are insurance broker. They're just going to use AI to be the best one. Sava is doing that with trust. It's a company that sets up trust, but it's doing it with AI. So I think that seems to be the new trend is going AI native and not just selling your agents, but using them to build the company, doing all of the stuff.

[John]

Yeah, we've talked to a couple law firms that have done that and also investment banks. Just people who have said, okay, we actually need to go do the core thing. I'm always reminded of Justin Kahn's company because it feels like Atrium was just a little bit early to that model and now everyone's working on it and it's starting to maybe work, and we'll see.

[Paul Graham]

Yeah.

[Guest]

I think if you go back, do you remember it was, I mean, it was like a decade ago now, but it was Balaji that started this whole thing with like the Full Stack startup.

He had this blog post and like, I don't know if you guys were in San Francisco at the time, but like, there was this moment where there was Doordash, which was delivering food, and then you had Spoon Rocket and Sprig, which were like the Full Stack version. Because what they did is they had these kitchens, like these vans, which had little kitchens driving around San Francisco cooking the food.

[John]

Right.

[Guest]

So I think like, like back in that era, it was seen as being the most ambitious thing to be a Full Stack startup. You didn't just sell your software, you did the whole thing. Ultimately, those companies didn't. It turned out that being a marketplace or selling software was just a better scalable business in that era. But now with AI, I think the promises were kind of going back to the Full Stack startup idea. But this time we're all hoping and kind of seems like these things will actually scale because you don't need to hire a thousand people to do the work. You just keep improving your agency.

[John]

Yeah, yeah. I mean, the food example is interesting because it feels like Travis Kalanick is maybe dipping his toe in like, oh, what if I did the Full Stack thing?

[Jordi]

Yeah, he's got Picnic and I think it's Otter and Then he has cloud kitchens. He's like, I can do it.

[John]

But maybe at his scale, maybe it's a scale thing, I don't know. But it is more complicated financially.

[Guest]

I think if you have Travis's access to capital and his background operating that, you can do that.

[John]

How are companies or founders grappling with what's happening at the largest foundation model labs? I remember there was some Sam Altman interview where he said, here's how not to get steamrolled.

If your entire business is just predicated on the model not getting better, you're going to have a bad time. But if you're doing something completely separate with the model, you're probably good. How are people thinking about it in the more modern context?

[Guest]

I think the framework people have on this stuff is that they expect Sam and the big lab companies, OpenAI in particular, to go after probably, maybe more of the sexy consumer ideas that capture the public's imagination and it's going to be hard to compete with them on that. But there are like the startups in the batch in particular focus on just like the unsexy verticals, like building an audit firm, building a legal firm, building insurance broker. Like the bet they're making is that like the best people at OpenAI are anthropic, are not going to be thrilled to build like auditing software or auditing agents, you know, and so.

[Jordi]

Or actually sell the. Or actually sell the end service.

[Guest]

Yeah, exactly. Like doing it like going like all the way and like learning what that customer wants and how to do it really well and like iterating on it a thousand times to get it.

[John]

Yeah. This is the whole thing with Google versus Amazon. Like Google did wind up building a shopping product, but they never really had that in them to be like, we're going and doing warehouses and we're going to compete with Amazon even though we want E commerce, like we don't really want it that badly. That sounds actually sort of miserable and it's just not.

[Guest]

People don't want to do it.

[Jordi]

Right.

[Guest]

Like the best engineers at Google don't want to build a shopping product. They are like back in the day they want to wanted to work on search quality. Now they probably want to work on Gemini.

[John]

Totally. Yeah. And there's also just cultural, I feel like culturally there are certain companies where like if you're like, we do 80% gross margin work and you show up and you're like, I'm the guy who does 30% gross margin work, they're like, you can leave the company actually like we don't like you at all. So yeah, you know, your margin is my opportunity. Both directions sometimes.

[Jordi]

What are some companies from previous batches that you really feel like are hitting their stride now? Uh, we had Kalshi on yesterday for their $11 billion round. I don't, I don't think a lot of people are even aware that they went through YZ because it was so, so long ago.

[Guest]

Right. Yeah, there was, that was 2019, I think. So. Yeah. I mean obviously Kalshi is like the prime example of a company that just made a bet on a space early and like had to just wait for the market to actually exist for it. And like those founders, like super tenacious went for it. But I think more recently there's a company that announced around doing customer support called Giga, which I think is a really exciting one. They're competing with Sierra and Decacorn, superstar founders of those companies. Tons of capital raise but they've been able to beat them on head to heads with customers like doordash.

Through technology really. So I think Giga seems to be really growing.

Another one like Non AI. That's where Post Hog is actually a little bit more under the radar. But they are sort of like taking the rippling approach of.

[Jordi]

Yeah, they're launching a new product like every week it feels like.

[Guest]

Yeah, it's like really interesting to see. They've done that from day one and it seems to actually be compounding and working the way that it has for rippling. So I'm curious to see if you start seeing more of that. Just like startups trying to build multiple products from day one and have like the compound startup effect.

[John]

I like animal themed companies. I like Post Hog, I like the Hog themed. When we did our first demo day stream we talked to a company called Pig and we really like Pig and it stuck with me and so I'm rooting all. Gotta check in with you all the swine themed startups. I hope they all do very well.

But thank you so much for taking the time to kick this off with us. Congratulations on the big day.

[Jordi]

Great to have you on for the first time. Yeah.

[John]

And we gotta do this more often.

[Jordi]

Come back.

[Guest]

I would love to. Thanks for having me.

[John]

Yeah, let's talk. Have a good one.

[Guest]

See you guys.

[John]

Goodbye. Our first guest will be Clad Labs, makers of the Chad Ide. First, let me tell you about Julius AI, the AI data analyst that works for you. Join millions who use Julius to connect their data, ask questions and get insights in seconds. We have Clad Labs while We wait.

[Jordi]

I have some other names for if you're launching a startup and you want a pig themes name, swine themed name. You could have Wilbur, Babe, Hamlet, Daisy, Peanut and Cookie.

[John]

Okay, I like that Ham solo. I like Babe. I think Babe mud pie. Okay, so we have the founder of Clad Labs in the Restream waiting room. Let's bring him into the TVP and ultradome.

[Jordi]

What's going on?

[John]

Look at this shirt. They look fantastic.

[Jordi]

Incredible.

[John]

Incredible shirt. You know you're winning me over already. Break it down. First introduce yourself. Tell us what you're building. Good to meet you.

[Guest]

How's it going guys? Yeah. I'm Richard, the CEO of Cloud Labs. We're building Chat ide, the world's first brain rot ide.

[John]

Okay.

[Jordi]

Why so great? So we exchanged some comments and wanted you to come on the show. I think you get the TVPN award for the best rage bait at the product level of the year. And I thought your response to the essay that I did was amazing. You were like, like, cool essay. Unfortunately, it doesn't apply to us.

[John]

Yeah. So why doesn't it apply? What are you actually building? Like why brain rot? Is it just for fun or is there something meaningful here? Do you think this turns into a real business? What's the plan? Yeah, the general thesis is that we're.

[Guest]

Able to subsidize the generation of code with affiliates and provide these state of the art models for much, much cheaper. Mostly for free actually, to most developers.

[John]

And so that's why you're putting. You're putting. So you're acting as a funnel to, you know, any affiliate that. So it could just be ads, but you picked specifically the most controversial ones, the gambling and the subway surfers. Like the stuff that feels more Brain Roddy. Because that would get into a reaction. Was that the plan?

[Jordi]

Yeah.

Yeah.

[Guest]

I mean, there's a. I mean, I think Jordy touched on this earlier. There is a difference between the marketing and the product.

[John]

Sure.

[Guest]

We actually started out with affiliates on these very normal sites and then a lot of our users actually requested saying, hey, we actually scroll on Rain bet. We actually go to stake during our generation time. Like, okay, we'll integrate that feature and then we'll use that as our marketing campaign. Okay.

[John]

It's incredible. I mean the debate was, are you making something people want? Is this in keeping with the Y combinator thesis and the values of the organization?

[Jordi]

Yeah, I guess. So break down what's actually happening.

You have the IDE and then you have this other column which you can Basically fill with anything. You could fill with an ad, you could fill with videos or rain bed or whatever. What are some of the most common ways that developers are using the product today? And what do you think really scales and becomes the most popular?

[Guest]

Yeah, the greatest thing about AI Native is that it completely changes the ad unit. So we have these AI Native ads that are in context and it's really great for code generation Here, let me give you an example. So let's say I code a website. Code me a website right now. Cloud code has this multi stage planning, right? It says, well, what do you want to code? Like how do you want to use a backend? If I say, well, maybe I want to use Supabase, say yes, Supabase. That's a Supabase conversion right there. So the ad is actually in the context, in the application layer. So we have multiple ad placements. But I think the most exciting one is how does ads scale at AI Native?

[Jordi]

Yeah, what was the name of the company that we had on? There's another company that's doing this and actually integrating the ad so that you see an ad, you're like, yes, I want this functionality. You press a button and the AI actually implements the product for you. And then you just. And I can just see that converting at a really high level and companies being willing to pay quite a lot to get in front of people like at the right time.

[John]

I mean, yeah, it makes a ton of sense to me on that level. A little bit less on the stake gambling while you're waiting. That feels like that would actually reduce developer productivity. Do you have any plans to actually assess whether or not this is a good decision? Because most developers are not solo entrepreneurs, they're employed by someone. And if I'm running an organization, do I really want my most valuable resource, my most valuable human capital tuning out every other second while they're waiting for the generation to come?

[Jordi]

One of those engineers might say, well, because I'm betting on rain. Bet with my personal dollars, you're paying less for the ide. I'm saving the company money. John.

[John]

Don'T you think that it would be better to show educational videos then something like that? Oh, we have that as well.

[Guest]

Yeah. So we have educational videos, learning about the code that you're actually writing.

[John]

Okay.

[Guest]

But I think our thesis is basically that we follow the YC advice, talk to your user. And the user wanted the gambling integration, so we made it for them. And at some point the user doesn't want it anymore, we'll take it away Right. So it's all about, I think for B2C is being close to the user, iterating close to the user and serving what they want.

[Jordi]

Have you been banned at any companies yet?

[Guest]

Actually the opposite. We had quite a few companies reach.

[John]

Out to us and say, hey, we.

[Guest]

Actually really want you to integrate our notion, our JIRA board, like the whole productivity workflow into the generation time. And we're like, we really, we're serving consumer right now. But I mean there's infinite possibilities here to scale at various business levels.

[Jordi]

How's the traction been to date?

[Guest]

It's been great. Yeah. I think I have to thank you.

[John]

Guys for that as well. Helping us go viral.

[Guest]

So we have a great waitlist of 11k.

[John]

11,000 people on the waitlist, successfully baited. Has anyone used it yet? Have you built it? Is it in the wild? Is it a VS code fork?

[Jordi]

Yeah. What were the metrics that you shared at demo day?

[Guest]

Yeah, so the messages I shared at demo day were 11k on the waitlist, 30k in revenue from ads we have people using right now in beta and we're going to give out codes today at demo day. So anybody who comes up to us in demo day, we're giving you a code. It's going really great.

[Jordi]

Amazing find Tyler. I know. He's probably in the same room. Let's get Tyler.

On Clad Labs or Chad Ide. We should hit the gong.

[John]

Yeah, we should.

[Jordi]

And.

How'S the round going?

[Guest]

It's going great. Yeah, we filled half the round. We have a lot of allocation to give out to people who are in interested.

[Jordi]

Awesome. All right, well, great to meet you, Richard. Thanks for coming on and breaking it down.

[John]

Appreciate it. We'll talk to you soon. Have a good one. Let me tell you about figma. Think bigger, build faster. Figma helps design and development teams build great products together. You can get started for free. We have our next guest coming into.

The ultra dome.

[Jordi]

This next company is absurd.

[John]

Really? Oh, wait, it is absurd. That's the name of the company they will be joining in just a minute here. We might need to pop back to the timeline while we wait for them to sit down. Jordy, you can take a view here. This is a live view into the very into ycw. So if we jump ahead of the schedule, we can always check in there.

[Jordi]

But there we go.

[John]

We have the founder of Absurd in the Restream waiting room. Let's bring him into the TVP in ultradome. How are you doing?

[Jordi]

What's happening?

[John]

Thank you so much for Taking the time to talk to us.

[Guest]

Of course. I'm doing good.

[Jordi]

How are you guys doing?

[Guest]

I know you guys are only taking on a couple companies today, so thanks for having me on.

[John]

We appreciate you.

[Jordi]

Fantastic.

[John]

Coming up, please introduce yourself. Tell us what you're building.

[Guest]

Yeah, my name is Philip. I'm CEO of Absurd. Absurd makes AI marketing videos. An ad that we've made you've probably.

[Jordi]

Seen on your feed is Kyle.

[Guest]

She's Mamdani versus Cuomo 1v1 basketball match.

[Jordi]

Which for the election.

[Guest]

We like to joke that we influence New York City elections.

[John]

Amazing. So walk me through the product. It sounds like you're more using the foundation models using Sora VO3 than training your own. But what are you building? How do you fit into the stack? Are you more of like a creative agency that I hire and pay a lot of money for an ad and you go out and use all the tools, or are you trying to build software as a service or train a foundation model? Where do you sit in the stack?

[Guest]

The way we're seeing how we fit into the stack is that we handle everything for a company in terms of AI native distribution. And the reason why we're doing it in that route instead of making an editor that anyone could use is because we can charge exponentially higher for that.

[John]

So do you want to ultimately productize this?

[Jordi]

This is what Harj was talking about. Basically, instead of building like an AI native accounting firm or an AI native law firm, you're effect effectively building an AI native creative ad agency where somebody comes and say, I want one launch video, please. And you say, sure, here's the fixed price. And then you guys use your internal tooling and whatever models you have access to to generate the best possible output and you deliver that end product.

[Guest]

Exactly.

[Jordi]

And what are you charging on like a per video basis today?

[Guest]

So a lot of that's confidential, but I can say we charge upwards of 30 grand per video.

[Jordi]

So in the same, you're effectively charging the same. Somewhat similar to what somebody would pay for a full day shoot.

[John]

Totally. You're in the proper video production realm, at least in terms of price. What are the secrets to using the video models appropriately to actually go viral?

What do you hire for? What are you focused on making sure that the video that you deliver is actually hitting upwards of $30,000 of value.

[Guest]

So in terms of the value we deliver, every video we've posted has gone viral. I mean, we average 300,000 organic views for every company we work with, regardless of Whether you have 200 followers on Twitter or you have like a million.

Second thing in terms of what we're prioritizing, what we're really thinking about internally is just how many videos per person per week. What's that throughput looking like? And then how do you drastically increase that week over week? So three weeks ago that was one video per person per week. Today it's 10 Super bowl quality ads per person per week made in parallel. Next week it's going to be 50, following week is going to be a thousand. I mean, there was a company that came to us, I can't say their name, but they said they won 1500 of our Kalshi super bowl ads in a month. And that's the type of quantity that we're talking about here. This is a lot of money that we turned down $200,000 in the past three days because we just, in terms of our bottleneck, we just had this huge technical bottleneck and we couldn't get it out in time.

[Jordi]

You turned that revenue down a few days ago. Why don't you just go back and say, hey, we have the capability, we have the capacity now. You just said it's rampant because we.

[Guest]

Still don't have capacity now. We could literally.

[Jordi]

So how many $30,000 videos have you sold? Did you find an infinite money glitch here or something? There's not even a thousand. There's not even a thousand venture backed startup launches a week.

[Guest]

So the way we were seeing things right now, sure, we start out with launch videos that we charge 30, 40 grand for, but now we're going towards more of like a retainer. Right. So now we're striking deals with companies like Kalshi Replit and wap and we're telling them, you know, we'll do a bundle deal, 10 videos a month for X price.

[John]

Sure.

[Guest]

Right. And eventually it's going to go to 50, then 100 and 200. A lot of this is going to be used in ads because the more you spend on ads, the more you have to switch out ad copy because of ad fatigue. And then we're going to go up and we're going to actually connect the orchestration layer to the actual metrics dashboard of all these ads.

[John]

Ads.

[Guest]

And then eventually we're getting at this point where we have this huge compounding data mode and our ad just get better and better. And you can think of an ad, I think for the first time in history as like you can create a thousand different variations with one click of a button. Because if you think about the ad in an AI ad, it's literally just like images and you're animating them. And as long as you have an agent that edits the images and changes the prompt slightly, you can create a thousand different variations and then test multiple things at once.

[John]

Will we see any absurd commercials during the super bowl this year?

[Guest]

I can't say. I.

[Jordi]

I think maybe that's a good answer.

[John]

You should make it up.

[Jordi]

He can't. He can, he can't. You know, people are going to look up his customers.

[John]

And what do you think about the role of taste of craft? A lot of of what's previously gone viral.

In the pre AI age has been someone coming up with a really unique concept, a really unique spin, and AI hasn't really been able to deliver those unique ideas. It's really good at reconstituting what's already out there and coming up with existing ideas.

[Jordi]

Yeah, Historically the best creative agencies have been the agencies with the best ideas. Right.

You pay to work with somebody that has a track record of generating great campaign concepts and then they'll oftentimes just outsource the work to people that are good at the execution layer, but not at the idea side. Do you feel like you guys need to develop an internal taste? Yeah, just ability to generate a high volume of good ideas now that the actual execution in terms of creative production is so much faster with AI.

[Guest]

Yeah, I think we think of creativity not as a monolith, but really in terms of two parts. Exactly how you put it. So there's a taste layer, then there's an execution layer. Our job here is we want to remove that bottleneck between an idea and a finished product. So internally what we're doing to solve that is like, sure, we're not going to replace human creativity, we're going to automate human labor. We're going to make it so easy for a comedian or a script writer or someone who just wants a part time job. And we're going to pay them a really high salary, really easy for them to create the seed of an idea that we can spin off tens and thousands of ads for.

[Jordi]

How much are you guys actually spending.

On the model side or within any of the applications that you're using to generate this, this content?

[Guest]

Well, for like a 30 second to 60 second video, really it's like 300, 400 bucks. We have an internal orchestration layer that picks the best models to use for all these specific use cases. It's paired with like a 50 page doc that has all our learnings that isn't available anywhere online. And we're able to use these models really effectively. So our margins like beyond just human labor because we're the ones making the videos and spending all these things up, we don't know how to price. That is like, like close to like 98%. But if you add in human labor, I think it's still like above 90.

[John]

How many different models are you using on an average 30 second video? Do you feel like it's worthwhile to stick to one model because you get more of a consistent look or are you jumping around? How do you think about the different models, what they're good at?

[Guest]

Well, it's extremely obvious that some models are just really good at some stuff and really bad at another thing. C Dream is good at specific use cases. Nanobanana is good specific use cases. Cling Juan all have their own unique use cases that we use. Something that's interesting beyond just the models is just like in terms of workflow orchestration before Nano Banana Pro came out. I'll give this an example. If you wanted to swap someone else's face.

I put in John's face and I say I want to put Kanye on that. That and that wouldn't work. So the way you do it is you'd actually tell Nanobanana to cut off John's head and then get that headless image and put Kanye's head on top. That's how we swap faces before nanobanana Pro. So there are all these little workflow things that we've learned just by experimenting and playing around with these models which play a huge role in making all our ads like the creation of our ads really affect.

[John]

Fascinating.

[Jordi]

Have we entered a post slop era?

Will we enter a post slop era? What is your post slop timelines?

[Guest]

I was speaking to Jess Lee actually about this. She was talking about how photography used to be seen as slop.

And because they used to say that photography was this way to treat artists who were actually painting something. But photography allowed people to realize that. That allow people to capture like a smile really quickly through slow motion. And something will emerge from this AI era where you can do something with AI video to capture some essence of human that you wouldn't be able to do otherwise. And we don't know what that is. But I'm pretty sure we'll be the first to figure that out. Especially if we're pushing all these videos every month.

[Jordi]

How big is the team today and how's the fundraise going?

[Guest]

It's just me and my Two co founders, Daniel and Damon. In terms of the round we closed, I can't announce how much, but we closed a week and a half ago.

[Jordi]

Incredible. John, hit that gong.

[John]

I will.

[Jordi]

For Philip and the Absurd team. Thanks for coming on Breaking it Down. I'm actually surprised there's not more companies trying to do this. Exactly. Exact sort of playbook. But it's cool to hear how you're thinking about this and excited to see more of the work that you guys put out.

[Guest]

Of course.

[Jordi]

Yeah.

[Guest]

And by the way, before I go, I'd love to make a launch video for you guys.

[John]

Let's talk about that.

[Guest]

I would love that.

[John]

I want to see what you can do. We have a benchmark here, bezel bench, where it involves a lot of watches on arms. We like to put this to the test with a lot of different, different AI video generators. It's a particularly hard shot to get right, but we can come up with a bunch of different ideas. Let's do it. That'd be fantastic.

[Jordi]

Let's do it. Perfect.

[John]

Well, have a great.

[Guest]

Dylan is an angel investor. He'll be in contact. Thank you guys for having me.

[John]

Talk to you soon.

[Guest]

Cheers.

[John]

Goodbye. Before we bring in our next guest, let me tell you about Adeo, the AI native CRM. Addio builds scales and grows your company to the next next level.

[Jordi]

Up next we have Lightberry with ar. I like social brains for robots.

[John]

Social brains for robots. Let's bring.

[Jordi]

I like the sound of that.

[John]

Ali. Lightberry. Yes. Very interesting to see what robots we're talking about, but we have him here in the studio.

[Jordi]

Welcome to the show. Hello, how are you? What's happening? Light berry owning yellow. Verticalizing yellow. I love it. I love it.

[Guest]

We have to wear something different. Everyone's wearing gray and blue and black and we need to stand.

[Jordi]

So yellow it is underrated color. Underrated color.

[Guest]

It is. It's awesome.

[Jordi]

Great to have you on the show. Why don't you introduce yourself, give some quick background what you're doing before starting Lightberry.

[Guest]

Yeah, of course. So yeah. Hi, everyone, I'm Ali. I'm one of the founders of Lightberry. We're effectively just building the operating system for all robots so that any person can use a robot. Before this, I ran a browser company called Sigma os. I was running product and design there and I went through YC in summer 21.

[Jordi]

Very cool, very cool. Talk more about that. This feels like a very big opportunity, but I'm not using a lot of robots in my day to day life today.

I assume that I will be much more in the future. But yeah, talk about what the business and the product looks like today and where you see the kind of category going.

[Guest]

Yeah, so we literally have a humanoid robot upstairs right now emceeing the entire event for demo day. And he's fully autonomous. He talks, we give him some instructions about how he should behave for the day. And he's just acting like a part of the event staff. Now you can go out there right now and just buy a humanoid robot from at least 50 different manufacturers.

[Jordi]

But if you do that, who did you buy yours from?

[Guest]

So ours is from Unitree. It's a Unitree robot.

[Jordi]

Did you buy it on Walmart.com? no, no, I know they sell Unitree.

[Guest]

No, no, no, no, not at all. No, no, we actually work directly with Unitree. And so like, you know, if you buy a robot from them or any of the 50 others, it literally doesn't do anything. It can't talk, you can't teach it anything. You can't. The only way to interact with it is by writing code. We thought that's insane. And so we're building a software layer that, that allows literally anyone to use a robot by just talking to it.

[Jordi]

What does adoption kind of look like with this? How are you actually selling it? Is this something that you want Unitree to kind of encourage their customers to adopt? Because again, I'm sure any manufacturer of robots doesn't want to just sell to developer hacker types that happen to want to go through all the different hoops in order to actually get value out of a humanoid.

[Guest]

I mean, that's exactly it. You hit the nail on the head. We're working directly with the manufacturers. There's over 50 of them. We actually just last week closed a deal with Unitree. They correspond to 90% of market share in the world.

[Jordi]

I'm giving you the air horn, but I have encouraged various government officials to ban Unitree from the United States.

[Guest]

Oh no.

Well look, the truth is they're the only. Like we want to work with the American companies too. We want to work with literally everyone but Unitree's shipping, they have market share, so it just makes sense to ship on them. We're going to be selling light berry powered robots with them all over the US but we're also working with other companies. Some European ones, some American ones.

[Jordi]

Yeah. What's happening? So I would imagine 1x has no interest in partnering with external providers. That would be my sense. Maybe that changes in the future. But I know they're trying to really Verticalize. And I'm sure they want to create great personality and some of the same feature set. But what about other players in the US Figure optimus, et cetera?

[Guest]

I mean the truth is they're just not shipping yet and when they want to start shipping and right now they currently don't have any software that allows you to interact with the robot. There's nothing that works in a public space. I heard that figures deal with OpenAI just fell through. I don't know if that's true, but that's the rumor. We'd love to help all of those companies get to market faster. It's just a race right now. So it's like whoever needs software so that you can interact with the robot, we're here to help.

[John]

What do you think the most dominant form factor for robotics in daily life will be in just maybe like two or three years? Do you think we're going to go through a wheeled robot phase or one robotic arm on Roomba phase? How do you see these? Because the self driving cars are sort of here, the roombas are sort of here. The full humanoid robot that feels a little bit farther out. But is there going to be more of a transitionary phase in your mind?

[Guest]

I mean if you look at sci fi as an indicator of what people want, we don't just want humanoids. There'll be different kinds of robots. You're going to have some small bipedal droids that we work with, a few companies that do that. You're going to have wheeled robots for delivery. That's just more practical in homes. I actually don't think you'll have humanoids because like why do you need locomotion in those cases? Humanoids are going to be the first general purpose form factor that's going to make it, in my opinion, just because.

[Paul Graham]

They look like us.

[Guest]

And the reason why we're building humanoids is because they look like people. And so we'll just be deploying them in people facing roles. So like shop assistants, manning booths at events, emceeing at demo day. Right. We have done this before. We deployed a fully functional autonomous humanoid at the 11 lab seven like three weeks ago. And it was just working there for ten and a half hours fully autonomously alongside the staff. So yeah, that's what we do. And we think that there's going to be tons of different form factors going to be like a Cambrian explosion of robots.

[John]

What are the compute constraints like, do you think on device inference is going to be really important?

[Guest]

So we run a hybrid Pipeline. We rely heavily on the cloud because that's where the best models are. And people prioritize the quality of interaction more than. Than the reliability of it. Now, we also run it, as I said, hybrid, so we have an offline version that's also running in the same time. So if connection drops or anything, the robot will still talk to you. It'll still understand. It'll be less smart, but it'll know about it.

[John]

Yeah. Have you had any luck?

I mean, how do you think about personality development? And I've been very fascinated by the fact that pretty much no lab has been able to hammer out of the model. Like, it's not this. It's that, like, they all have this specific LLM flavor to them that I don't think most humans. Maybe I run into one out of a million people that talk like that, but they all kind of. All the robots talk like robots. And I'm wondering if you have any thoughts about where that all goes.

[Guest]

I think prompting is just. I mean, these models are getting more and more steerable, and they're better at following instructions. So as long as you do a great job of spending time on designing those interactions, you'll be able to get these robots to behave less like robots. Now, we're not trying to make robots that behave just like people love C3PO, but C3PO is very obviously a robot. It has a robotic voice. It's a bit awkward in the way it speaks, and that's the inspiration. It should just be smart enough, but it should still behave and follow our social norms. The robot should look at you when you're speaking to it. The robot should be wearing the outfit of the staff members that it's representing if it's at an event. And that's what we're here to do. We're just here to make that easier for all of those manufacturers because they're racing on hardware. They don't have time to think about the software and the interactions.

[Jordi]

Are you excited about robot pets as a category? I know dogs are substantially cheaper, and that feels like something that a robot pet doesn't need to necessarily add any value outside of companionship. And so it feels like potentially a. An area where we could see a lot of growth in the near term.

[Guest]

So we actually have, like, a little pet droid in our office. It's like a bipedal that kind of looks like R2D2. We brought a bunch of little robots to the event, too. So there's like six of them in the demo. For anyone who's here. Yeah, I think robot pets are going to be really big. It's just we started working mostly with humanoids just because the price point is so much higher that we could just focus on quality rather than trying to optimize for cost. Obviously as these robots get smaller, the cost gets lower and so for us, we just care about quality. The models are going to get cheaper too. So we'll be able to deliver on toys, pets in the near future.

[John]

Yeah, the toy market seems really, really interesting.

[Guest]

Yeah, our first customer was a toy company actually.

[John]

It's just so much larger.

[Jordi]

What about security? I feel like there's a potential use case for humanoids just having a human shaped thing just moving around.

[Guest]

So literally the landlord of our building, when he saw that we moved in, he stepped into our office and on day one he just asked us like, oh, so these things can talk and they can walk around, they can map the world? He's like, yeah. And he was like, you know what, I would love to deploy them for security. How much does it cost? And I told him it's going to be like around 60 to 70K. He's like, I want four. I was like, okay, deal. So like he already pre ordered them. Like, people want this for security, not because it can fight, not because it can harm people. These things can't.

[Jordi]

But they're like the best deterrent. Yeah, it's just, it's the best deterrent.

[Guest]

And like, you know, we can literally talk to weird people in the evening and say like, who are you? And like run facial recognition, like, are you meant to be here? And then just alert like whoever's on, like on guard at staff and just call them and ask for help. Like that's how it should work, right?

[John]

Yeah.

[Guest]

Robots to help people, not to replace them.

[Jordi]

Yeah. I do think it's interesting that a lot of these humanoid companies are focused on the hardest possible thing, which is replacing like a house, a housekeeper who is already not the highest comps person doing the most like intricate, specialized tasks where somebody that's a security guard, their primary job is to just stand there and look like they're paying attention. And that's like the job. And they make like the same amount as a housekeeper.

[John]

Yeah.

[Guest]

I'm sorry, we don't think the chatgpt moment for robotics is going to be the day that your robot will know how to fold your laundry. We think it's going to be the day you start seeing robots everywhere in the street or like in shops, in coffee shops, in events, like talking to People, and that's just really soon.

[John]

It's going to be fun.

[Jordi]

Very cool. How big's the team?

[Guest]

We're just a small team of three people. We have a few people that we're working with that are helping out on top of it, obviously. But yeah, it's just a core team of three founders.

[Jordi]

Amazing. And how's the round going?

[Guest]

It's been very fun. I mean, we managed to close it like pretty early. There's a lot of. There's some interest now because, you know, with the unitree deal, we're pretty close to a series, a milestone. So we're trying to like, discuss that.

[Jordi]

There we go. Series A time.

[Guest]

Love it. Let's go.

[Jordi]

Really great to meet you.

[Guest]

We'll talk to you soon. Thank you for coming.

[Jordi]

Coming on and on.

[Guest]

Have a good one.

[John]

Cheers. Bye.

Up next, we have Dome, a unified API for prediction markets. This should be fun. It's trying to sit on top of.

[Jordi]

Pick a favorite, pick a favorite and Poly or polymarket.

[John]

Well, there is a lot of arbitrage to be done.

On the topic of robots. I'm super excited about the lamps that are happening. Have you seen that there's two robotic lamp companies now? They're like, they're.

[Jordi]

One of them was just cgi, right?

[John]

I don't know, maybe both of them were cgi.

[Jordi]

Isn't Apple making their own robotic.

[John]

It just feels like something that can be done. Whereas if it's full humanoid tomorrow, for this much money, that feels like a taller order. It's going to be a couple years away. But the lamp, I feel like we can do today. The lamp can talk to you. It's going to be funny. It's going to be awesome. I'm excited. I'm really bullish on the lamps, but I'm also bullish on a unified API for pro prediction markets. So we'll bring in the founder of Dome. Welcome to the show.

[Jordi]

What's going on?

[John]

Welcome to the TVP and Ultra Dome. You're in the Ultra Dome and company's Dome. Please introduce yourself and your company. What do you do?

[Guest]

Hi, my name is Karoosh. We're basically Dome. So Dome is a unified API for prediction markets? In a nutshell, what that means is we allow users and developers to trade and analyze across multiple platforms at once.

[John]

Okay, who's the customer? Are you talking hedge funds or like the most advanced traders?

[Guest]

Yeah, honestly, it's all of the above. A lot of our current customers are folks building applications in prediction markets. So these are folks building prediction market skins. Or markets themselves or copy trading and agentic trading is really popular right now. We talk to a lot of sports books and hedge funds as well. They're getting interested in high frequency trading and also platforms like think sweepstakes apps, folks who are trying to price internal parlays. So there's a lot of applications currently being built right now.

[John]

Now it's crazy.

[Jordi]

Very cool. Who's your favorite? Polymarket or Kalshi? I'm just kidding, I won't make you answer that.

[Guest]

I was about to say that's the million dollar question.

[Jordi]

Yeah, yeah, no, no, I mean it's, it's unfortunate, you know, it's unfortunate that the timeline is just so incredibly toxic right now. But I feel like you're able to kind of like sit back and be hopefully like Switzerland and support a variety of different exchanges. How do you think this market actually shapes out? Right. I think the big news from last week is that Robinhood is getting into the game themselves. They actually want to not just be a broker, they want to be the exchange. But how does this evolve?

[Guest]

Yeah, I mean we're supporting currently Polymarket and Kalshi, they're both great. Obviously we don't pick a winner in the fight. We want everyone to do well. And what we're currently seeing is there are a bunch of new platforms launching different regions, different specific verticals. Some folks are just like only sports, some are doing crypto mention markets. So what we're actually seeing is there's going to be a lot of players coming in, each trying to find their specific wedge, find their little market, their community there. And so in addition you have Kalsheet Polymarket, you'll have Robinhood and a bunch of other big players that are probably launching soon. But you'll also have a lot of these smaller players in different specific regions and verticals. And so, so we're excited to see the whole world basically start adopting this.

[John]

Do you have a reference point for how cross market transactions. Is there a public markets equivalent to you or some sort of.

Layer that's not necessarily a hedge fund? But I remember reading Flash Boys and in there they're talking about trading on the commodity markets in Chicago and then also the stock exchanges in New York and but it's done, this is all done by the hedge funds. There's not some sort of intermediary. Why do we need an intermediary here in this markets particularly?

[Guest]

Yeah, I mean it's a great question for what it says. Flash Boys is my co founder's favorite book.

But yeah, Absolutely. So one, as you get a lot more providers in right now, a lot of the liquidity is fragmented.

[John]

Okay.

[Guest]

So if you actually look at just calcium polymarket themselves about like 80% of their markets, their underlying contracts are the same with event. So you actually have a good amount of overlap there. But you also hit it on the nail as well is like there are other markets you can match against. Like sports books are obviously very, very clear. There's a lot of prediction market overlap there, Crypto prices, perps and all of these things. So by kind of taking all this data in, creating, creating that centralized source, it really helps out the hedge funds and those other professional traders who are trying to trade across multiple platforms because everything's in one spot.

[Jordi]

Or is some of your volume people just arbing markets on the different. Basically seeing like, okay, what are the odds on Kalshi, what are the odds on polymarket and trying to find alpha through that?

[Guest]

Yeah, I mean arbitrage is a very common request from a lot of our customers. Right. We actually had a customer that charted using our APIs, like the different prices across the platforms. It's a really cool visual because you can see the gaps over time of free arbitrage. And so arbitrage is a very common platform. One thing that we do really well is we make sure like when we are matching markets across platforms, we tell them like, hey, this is for sure one to one market versus like a maybe one to one market. Because personally the way we got started was we were trading ourselves and got burned as well when, when two markets look similar but they're not perfectly similar and you lose a lot of money. And so that's because you think your.

[John]

Head, you think you're just squeezing 1%.

[Jordi]

Here's the, here's the issue is you can have the same event and like different criteria market based on the platform and where what exchange it's hosted on. A lot of people have been seeing the rounds coming together for the different prediction market platforms and having flashbacks to OpenSea in 2021 and 2022. Why do you think NFTs, which also saw explosive growth in volume, are kind of not the right comp for this industry?

[Guest]

Yeah, great question. Biggest answer is we've kind of seen this exact playbook before. Both my co founder and I, we were founding engineers at a company called Alchemy. So they're the blockchain infrastructure layer for anything you're doing in Web three. They did extraordinarily well and prior to them, really the only really big businesses in Crypto was exchanges. After they came and solved the infrastructure problem, you saw a bunch of companies build on top of them, including OpenSea and Polymarket. So we've seen this wave. We've built a lot of the similar technology, the infrastructure layer at these previous companies. What you typically see is there's a huge hype and boom cycle. Everyone's excited, and then like interest kind of fades away, but people keep building. And then the next hype cycle, you realize, wow, the floor is raised. And so with, with prediction markets, you saw this during the 2024 election. Everyone was super excited. They thought this was the future. The election ended, everyone's like, oh, this is fine, we'll see you in 2028. But people kept building. And then the first week of the NFL Sunday, they did more volume than they did during the 2024 election. And so that's just more proof to say like, yes, there will be boom and bust as far as interest, but the overall market will continue to grow.

[Jordi]

Are you actually routing trades.

On behalf of clients or just providing the data layer? Because I imagine it could get quite difficult when some exchanges are using digital asset stablecoins, others are using traditional fiat. Rails, I'm sure you would need to integrate both. What can you say there?

[Guest]

Yeah, so first things we start off with is you got to solve the read layer. You got to give developers the tools they need to build.

[John]

Right.

[Guest]

So that was the first version of product is just give them data, give them prices, APIs, tools, whatever they need to display on their application so that they can build applications. Right. The next part of our plan was then, okay, let's actually start doing order routing and routing these requests to these different platforms. And we actually just recently launched our order router as of last week. And so we will be doing. We first are starting off on the crypto angle, like processing orders through on chain portions. And then eventually we'll also do off chain and traditional fiat as well.

[Jordi]

Well, do you think it's interesting that a lot of the sports books are funding lawsuits against the prediction markets while also starting prediction markets products themselves?

[Guest]

I think it's super interesting. I think a lot of these sports books and sports companies are also very smart and aware. They understand, they kind of see the writing on the wall. There's so many more advantages to having a pure prediction market, a P2P experience. It's a lot better for the end consumer as well. So I think they kind of see the writing on the wall. I think while the lawsuits are like the equivalent of like maybe the taxi industry suing Uber back in the day. I think eventually most of this industry will move towards prediction markets.

[Jordi]

How's the round going?

[Guest]

Round's been good. We actually closed up yesterday and so super, super excited. We're excited to get back to building.

[Jordi]

I had a feeling. I had a feeling. I appreciate that.

[Guest]

Yeah, I appreciate the excitement. It's been an exciting journey so far.

[John]

Well, thank you so much for coming on the show.

[Jordi]

Great to meet you.

[John]

Congratulations.

[Jordi]

Celebrating domes.

[John]

Yes, we appreciate you. We'll talk to you soon.

[Jordi]

Cheers.

[John]

Have a good one. Before we bring in our next guest, let me tell you about none other than Turbo Puffer, Serverless Vector and full deck search built from first principles and object storage, fast 10x cheaper, extremely scalable. The Forbes 30 under 30 came out today and liquidity is having some fun because one of the guys who made it, he performed 150% equity growth since 2019. But the S&P is up over 172% over the same period.

[Jordi]

He made money for his investors.

[John]

Well, this is the thing. He might have taken less risk and so if he took less risk and made almost the same amount of money, then that's good. There is a steel man for this particular person making the four.

[Jordi]

It's always a steel man.

[John]

But there's some good folks on the 30 under 30. We'll have to take you through them at some point. But until later we will go head over to Source and we're going to talk to David who's building Tinder for Jobs. David, good to meet you. Welcome to the show. Thanks so much for taking the time. Introduce yourself, introduce the company.

[Guest]

Yeah, thanks for having me guys. My name is David. I am one of the founders of Source. Source is like Tinder but for jobs. So you just upload your resume, swipe right, and AI will apply on the company's website for you.

[John]

Okay. How is AI actually helping there? Because I'm still doing the swiping myself. If I'm looking for a job, the AI is just doing the application. Is that correct?

[Guest]

Yeah, yeah. So you basically fill out one job application when you first set up the app and then when you swipe, then we have browser agents that will actually fill out the applications. Got it. It just saves the filling out form time.

[John]

How's the traction?

[Jordi]

What is what? Yeah, so talk about, can you talk about the state of the hiring market? Because I feel like the, the number one complaint that candidates and people that are applying for jobs have is that like seemingly nobody reads, nobody actually looks at job Applications. And a lot of roles don't actually end up getting hired based on traditional job boards. But yeah, what can you say about kind of what you're seeing in the market?

[Guest]

Yeah, I guess it very much depends on the company and the role in the sector. But in general, people definitely still get interviews from just inbound applications. A lot of it is automated and recruiters do kind of like sift through the applicants applications. But I think the number one meta point is that it's definitely a field that's like ripe for disruption. Like you are applying with many, many other people and there's typically other ways to get in. Like, a lot of people email themselves into a job or a lot of people refer their way into a job. But the inbound is definitely still something that companies use because when you're hiring people at scale, there's just no other way to do it. Like, if you're a company that's hiring like 200, 300 people a month, it's impossible to do it through inbound.

[Jordi]

Yeah. So what kind of jobs and markets have you been focused on? Because maybe it's not like other companies in a YC batch. Maybe that's, that's, maybe that's incorrect. But where, where's the focus been?

[Guest]

Yeah, yeah. I guess a misconception about Source is that we're not very directly working with these companies. We're just a traditional job board, like an indeed or a LinkedIn. So we directly scrape the ATSS. So right now there's like, like a million and a half jobs on the app and those are scraped from atss, like Workday or Greenhouse or Ashby. So if your company uses that system as an ats, then we've probably scraped your job and you're on source.

[John]

Are they okay with that? Is that fine? Does just scrape these? Because I know LinkedIn used to be amazing for scraping and then I'm assuming.

[Jordi]

Yes, because they're like, you're going to get more jobs.

[Guest]

The ATSS themselves aren't like advertising or marketing. Like, they're just SaaS.

[John]

Right.

[Guest]

So there's kind of a contract in this industry that ATSs are there to be scraped. Like indeed. 80% of the jobs on Indeed are scraped. Most of the jobs on LinkedIn are scraped.

[John]

Sure.

[Guest]

Job boards themselves obviously don't want to be scraped. Like we wouldn't want to get scraped script. But the ATS themselves, obviously, they are just like sending out emails to candidates and managing that whole pipeline, so.

[John]

Got it.

[Guest]

Yeah, that's completely fine. And as for how the Companies are reacting to it to answer someone's question.

Like, we've helped get over 25,000 interviews in the past year. And those range from.

Those range from.

[John]

Thank you, that's fantastic.

[Guest]

From like, I guess there's a very wide range of companies. Like you've helped somebody get a software engineering role at Enduril like a couple months ago, but then very often you'll see someone get like a, like a line cook job. But it's really just the universal fact is that filling out the form is very, very pointless.

[John]

How do you do top of funnel? Like how do you get people to be aware of your app, actually install it, download it? How are you driving attention on that side?

[Guest]

Yeah, we've gotten very good at going viral and getting views. I think over the past year we've done over 100 million views on social media, mostly on TikTok and Instagram. And that again is mostly just like me and my co founder making videos on TikTok and Instagram. We have like, I think like 72K on Instagram right now. And that's just from us pulling out the camera and telling people about what we're doing and people like it.

[Jordi]

So that's very cool. How, how are you going to make money? Are you making money already?

[Guest]

So we actually launched this while we were in school. Like I just graduated in May, but we launched this last like at the beginning of the fall semester. And we used to make money from charging people for, or by charging people for more swipes. We recently have gone like very, very free. Like you really don't need to pay to apply to a lot of jobs anymore. But yeah, we used to make money from that. Since we've taken that down, we don't really make money from that anymore. And in the future, obviously we plan to, to take the traditional job board route and work directly with employers. Just faster matches, get more applicants, etc, but right now we're very much just like product focused and we're kind of willfully ignoring revenue.

[John]

Yeah, yeah.

[Jordi]

How, how's the, how's the round going?

[Guest]

Round is basically done. I think my co founder is talking to investors, but it's really just for fun. Like we're, we're not planning on raising any more money.

[John]

Tell them to get back in the grind. You don't need to be talking to investors if you close the round.

[Jordi]

I don't like tinder 4x.

[John]

Oh sure.

[Jordi]

I'm sure that that actually resonates really well with consumers, but the product experience makes a ton of sense.

[John]

People Think swiping. They know swiping.

[Jordi]

Yeah, they know swiping.

[John]

They're not getting the way for that. Makes sense.

[Jordi]

But anyways, very, very, very cool. Congrats on all the traction and.

Hopefully we find some people on source at that point.

[John]

Yeah, that'd be great.

[Guest]

Absolutely.

[John]

Thanks so much. Have a good rest of your day. Let me tell you about Gemini 3 Pro, Google's most intelligent model yet. State of the art reasoning, next level vibe coding and deep multimodal understanding.

[Jordi]

Before we go to the next.

To the next guest, Juro in the chat says, I don't know if anyone said it, but the Ryzen X3D is the only way to go for your racing sims.

[John]

Ooh, that's an AMD chip. That's an AMD chip. We might have to do amd.

[Jordi]

I had a few emails ago, give us some. We've been talking.

[John]

I think we're dealing with an expert here.

[Jordi]

An expert talking to our friend Paul who's a racing enthusiast. Getting some recommendations there, but putting together some rigs for the team.

[John]

Yeah. Well, up next we have Matorial with Kareem, the integration layer for AI agents. Welcome to the show. How are you doing, Kareem? Thanks for.

[Jordi]

Finally somebody that is integrating agents.

[Guest]

Great job.

[John]

Almost. Almost. Correct. Okay, break it down. What are you doing? So we Basically give your AI agents, so your LLMs access to these apps and data sources. So anything from your Gmail to your SAP to your sales. Okay. I was just, we were just talking to somebody. Oh, Jason Fried, right. He was saying that OpenAI just wound up building a basecamp integration out of nowhere one day they just kind of told him, hey, it's live now. You didn't have to do anything. Is that not happening fast enough? Like in what scenario would I need your service? If all of the. It feels like there's a massive war going on between the LLMs. They all want to do the integrations as fast as possible. How is this going to play out?

[Guest]

I mean, actually one of the OpenAI member of technical staff reached out to us for our product.

[John]

This makes sense. There's that.

[Guest]

But basically one way to think about.

[John]

This is first of all, OpenAI won't.

[Guest]

Give you AI integrations for the other providers.

[John]

People still want to be using Gemini, they want to be using Anthropic or any of these others. So we basically provide you with the developer tooling to use any LLM model.

[Guest]

With any AI integration. And it's not just integrations, it's also.

[John]

These things like access control right. Because these Fortune 500s can't just unleash an LLM with access to whatever your Salesforce SAP to all the members in their organization. They need to think very concretely about who has secure access to which models and which data sources. Yeah, that makes a ton of sense.

[Jordi]

What were you doing before this?

[Guest]

So I just graduated from NYU Abu.

[John]

Dhabi in May and before that I ran a different Abu Dhabi based ticketing.

[Guest]

Startup for around three and a half years.

[John]

Oh, that's cool.

[Jordi]

Very, very cool. What's traction been like you said member of the technical staff at OpenAI reached out.

[Guest]

That's a development from yesterday.

[John]

So not too much, not too many updates on that.

[Guest]

But we are open source with over.

[John]

3,600 GitHub stars and we have close to 1,000 weekly active users just since.

[Guest]

Launching around five weeks ago. And then we are also in final stage discussions with some Fortune 500s and.

[John]

Unicorns who would deploy this across the organization. Good side effects across the organizations of 80,000, 100,000 members. Is MCP complementary, competitive, substitutive? How does MCP fit into this? So here's how we think about it, right? So LLMs 10 years from now will still need access to apps and data.

[Guest]

Sources with access control.

[John]

Right now the standard for that is mcp. So we basically have this middleware layer translating between our platform and mcp. But if the standard changes a year from now, we just switch to that new standard. Right, because the long term bet here is not an mcp.

[Guest]

I think that's what a lot of.

[John]

These other companies are getting wrong, where they're building 100% on top of MCP, but they don't actually think about what these companies need. They're just kind of following the hype train of oh, MCP is the next cool big thing. Which we are not fully in agreeing. Can you take me through sort of like the top five agent categories that are interesting to you? I imagine coding agents are probably at the top. Maybe knowledge retrieval, deep research agents. Maybe. Yeah, we don't actually think about that. Okay. We are completely unopinionated about how you build your agent. We just provide you with the integrations.

[Guest]

Sure, right. Because every agent will need to do.

[John]

Read and write operations on these apps and data sources. And if you can take a tax on that, you figure out how big the market is.

[Guest]

Yeah.

[John]

Is there, I mean, I guess to flip the question around, just what agentic capabilities are you excited to see out in the world in 2026?

[Guest]

Honestly, I really like seeing all these.

[John]

New verticals where basically people just what do they call them? Those full stack AI native firms were free. People go in there, use these LM capabilities and these for example legal agents or healthcare agents to compete with unicorns or large established players. I think that's really exciting.

[Guest]

You kind of got this Goliath story there.

[John]

Okay, so walk me through that. If I'm a lawyer and I'm leaving my firm to start an AI native law firm, I might buy some AI legal SaaS, but I also might need to integrate with some more niche tools or some more legacy tools. Are you the firm that I would go to to do those integrations for me? Yeah.

[Guest]

We basically want to become the substrate for your integrations. So really long term we want to have a sort of Oracle story here.

[John]

How similar to how Oracle became the substrate for enterprise databases and then sold.

[Guest]

Those extra things like enterprise Java, et cetera on top. We want to be the substrate for the integration layer and the access control layer and then add these additional things like the workflow build or also hosting your agents. Right, yeah, that's kind of the long term vision here.

[John]

I always like to take the temperature on YC folks on like what's breaking out in their supply chain. What's a tool or company or service or technology that you are leveraging to build this company that you're particularly thankful for.

[Guest]

See this might surprise you, but kind of of compared to a lot of.

[John]

Other people, we are very OG software engineers. And what I mean by that is my co founder and I have had.

[Guest]

Formal computer science education for over 11 years.

[John]

Sure.

[Guest]

So we met in Austrian technical high.

[John]

School at 14 years old for basically computer science.

[Guest]

And that really allows us to think about first principles.

[John]

So in terms of building out our.

[Guest]

Entire infrastructure ourselves, thinking about the API design from scratch.

[John]

And we don't really use that many.

[Guest]

Tools that are available out of the market right now because what we find.

[John]

Is that they speed up the process a little bit. But we have been doing it for so long that we can just do it better ourselves. So really we invented a lot of new things here as well, which kind of the other competitors who are mostly only wipe coding can't even do with their.

[Jordi]

You need to get an organic certification on the website. This is organic code, zero value.

[John]

You can get the Austrian amagel.

[Jordi]

Let me guess, the round's already done.

[John]

Yes, very fast.

[Guest]

Actually wrapped up in around five days.

[Jordi]

Five days. I knew it. I knew it. Oh, I knew it. Congratulations, love.

[John]

Thank you.

[Jordi]

Yeah, loved hearing how you're thinking about the opportunity and how opinionated you are. So congrats on all the progress. Excited to follow on.

[Guest]

Appreciate it.

[Jordi]

I'm sure you'll be back on the show soon.

[John]

We'll talk to you soon.

[Guest]

Thank you so much.

[John]

Have a good rest of your day. Before bringing our next guest, let me tell you about Fin AI, the number one AI agent for customer service. Automate the most complex customer service queries on every channel with Fin AI. And we have Philip from Crunched. What a great name for an AI analyst for an AI. Excel Analyst for Excel power users.

[Jordi]

Just for power users.

[John]

Have you ever been an Excel power user? You always had to have one hand on the mouse. You were never just on the keyboards.

[Jordi]

Guy always had one. Very soft, very soft.

[John]

I can hear Andrew Reed losing respect from you all the way from here.

[Jordi]

He's getting cooked all the way from the Valley.

[John]

Indeed. Well, he is in the restream waiting room. Let's bring in Philip from Crunch to the TVP at Ultradome. Philip, welcome to the show.

[Jordi]

What's happening?

[John]

Thanks for joining. Please introduce yourself and the company.

[Guest]

Hey guys, pleasure to be on. Great to meet you. Michael actually from Crunch did a last minute switch here. Oh, okay.

[John]

Good to see you.

[Guest]

Michael, the co founder as well. Coo. COO of Crunch.

[John]

Fantastic.

[Guest]

Maybe I'll give you a two second description of Crunch then. Crunch is your Excel AI analyst built by and for power users. So it's like this side panel chat in Excel, basically cursor for the world's most popular programming language and then you're just chat it in natural language and it makes modeling for you.

[John]

Makes a ton of sense. Very clear value prop. I think everyone who uses Excel wants a copilot. But there is a company that's trying to build Copilot and they happen to own Excel. How are you imagining this plays out? Are you going to live in plugin world? Are you going to live at the OS level and be screen scraping? Are you worried about sharp elbows for Microsoft? How are you dealing with all that?

[Guest]

It's a great question. I think Microsoft is for sure going to build a great product. They're building a co pilot for 2 billion Excel users and they're in competition with Google Sheets. Right. I think we're building a tool specifically for the top 1%. Finance professionals, investment bankers, private equity associates, management consultants of the world who use Excel in a very specific way. Right. So this is more of the 5 million of the Excel users. The top 1%. So that's a bit of the. Bit of the difference.

[Jordi]

Lot of Big, big market, big opportunity. If you build a great product, there's tons of people that will happily pay for it. There's also tons of startups as well going after this opportunity.

What do you think they're getting wrong?

Or is this just going to come down to actual product quality and working super closely with these power users to to make something that actually integrates into their everyday Excel life?

[Guest]

Yeah, absolutely. So I think we have plenty of startups going after this opportunity. We don't think about competition too much. But out of the big ones with the most traction, we're the only one with a team that has 10,000 plus real life Excel hours in our previous jobs, me and my co founder Philippe in McKinsey and another finance guy gigs and I think that really shines true in the product. I think also crunched is modeling more like a real life analyst and performing more of the real tasks that you do on the analyst floor versus like some of the bit artificial benchmarks you see around. So for example crunchcan detect mistakes in workbooks. Plenty of time is spent in like private equity firms on actually reviewing Excel and making sure they are correct. As much time as modeling from scratch. Right on the. And these professionals typically work with templates, right. And they need crunch to fill out and augment their templates, not build like basic analysis from scratch. We can do that as well, but we're great at working with large models and these sorts of things.

[John]

How do you think about the enterprise flywheel here? It seems like one of Cursor's main advantages is that they have a really solid data flywheel. Now from open source developers and developers who are not in a enterprise level contract, I imagine that the top 1% of investment bankers, consultants on day one, they're going to not want you to train on their data because it's going to be not just some code that builds a front end website, but extremely critical financial information, private information. It is probably a higher bar to not letting that leak into a training run. So how do you get a data flywheel going? How do you improve the product iteratively?

[Guest]

It's a good question. Right? And as you say, security is top concern I think for all of our customers. We live with global consulting firms.

[Jordi]

But also let's give it up for global consulting firms. They don't get enough love. They don't get enough love. Except here. Except here. Except here.

[John]

They just don't.

[Jordi]

I will, I will defend Accenture.

[Guest]

Exactly.

That's good. But they are obviously super concerned about their security. Right. And do live public deals. Right. All of this stuff. And so like in principle we do not train on the data of our customers and we can cannot see what the prompter do.

[John]

Right.

[Guest]

At the same time, what we want to do now and just in record time, close our fundraise, we want to make sure that we tailor Crunch to every single firm.

As well. That's great.

And then.

[John]

When we tailor it, we.

[Guest]

Have discussed with a few customers the opportunity to for some of the large organizations, they do enough modeling work on a global basis that is possible to tailor, do some fine tuning and tailor to their specific organization. But as well, we come in in a forward deployed manner. Right. And do customization, whether that is formatting or solving for their specific workflows and linking into their templates. How like the sims that they get, how can we link that into their specific LBO template and then transfer that from like the simple LBO to the advanced LBO and these sorts of customers.

[Jordi]

What's the biggest deal Crunch has supported?

[Guest]

The biggest deal we have supported. That's a good question.

[John]

I can tell you about the.

[Jordi]

You don't need to name the company. Yeah, you don't need to name the company.

[John]

It was a $500 billion company. They were doing a $1.4 trillion deal. They were doing about 20 billion in revenue. I'm not going to say who it was.

[Guest]

Exactly, exactly. But I can tell you a real story about a mistake we caught though. Crunch has this error detection system and on a live deal for an associate that one of our, our private equity clients in London used the sort of crunch mistake detection system to, to identify or like scan his, one of his previous models on a real transaction and identify the mistake in the working capital that overvalued the deal by £10 million.

[Jordi]

So.

You saved his job.

[John]

Send him an invoice for 5 million right now. You just saved him 10. Give me 50% of that. That's your seed round right there. Exactly, exactly. Well, congratulations on a fantastic demo day. Thank you so much.

[Jordi]

Yeah, great. Great to meet you, Michael.

[John]

We'd love to talk to you again.

[Jordi]

Congrats.

[John]

I live for Excel agents. I'm so excited about this category. It just feels like we should get your access. I would love to. Thank you. Well, have a great rest of your day. We will talk to you soon.

[Jordi]

Great hanging, Michael.

[John]

Goodbye.

[Guest]

All right, thanks guys.

[Jordi]

Numeral what $500 billion company could that be?

[John]

Numeral.com compliance handled numeral worries about sales tax and VAT compliance so you can focus on growth. Speaking of Growth.

There's some folks putting Menlo Ventures in the truth zone. Enterprise large language model API market share has been falling for OpenAI. It's been climbing amongst anthropic according to Menlo Ventures. Ev Randall puts it in the truth zone over at benchmark multi time TVPN guest EV Randall. He says people are quoting this Menlo Ventures chart and extrapolating from it like it's official data from the Federal Reserve or something. It's a small sample survey conducted by an investor in Anthropic. Please calm down. I like that he's pouring some cold water on this.

[Jordi]

This was from November 3rd.

[John]

At the same time, is it possible that OpenAI's Enterprise Large Language model API market share is falling? Sure. They were the only game in town when they launched and so you would expect their market share to fall a little bit over time. Will be interesting to see. We'll get more data on this. All these companies are going to be public in a couple of years and so we'll know exactly how it's breaking down. Can't wait until, until that happens. We will return to our coverage of YC Demo Day 2025. We have Sava, the AI powered trust company. Welcome to the show. How are you doing?

[Jordi]

What's going on?

[Guest]

Hey, I'm doing great. How are you?

[John]

We're fantastic. Please introduce yourself, introduce the company. Tell us what you're building.

[Guest]

Great. Yeah. I'm Nimit Maru. We're building Sava. We're building a new modern agentic trust company that administers advanced trusts.

[John]

So is this specifically like will and Trust?

[Guest]

Yeah. So it's trust like will and Trust. Yeah, exactly.

[John]

Yeah. Not because people would say a trust company could be somebody that makes sure your password doesn't get leaked or something. But this is.

[Jordi]

How old were you when you realized you wanted to use AI to spin up trust? I'm just kidding. What were you doing? What were you doing before this? This?

[Guest]

Well, my, my previous company was actually In John's batch summer 12 batch.

[John]

No way.

[Jordi]

No way.

[John]

What company?

[Guest]

Yeah, and we, we. So at the time we were building Yeli, which was a. We're like, you know, like the front facing camera had come out on the iPhone. So we wanted to build like a telemedicine but we pivoted to being an early code education and like tech education company and that's how we kind of built that and then sold it in like 10 years later.

[Jordi]

If you hadn't pivoted, you could have been selling meth at scale like some of the other telemedicine companies. But I'm glad you did.

[John]

I'm glad you went the code of.

[Jordi]

Very cool.

[Guest]

Did you say selling, did you say selling meth?

[Jordi]

No, no, I'm just, I'm just, I'm just joking because I'm not sure if you're.

[John]

But there were some pill mills, there were some telemedicine companies that went a little bit too far and one of the founders is in jail now.

[Jordi]

They're like check if the patient breathing.

[John]

If they are give matter all that's that was going. Yeah.

[Jordi]

No more seriously talk about what's are these Nevada trusts? Like what's, what's, what's the, what's happening at the actual like entity layer?

[Guest]

Sure. Yeah. So we, so we're not drafting the trusts. We will basically like a, an attorney or a, like a fintech or legal tech that uses LLM to draft trusts. So they would create the trust documents and then once they need someone to administer the trust to be an independent trustee, that's when we would take over. We're getting our charter in Nevada, so we're going to be chartered in Nevada. You know, maybe eventually we'll go to other states, but that's where we're going to be right now. And yeah, we work directly with attorneys, wealth managers, fintechs to serve as the trust administration there.

[John]

So so would you. Do you have like no consumer facing brand essentially it's like purely B2B at that point.

[Guest]

Well, I mean it is consumer facing in the sense that the people who would be using it are also the families who have these trusts. But the reason I say we work with attorneys is because generally the families are taking advice from the attorney or the wealth manager about which trust company to choose. Because I mean how would a family know even what a trust company is or so. So we think of them as the icp.

[John]

Sure.

[Jordi]

Are trusts underrated?

[Guest]

Yeah, I think they're underrated and they're underutilized and also right now they're very annoying and expensive to create and manage. And so I think people don't use the power of trusts enough. And that's not to say like you know, every American or every person can be using them, but definitely a big slab of people in know kind of below where right now they're being utilized.

[John]

Yeah. Do you have a ballpark cost figure for you know, doing a trust? Like what, at what scale does it start to make sense for customers to even participate in the market to even consider a trust?

[Guest]

So I think creating a revocable Trust that, you know, owns your house or other assets that's applicable and at reasonably, almost any level, when someone would own some property, just as soon as you own a house.

[John]

Makes sense.

[Guest]

Yeah, but then using something like Sava today, it's generally people who are trying to make irrevocable trusts. And so they would tend to have.

Some millions, maybe low single digits or maybe mid single digits, millions in assets before they start utilizing that. That I think that as tech makes it a, a lot.

Easier and cheaper to create trust in a good way. And also, you know, people like us can make it a lot more friendly and modern to administer trust. Like, I think more people will be able to use them.

[John]

It should just get way cheaper. I mean, if you think about just the YC story of, of how much it cost to set up a corporation and raise a seed round in 2005 or something, you were looking at like 20,000, maybe 50,000 in total fees across everything. Now it's like Stripe Atlas. One click, they charge you, what, 200 bucks or something?

[Jordi]

500 bucks.

[John]

500 bucks. And then the safe is one second and administered by a bunch of folks. It's like really, really low, low cost. And that's obviously led to just more entrepreneurship. You would imagine that something similar happens.

[Guest]

When the infrastructure gets better. Like usage goes up and even the safe, like I was talking to my co founder the other day, like, the safe is an incredible invention that makes this, like, early stage of fundraising, you know, so much smoother. Like back when we did it in the summer 12 batch, it was like all convertible notes and, you know, even that a lot of investors wanted price round at this stage. It's like a pretty difficult thing. So, yeah, I think when the infrastructure gets better, like, more people utilize it and like, more people can take advantage.

[John]

Well, congratulations on the progress. Thanks so much for coming on the show.

[Jordi]

Yeah, excited to check the product out.

[John]

And we'll talk to you soon. Have a great rest of your day.

[Guest]

Thank you. Cheers.

[John]

Goodbye.

[Guest]

Thank you, guys.

[John]

Let me tell you about Profound. Get your brand mentioned in ChatGPT. Reach millions of consumers who use AI to discover new products and brands. I want to pull up this chart of the day from CO2. They say, hey, look, there's no code red here. It's all Baja Blast. Because chatgpt traffic historically dips this time of year. And it's a fascinating chart. If you actually zoom in on this Gemini 3 launch day, it looks like people stop using LLMs around Christmas. The turkey's going around the tryptophan is coursing through their blood.

[Jordi]

They're getting a little sleepy.

[John]

They're getting a little sleepy. They're having an extra bottle of wine and they're taking time off from their chat app, specifically from ChatGPT. This is bizarre that this chart tracks so much with when people do work. You can see that ChatGPT grows.

In the spring every year up until summer. Then it completely flatlines during summer. Then it peaks when school year starts again and work starts back up. Then it crashes on.

[Jordi]

Students and students.

[John]

And workers, people with jobs. That's everyone. That's everyone. Come on. That's everyone.

[Jordi]

What about the unemployed?

[John]

Oh, yes, I don't know. Well, they're the ones that are holding it up. They're holding it down during Black Friday. They're like, I'm still grinding. But clearly folks did not get the great lock in memo because the whole point was that you were supposed to continue to use all the AI apps anyway. It's a fascinating, it's a fascinating chart. I'm sure we'll be digging into it more, reading the tea leaves. But up next we have Ben from SF Tensor Sversel for GPUs. Welcome to the show. Thank you so much. Please introduce yourself and the company.

[Jordi]

Great to have you.

[Guest]

Hi. Yeah, thanks. I'm Ben. We're building the infrastructure layer for AI researchers. So basically from training models, from small experiments all the way up to large scale frontier training runs, we basically deal with infrastructure to allow you to do all your training runs.

[John]

Okay, so there's a bunch of different layers going down to somebody that owns the ground, somebody that builds the data center, somebody that racks the GPUs. And then there's the NEO clouds. Are you interfacing with multiple NEO clouds? Are you a NEO cloud? How are you positioning yourself?

[Guest]

Yeah, so we work with all sorts of NEO clouds and hyperscalers and we basically just say we're building above all of them. And so our customers should only be worrying about what they want to be researching or training and not like how the actual technology, like the underlying stuff works. And so we deal with, you know, finding GPU allocations, optimizing for different GPUs. So we also allow you to work with TPUs or AMD GPUs or any of this stuff to allow you to train your models.

[Jordi]

Okay, so this is specifically for research and training runs and less focus on like actually inferencing on the product side.

[Guest]

Yeah, so we focus exclusively on the, on the training side. There's great companies even, you know, from last batch, for example, there's Luminal. They do great things for inference. We focus just on training because we think training is a problem that's not been solved by anyone. And there needs to be way more training happening.

[John]

What are your clients.

Like? What's the shape of them? I guess there's a lot of focus. When people think training, they think, think OpenAI, Anthropic, Google, DeepMind. Right, but take me through the variety, the landscape of folks that you talk to who are actually doing training runs. Who are these folks? You don't have to give exact names, but tell me the shape of their workloads, what problems they're trying to solve, the scale of their training runs. Take me on a little tour.

[Guest]

Yeah, so there's a huge variety. I mean you have on the one hand you obviously have like the academic or researchers at home who are trained like small models, and then you have larger scale academic research happening. But then you also have startups that have raised maybe call it $10 million. There's some companies from YC as well who are training models for super niche use cases. And then there's also companies that have raised hundreds of millions or up to a billion dollars. There's a bunch of labs actually in that area who are training their own models. You don't just have anthropic, I mean like the text based models like LLMs. There's not an awful lot of competition going on there anymore. Things have sort of converged at the top there. But for everything else, like you know, drug discovery or you know, protein folding, all of these things are still problems that have not been solved by anyone.

[Jordi]

Is it correct to say that SF Tensor is a bet that there will be millions of.

Smaller models for specific use cases or one day billions?

[Guest]

I wouldn't say billions, but definitely a lot more than there is today, especially just in the modalities that haven't been explored today. I mean we're all focusing on text and text is great for a lot of things. But I can't really use a text based model to do things like, you know, text to speech for example is another type of model. Or we have protein folding models or all of these things can't really be solved with text. We need models that are specialized in those topics.

[John]

What about, I mean, we were talking to the CEO of AWS yesterday and he was saying that AWS launched a product that is actually a checkpoint 80% of the way done on an actual foundation model. And then a company can come in and add their own data to the pre train and then they can do everything else with it. And that felt like an interesting proposition. When you think about if you do want a text based model and you want it to be to really know your company's data at the core in the pre train, really know it, not just drop it in the prompt, not just fine tune on it, actually bake it in. That feels like we're going to see a Cambrian explosion of every company wants their own trained model earlier, they're going to want training workloads for that. Is that something that you think you can play in? Are there already other companies that are working there? How do you think about that?

[Guest]

So it's a very unexplored area so far. The idea of basically saying you have like, you know, 80% of the way the model can already form coherent sentences, have basic reasoning abilities and then I add my own information. I think that's going to be very important in the future just because it allows me to take a base model and then not just do like post training, but sort of, you know, continuous pre training almost, you know, continuing the pre training. I think there's going to be a lot of use cases that come out of that and I think we can help there. I mean, we don't really care what you're training on the hardware. You know, if, if it's an AI training.

We can help with that. So that's definitely something we're looking into.

[John]

Do you want to ask about progress?

[Jordi]

Yeah. What kind of metrics were you sharing today during demo day?

[Guest]

Yeah, so the metric we're sharing is we launched like two weeks ago and we did $41,000 in usage based revenue since then.

[Jordi]

There we go.

[Guest]

Love it.

[Jordi]

And how's the round going?

[Guest]

We closed the first day of fundraising.

[Jordi]

First day of fundraising.

[John]

There we go.

[Jordi]

There you go. I'm not gonna dox, but a friend.

[John]

Of ours, we got a text message about you.

[Jordi]

We got a text message about you. A friend of ours just backed one company, this batch and he's known for backing great companies and he just backed you. So I'm excited for you guys to announce the round soon and come back on and do it on tbp.

[John]

Thank you so much.

[Jordi]

Awesome.

[John]

Great to meet you soon.

[Jordi]

Cheers.

[John]

Have a good night to meet you. Good to meet you. Let me tell you about getbezel.com shop over 26,500 luxury watches.

[Jordi]

Super intelligence for your wrist.

[John]

Fully authenticated in house by Bezel's team of experts. Brad Gerstner on Trump accounts. POTUS was elected on Main, on a Main street agenda to get the rest of America into the game. And that's exactly what this does. Bill Gurley, showing him some respect.

We didn't cover it yesterday, but Michael Dell donated $6.5 billion to these Trump accounts. The accounts where children get them, they can't be touched, they're invested and they.

[Jordi]

Compound over $250 for a bunch of individuals.

[John]

And there was some pushback. Some people are saying, well, if you compound at the S and P, even if you compound at 10% for 20 years, like it's only a thousand bucks or a couple thousand bucks, it's not that much money. It's not life changing. But you know, it's like a piece of, it's one that's, that's just Dell's contribution. Like there's going to be other people that are contributing corporations.

[Jordi]

There's $1,000 from America and.

[John]

Yeah, yeah. And there's a whole bunch of other ways to add money to the account over time at birthdays and Christmas and.

[Jordi]

Stuff like targeted donations.

[John]

And the most important thing is that there's a lockbox. So it's psychologically a lockbox. So I still stand with the, with the Gerstner accounts.

[Jordi]

It's incredible.

[John]

But we have our next guest in the Restream waiting room from Locus Payment Infrastructure for Agents. How are you doing? Please introduce yourself and tell us what you're shirt.

[Jordi]

I like tvpn. We've done over a thousand interviews.

[John]

I don't think we've ever seen one. This is unique. I like it.

[Jordi]

It's a first. It's a first. Thank you.

[Guest]

Yeah. And, and they, you know. Thank you. And so they actually switched me up, up with the other guy. Oh, Henry.

[John]

Icarus. Sorry.

[Guest]

Great.

[Jordi]

Well, Henry Tydai, welcome to the show.

[Guest]

Yes, sir.

[John]

Henry from Icarus, please introduce yourself and tell us what you're building.

[Guest]

Yeah. So I'm Henry, founder and CEO at Icarus. My background, aerospace engineer at Georgia Tech. Built drones for NASA and satellites at Orbital Icarus. We're building solar powered autonomous drones that fly at 60,000ft for weeks at a time.

[Jordi]

Close to the sun. Close to the sun. Not the closest, but not the closest.

[Guest]

And in fact, if we flew any higher, we'd actually fall out of the sky. So we want to stay at 60,000ft.

[John]

You're like, you're like. But we're going to try flying a little higher.

[Jordi]

Okay. How many hard tech companies were in this batch?

[Guest]

I believe like five or ten.

[Jordi]

Yeah, that seems about right. So I Feel like it's been like steadily at 5 to 10 for.

Forever, basically.

[John]

Take me through the bear case for stratospheric drones. What I've heard is, you know, people always, always refer to the SR71. It's such an amazing plane. It flies, I think around 60,000ft. The SR71 Blackbird, it's this amazing Lockheed Martin plane built at Skunk Works. Flies super fast. We can't build planes like that anymore. We don't have it in us. And when I talk to folks who are like, yeah, it kind of sucks, we can't build that because it was really cool. But we have satellites now and satellites go way higher and way faster. And so if you need to put a camera over something, we usually just use a satellite. So why not satellites for this use case?

[Guest]

Yeah. From first principles, you're 20 times closer than low earth orbit and you can say fixed urban area. So just from an engineering perspective, it makes a lot of sense. The bare case is pretty much like none of this is new. Even what I'm doing, the solar powered version.

It'S all been done. It's just been too expensive.

[John]

Sure.

[Guest]

So the question is like, can you get the cost down?

[John]

Can you? How are you doing that? Is it just being a starter up? Are you using cheaper materials? Are you standing on the shoulders of giants? Like, what are you leveraging to actually make it cheaper?

[Jordi]

Well, we'll talk about the form factor first because I'm on the website and this thing just looks like a massive, really skinny bird. Yeah, it's very unique. It's Icarus1.com or sorry, Icarus.1.

[Guest]

Icarus.1. Correct. Yeah. To your point, John, it's about getting the right product specifications for the first go to market. And so yeah, our first product is it's a 20 foot solar powered bird. Flies for weeks at a time. Yep.

[John]

The bird noise is perfect.

[Jordi]

So it's effectively like a loitering drone that's just sitting at 60,000ft. And I'm assuming it's incredibly light.

[John]

Station.

[Jordi]

It has a battery, but it can generate solar power on the fly to increase the battery life effectively. Like it's not sufficient to hold it in the air forever yet, but yes, but it can stay up over a specific area. So is this primarily like defense applications early on? Who are you trying to sell this to?

[Guest]

Yeah, act one. It's all defense. I do think this is much bigger than a defense company. I do see the stratosphere as a category. And.

Once you kind of are able to.

Make the stratosphere affordable, then there's many things you can do. So one easy example. Today you can't really carry very heavy payloads. You can't carry and deliver a lot of power. But the future looks like, and there's no laws of physics that says you can't do this. You can essentially take a Starlink satellite and have that in the stratosphere and imagine if you had this StarLink satellite that's 20 times closer and fixed open area. So then that's like that's the future. And what you can do from that, it's, I don't know, it's anyone's imagination near term. There's a lot of clear direct line of sight towards defense and a market there. Again, it's like really difficult. It's not like a category yet. Today.

There'S no real markets. But with defense there's a clear need.

[Jordi]

Very cool. How do you actually get the drone up? Is this something that you launch like a rocket and then it sort of spreads its wings at some point? Like how do you actually get a 20 foot drone 60,000ft in the air?

[John]

Into the air?

[Guest]

Yeah. So we use a balloon.

[John]

Oh, you use a balloon. Okay, okay.

[Jordi]

That seems less violent than yeeting a 20 foot drone.

[John]

Some drones are yeeted. I believe this is a real thing.

[Jordi]

So you use effectively like a weather balloon to take it up.

[John]

Are you a beneficiary of Starlink?

[Guest]

Are we a competitor?

[John]

No, no, no, a beneficiary. Oh, beneficiary. Can you use Starlink effectively as the backbone for communications?

[Guest]

Yes, that is our beyond line of sight method.

[John]

Sure.

[Guest]

Yeah. So we have Starlink on it as an option.

[John]

Yeah, that's very cool. Yeah. Fast. So how close are you to actually getting this up in the air? Have you flown? Is it just test at this point? Are you actually going to sell these things?

[Guest]

We are selling them today to the army and yeah, we've done over 30 successful stratospheric flights, successful demos with Special Ops Command, SOCOM and the army as well. And we have. Oh, there you go.

[Jordi]

There you go.

[Guest]

There we go. Yeah.

[Jordi]

Super impressive traction. I know noticed. Is it Ronick on your team? Was that Red Bull racing before this? How cracked is that's awesome.

[Guest]

He is, he is very, very hardcore.

[Jordi]

Like I imagine if you want to make something that's ultra light, ultra durable, he's your guy.

[Guest]

That's right, that's correct, yeah. So a third of Our team is SpaceX, Tesla Ronix. Worked at Tesla before Red Bull racing. And also SpaceX as well. But he's definitely a character. Yeah.

[Jordi]

Awesome. Well, great to meet you. I'm excited to follow along round. Already done. How's it going? Yes.

[Guest]

Raised a lot of money.

[Jordi]

There you go. Hit the gong again, John.

[Guest]

There we go. Yeah, buddy. Yeah, buddy. Yeah.

[Jordi]

Just coming on. Absolute legend. You're a TVPN legend.

[Guest]

Yeah. Thank you.

[Jordi]

We might have to make a TVPN tie dye shirt in your honor.

[Guest]

Yeah, let's do it. I love it.

[John]

Thank you.

[Jordi]

Very cool.

[John]

Well, have a good rest of demo day. Congratulations on all the progress. Very excited to see these up in the stratosphere. Just don't fly them too high.

[Guest]

Yep, exactly.

[Jordi]

Perfect.

[Guest]

All right, Jordy, John, thanks so much.

[John]

Have a good rest of your day. Goodbye.

[Jordi]

What a legend.

[John]

I need to know from you if we have some breaking news that we can share right now. It sounds like we might have some surprise guests joining the stream. So stay with us. I will also tell you about adquick.com out of home advertising made easy and measurable plan buy and measure out of home with precision. Also, people are calling for Google to make glasses now because Google Glass, they did this, this 20 years ago, practically Google Glass, but they're still working. Yes, yes, yes, yes. But through partners. Through partners. So they've done the Google Pixel, they've done a variety of hardware devices and they are working on some augmented reality glasses again. But they're certainly not making as much of a big push media push as they did with the original Google Glass, which was like, it dominated the news and it was like the future is here. And then the product didn't really get to Escape Velocity and is sort of remembered as a failure. But it wasn't a failure. They were just early. They were just early. And that's the important thing to remember. But we have our next guest here in the Restream waiting room. Let's bring him in from Locus. Welcome to the show. Thank you so much for taking the time to join us. Please introduce yourself and tell us what you're building.

[Guest]

Yeah, for sure. So I'm Cole Durmont. I'm the CEO and co founder of Locus. We build payment infrastructure for AI agents.

[John]

Okay. MCP currently doesn't have payment infrastructure. That's why you exist. That's what's going on.

[Guest]

Yeah, Basically, plus trust.

[John]

Okay.

[Guest]

Trust is a huge part of it.

[John]

Okay. Interesting. How are people, Are people actually like solving this manually right now? Are there, are there payment? Are there like, are there like agent to agent payments that are happening right Now? Or is this something where we're thinking, like, in the future they will all be flowing stablecoins to each other in the future?

[Guest]

I think agent to agent isn't really adopted yet. What we're looking at right now is more so developer use cases of. If you're familiar with X402, paying for.

[Paul Graham]

API endpoints on a pay per use.

[Guest]

Basis for potentially doing payouts to people. The way I like to explain it is historically payment automation has been deeply rooted in conditional automation, a series of ifs, ands, ors, et cetera. Now, with agentic payments, you open up this new frontier of contextual automation. Right. And that's a pretty huge evolution.

[John]

How do you imagine the first adoption of agent to agent payments, or even just payments for agents broadly playing out? Me and Jordi have been talking about this with the agent of commerce stuff. We're using ChatGPT, we're using Gemini. There's all these times when I run into a paywall and I can tell it's running into a paywall. It's like, actually, I can't tell you about what's going on on that website. And I'm like, you actually could if I gave you my credit card. I know you could, but they can't. And it seems like that's something you could potentially help with. But how do you see the first early adopters using your service?

[Guest]

I see the first ones as really.

[Paul Graham]

Developers building these autonomous agents.

[Guest]

Right. Being able to essentially pay for services as they do their workload in the wild and discover those services. Services autonomously. Right. In terms of like, the more commerce side, I think that'll be an industry that evolves over the next few years as trust is really developed, because frankly, on a, on a wide scale consumer basis, that's really the biggest barrier right now is trust rather than tech.

[Jordi]

Yeah. What kind of numbers did you share during your pitch or are you planning to share?

[Guest]

Yeah, so we processed around 3,500 transactions and have around 80 projects built using Locus so far.

[Jordi]

Amazing. What were you doing before this? John's got the gong for you. Hit it, hit it then.

[Guest]

What do you do?

[Jordi]

What were you doing before this?

[Guest]

Yeah, so I interned at Coinbase.

[John]

I was one of the people who helped build Coinbase business over there.

[Guest]

My co founder was one of the six software engineering interns at Scale AI, studied CS at Waterloo business at Wilfrid.

[Paul Graham]

Laurier, was the finance lead at Waterloo Blockchain.

[Jordi]

So Waterloo mentioned.

[John]

Fantastic. Well, thank you so much for coming on the show. Congratulations.

[Guest]

No problem.

[John]

And I'M sure we'll be seeing you soon. Have a good rest of your day.

[Guest]

Thank you.

[John]

We'll talk to you soon. Let me tell you about wander.com, book a wander with inspiring views, hotel grade amenities, dreamy beds, top tier cleaning and 247 concierge service. It's a vacation home, but better. And we have some surprise guests I believe joining in just a second.

[Jordi]

We will have them in Jessica and Paul. You may know them. They started a small startup accelerator called Y Combinator.

[John]

Yes, that's right. And it's Jessica's second time on the show. We had a fantastic conversation with her the last time she was on the show. We talked about the get your bag culture and the carpetbaggers and just all the cultural ebbs and flows of Silicon Valley and where we are culturally. So very excited to bring in Jessica and Paul, the founders of Y Combinator. They are living legends situated. Living legend. Everybody, this way. You're live now. Welcome to the show. Thank you so much for taking the time to talk to us.

[Guest]

Hi, guys.

[John]

Hi, guys.

[Jordi]

Good to see you.

[John]

Good to see you.

[Guest]

This is so fun with you guys.

[John]

Here at demo day. It is, it is. It's always great. This is our fourth demo day livestream, talking to tons of founders. It's always fun picking out.

[Jordi]

I can't wait for the 400. We got a ways to go, but I'm excited.

[John]

100 more years, we'll make it great.

[Jordi]

To have you guys on. What's it been like today?

[Guest]

It's been crowded, it's buzzing. And by the way, this is our first demo day that we've been to in a few years because we're in.

[Jordi]

England and can't manage to come back for it.

[Guest]

It is just buzzing. The energy here is just kind of like what I remember in the early days of YC and the investors are.

[Jordi]

All excited to be here.

[Guest]

It's magical. I'm on a high.

[Jordi]

Incredible.

[Paul Graham]

There's a lot of stuff happening.

[John]

Yeah. How are you thinking about? There was this moment a few years ago where I think in tech, maybe we were afraid to admit it, but it felt like a lot of founders and a lot of entrepreneurs were sort of grappling with this idea that OpenAI might just build every startup and there might be no more ideas. And people were a little bit nervous about that. Of course they went and built companies. But it feels like now, now things have calmed down a little bit and the founders that we talk to are building with more confidence. Have you noticed anything in the founders that you talk to in an ebb and flow of just the confidence with which they view the future right now?

[Paul Graham]

No, no, they weren't. Founders weren't really worried that OpenAI was going to eat them. I mean, maybe they were in denial, but whatever reason, they weren't worried about it. They're too busy working on their companies. They're making their thing, they're trying to get users. Is OpenAI eating them in some theoretical future three years from now? Like, they're not thinking about anything three years from now, so they're not thinking about that.

[John]

We had another. We were talking to Harj about this, this idea that potentially, I don't know, we're just in a new era where.

It has become easier for a small team of scrappy entrepreneurs to sell to Fortune 500 companies, to sell to the government even. And do you feel like something has materially changed in go to market for YC companies?

[Paul Graham]

Well, if you're an AI company, all these big organizations now have some bureaucrat who's been told you're supposed to AI ify our organization, right? And he's thinking, damn, I have no idea what to do. And so some startup shows up and says, we'll AI ify organization like, great, come in here.

[Guest]

Very different from the way it used to be.

[Paul Graham]

I mean, if you show up with other products for the big company, they'll still tell you to talk to the hand, but nobody's coming to them with AI things except startups. So they have no choice but to talk to startups.

[John]

What about this tweet that you put out just recently? We were sort of debating it earlier. This idea of the circular economy selling to other startups. There are a ton of benefits. Obviously, startups are very discerning. If you mess up and don't deliver the product that they buying from you, you might hear about it publicly. They'll churn, they'll talk to you, they'll talk to their friends. But are there any risks from that that you caution entrepreneurs on? If they are going to be selling to a lot of startups, do they have to message anything differently? Is there anything that they need to be doing differently?

[Paul Graham]

Well, you have to not suck because startups are discerning. You can't have some bullshit product and sell it based on a bunch of hype. It's got to actually work because they don't have time to mess around with things that don't work. And they're very sharp observers of technology. They're run by the founders themselves, usually at that point. So you got to actually be good.

[John]

I'd love to reflect on how marketing and launching startups has changed over the last few decades. Jordi. We had clad labs on which we had a really fun time talking to them. But they, they sort of went viral for the wrong reasons. They were.

[Jordi]

Well, in their view is the right reason.

[John]

In their view is the right reasons. They were offending people by putting gambling in your ide. So the software engineer can be gambling while they're coding? I guess.

[Jordi]

Yeah. And it felt like this year the concept of using rage bait, both at the marketing level and the product level kind of exploded. I guess the question is, has intentionally pissing people off been something that YC founders have utilized across the eras to get attention? Is it really new?

[Paul Graham]

That sort of technique sounds like the technique that would be popular with someone you describe as a bit of a scammer. And the thing about these scammers is they don't make the giant companies. They don't have a long term focus. They're not earnestly doing engineering. They're thinking about what's some gimmick I can use to get ahead. Right. And so long term they don't matter. You can skip the companies that do random shit like that because they're never going to be that big.

[Guest]

And of course, I haven't heard of.

[Jordi]

The term rage baiting either.

[Guest]

Of course.

[Jordi]

It'S the Oxford word of the year, so you can go look at their definition.

[John]

It's so interesting.

[Paul Graham]

It's getting attention by making people.

[Jordi]

I know what it means. Yeah. And I had written an article and Gary and I had a nice back and forth where I basically said, like in startups, in startups you need to build a coalition of people that want you to win. This is like talent. The media, investors, customers.

[Paul Graham]

You don't even have to do that actually. All you have to do is make something really good and find the people who want it. You don't even need a coalition. You think like when Facebook was taking off at Harvard, there was some coalition of investors and media wanted it to take off. All that mattered was that Zuck had this thing and everybody at Harvard wanted to use it. That's all that matters. That small, intense fire, Right. Where when Apple was getting started and the users were like the people at the Home Brew computer club. Right. The media didn't know about that.

[Guest]

There was no coalition.

[Paul Graham]

Media guys do a little rage.

[Jordi]

Zuck kind of did a little rage bait.

[John]

Rage bait with the Hot or Not app. That definitely enraged a lot of People who didn't want.

[Paul Graham]

Yeah, but he didn't do it deliberately.

[John]

No, exactly. Exactly. And I was thinking about the Airbnb example, like the whole Obama O's and Captain McCain's crunch. Like those cereals that they made. That was sort of a side quest for them.

[Guest]

That was simply to get attention from the press.

[John]

Interesting.

[Guest]

That's all.

[Paul Graham]

And make. No, actually it was to make money.

[John]

It was to make money.

[Paul Graham]

That was before yc. They didn't have any money. Remember, they were dying, they needed to make money. They went and got these like off brand Cheerios and they glued together the boxes themselves. But I don't think money. I don't think they knew they were.

[Jordi]

Going to make money.

[Guest]

We're going to have to consult.

[Paul Graham]

That was mainly done.

[Jordi]

What's it like being back in San Francisco, Sunny?

[John]

It's fabulous.

[Guest]

The energy is so great here. I'm just so. I'm so happy to be back and so happy to be around startups right now. I'm having a great day.

[Paul Graham]

If you can't tell. It gets better every time we come back. Like, Daniel Lurie is really cleaning up the city. Every time we show up, it's like a little better.

[John]

That's great news.

[Paul Graham]

I was asking, like, how far back have we gone? Have we gone all the way back to when Ed Lee died? Not yet. We're like. But we've turned the clock back to maybe two years into London breed.

[John]

Oh, that's good. Okay. Yeah, that's great.

I have one more. I want to think through this concept that's been sort of lightly bandied about in the startup discussion ecosystem. This idea of the deals guy era, that you can actually build a business now by being more of the business person, the more of the deals guy and less of what I remember about the. The Y Combinator promise, which was just. The earnest hacker. The earnest hacker. The earnest hacker. And it feels like there's a lot of people that are saying, yeah, but there's actually a way to go and get this person, just marshal the capital and do something that's just been forgotten, not necessarily discover something new. And I was wondering if you have any reactions to.

This idea that increasingly there are entrepreneurs that that sort of get really big, who knows if they win, but they seem to win on the back of just raw deal making talent as opposed to raw engineering leadership.

[Paul Graham]

Maybe an enterprise.

[Guest]

More.

[Paul Graham]

Like enterprise. You sell crap to CTOs instead of selling good stuff to programmers.

So salesmanship has always mattered more in enterprise.

[Jordi]

I have one observation from this morning.

[Guest]

Session of Demo Day.

Most everyone that presented this morning is an earnest hacker. I said to the person next to me, they're all nerds this time. Like 100%. I love it. Yeah.

[Paul Graham]

You know, if anything, YC drifted too far away from funding earnest hackers. And so YC for the last few years has been focusing more on, like, getting back to the essential, back to the roots. And so, if anything, I would say YC batches are more like a higher percentage earnest hackers now.

[Guest]

Yeah.

[Paul Graham]

You know, honestly, I would still bet on earnest hackers.

[John]

Yeah, I agree with you. Do you think that that is what the essential skill set of YC leadership needs to be? Because I don't want to discredit all the hard work you did in the early days, but you didn't have to fight the fact that there were people out there writing blog posts of how to reverse engineer and make it look like you're an earnest hacker, when in fact you are the, you know, the carpetbagger. And now there's a whole industrial complex for how to fake your way and make it appear that you're in earnest, when in fact, you're not.

[Paul Graham]

If the YC partners are themselves hackers, you can sniff out a faker like that. It's not even a problem.

[John]

Yeah, yeah, yeah. But that seems like the main way that YC creates value these days will just be continuing to. To hold that line, essentially.

[Jordi]

What do you think is your most.

[Paul Graham]

I think, you know, here's something that will reassure you if you think, okay, is the earnest hacker thing. Did that just work for a while, and now maybe it's over? Isaac Newton was an earnest hacker. It's way older than startups.

[John]

Yes, yes, yes.

[Paul Graham]

This is what wins.

[Jordi]

What do you think is your most underappreciated essay?

Because a lot of them are sufficiently appreciated.

[Paul Graham]

The thing is, I don't know how much people appreciate them. I don't know how much people appreciate different ones. So it's hard to say.

[John]

How to.

[Paul Graham]

Do great work is pretty good. But I think people like that one, right?

[Jordi]

I read life is short at least.

[Guest]

Once a year, but people like that one, too? I don't know.

[Paul Graham]

I don't know. That's a weird question.

[John]

You'd probably have to look at inverse page views. Which gets the least page views historically over the past year?

[Paul Graham]

Let's say if I was looking at a list of page views, I can tell you.

[John]

Okay, well, maybe I never.

Breaking news. Yeah, that's very funny. Do you have anything else?

[Jordi]

What else, Paul? Are we In a bubble?

[Paul Graham]

No, no. Everybody is always saying we're in a bubble. You know, like every year people say we're in a bubble. Every year people say, like, the valuations at Demo die, they're too high now. They were saying this back in like, like 2010 when the valuations were like $4 million, and now they're like, what, 30 or something, typically. So people are always saying stuff like that. And I don't know. I don't think so.

[Jordi]

I think.

[Paul Graham]

I'll tell you, I think, like, AI is very highly priced, but it might not be overpriced. That's the interesting thing. Is it as big a deal as prices seem to suggest? It could be. Maybe even bigger. It's definitely real. It's not hype. The AI is real.

[John]

Are foundation models good at writing Lisp?

[Paul Graham]

You know, I've never. I think they would be good at writing Lisp. Yes, yes. Because they're good at writing things that have a lot of. A lot of. A lot of training data out there. Right. And there's a lot of Lisp source code. So I think they'd be fine at writing that.

[John]

How are you using AI in your life?

[Paul Graham]

I just use it like ordinary people do I ask you questions?

[John]

Sure, sure.

[Guest]

Very boring answer.

[Jordi]

That's a good answer. It's not like, oh, I stringed. I'm training my own model to do a better Google search. No, no, no.

[Paul Graham]

I haven't actually written anything using AI. You know, I feel bad. I really should write an ll because you can't really understand this stuff unless you've written one. I should write an LLM, but I haven't done it.

[John]

Yeah. Didn't Karpathy publish a whole? Yeah, he did it. First principles type of thing, you know.

[Paul Graham]

To teach himself, you know, that's why he did it.

[John]

Well, he has a new company that's an education technology company, and I believe that the. The main course will be teaching yourself to build an LLM, teaching yourself to build a chatbot effectively, which would be very.

[Paul Graham]

That's what I tell high school kids. I get all these emails from high school kids, say, I'm working on a startup to introduce founders to VCs or some crap like that. And I say, don't start a startup, get good at technology, write an LLM, then you can start a startup.

[John]

Do you think, reflecting on the history of yc, do you think it's fair to try and create a concept of eras around.

What the key insight was? I remember a lot of people saying one of the first key insights was j. Just this idea that you could take someone fresh out of college and actually give them money and they could go and build a business. They didn't need $10 million. They didn't need 10 years of experience in the enterprise.

[Paul Graham]

Or an MBA.

[John]

Or an MBA. And then maybe the second era was thinking that maybe the same rules applied internationally. And that was like a second wave of entrepreneurial energy that was unlocked by the ycu.

[Paul Graham]

We had Internet, we always had international. We understand countries aren't all that different.

[John]

But do you think there are any other underappreciated aspects of the YC strategy or is it really just as simple as.

[Paul Graham]

Well, there were things we didn't appreciate in the beginning. So for example, we didn't understand that as a byproduct of funding all these companies, we would create this alumni network. We had no idea. But the alumni network is enormously important. It's out there now. All these alumni are investors.

[John]

Yes.

[Guest]

It's staggering how many are investors now, actually. Yeah, it's amazing.

[Paul Graham]

It's like taking over Silicon Valley and we never had any idea that was going to happen.

[John]

Okay. On the alumni network. Is it fair to characterize YC as a bit of a union against venture capitalists?

[Paul Graham]

Yeah, it's a lot like a union.

[John]

Yeah. Because if you attack one individual, one founder, if you fire the founder after investing, you get board control from them and you oust them, that might make its way into the rest of the YC community and it overall raises the level of founder friendliness. Is that correct?

[Paul Graham]

You know what though? It's not simply one sided. Because if founders screw over investors, if they like do a handshake deal and then refuse to go through with it, we would tell them not to do that too. We want everybody to play by the rules.

[Guest]

Yeah. And behave well.

[Paul Graham]

Because the big wins don't come from breaking the rules. The big wins don't come from little cheats that get you 2x multiples in a world of thousand x returns.

[John]

Right.

[Paul Graham]

It's for the same reason people in Silicon Valley don't focus a lot on tax evasion. Because what's tax evasion going to get you? Like 2x returns in a world where getting the right, right startups, we'll get you a thousand extra turns.

[John]

Do you think that the process of founding a company, raising money, is at the end of history in terms of efficiency? Like the safe is the most efficient document we will ever have, or do we need to speed things up even further?

[Paul Graham]

Well, C. Levy. Carolyn Levy. Invented the safe, and she also invented the convertible note that everybody used before it. So she has twice rewritten the rules. She's twice recreated the chessboard that the game is played on. If she thought there was a better thing than the safe, she probably would have created.

[Guest]

Maybe she has a third one in her.

[Paul Graham]

We should ask.

[Guest]

We'll ask.

[Jordi]

Trilogy.

[John]

Very popular.

[Guest]

Oh, yeah. Okay. Yeah. And you could ask her that, John, when you come on our podcast.

[John]

We'd love to. I'd love to. Love to. I can't.

[Paul Graham]

Is there anything wrong with the safe? And, like, if there is, why hasn't she fixed it already, you know?

[John]

Yeah.

[Paul Graham]

So probably not. Because C Levy's not slack. If there was anything missing, she would have, like, made a new version.

[John]

Yeah. I mean, from my perspective, it seems like it's worked.

[Jordi]

What problem in the world did you think a YC startup would have fixed by now? Think, like housing affordability or any of these sort of major, you know, we.

[Paul Graham]

Don'T have any grand strategic vision for what the startups do because the founders know that, not us.

[John]

Right.

[Paul Graham]

That would be like asking a publisher, what. What novel do you think, you know, would you have expected someone to write? Right. Good publishers, they just, like, they let the. They let the novelists write the novels. So we would just. We just try and find good people. What do they do? Whatever these good people are interested in. Anything. Any preconceptions we had about what they should do would just be adding noise to that.

[John]

How do you think about coaching folks through pivots? It feels like we're in an era where there's a lot of companies that are still finding product market fit. Pivots are probably just as common as they always have been, but everyone has an order of magnitude more money, if anything.

[Paul Graham]

More common?

[John]

Yeah.

[Guest]

Yeah, I think it's more common.

[John]

You talk about new ideas with startups.

[Guest]

All the time in your office hours.

[Paul Graham]

This is one of my. My specialty.

[Guest]

Yeah.

[Paul Graham]

When people are just dead in the water and they need to get a new idea, they often get sent to talk to me and we cook up something.

[John]

Has the advice changed? If someone comes in and says, hey, I have $200,000 raised and I have. Me and my co founder are living in an apartment together and we need to pivot versus I come in and I say, hey, Look, I got 5 million bucks and I got 20 employees already or something like that.

[Paul Graham]

20 employees.

[John]

It's happening, right? It's a.

[Guest]

You.

[John]

You do see this, right?

[Paul Graham]

Well, no, usually they don't have 20 employees.

Usually usually, I mean, that would be, that would be alarming. That would be very alarming because there's.

[John]

So many companies, those 20 employees constrain.

[Paul Graham]

The idea you're going to have. If you just have the founders, you could do anything. If you already have 20 people, you either have to fire them them or do something that those 20 people can do.

[John]

Right.

[Paul Graham]

Which really constrains your options. So it's the problem with the 20 employees is not the cost, it's that they change what they limit what you can think of.

Which is why you shouldn't hire.

[Jordi]

What kind of guidance do you give to founders that are feeling a pressure to go from 0 to 100 million in ARR in 3 years or whatever? Like the new gold standard is what.

[Paul Graham]

I tell startups over and over and over is all that matters is growth rate, not the absolute numbers. Because mathematically you'll see if you try simulating it, if your growth rate is high enough, doesn't matter what the absolute numbers are, you'll get there. And so you just get a really good growth rate. And so the great thing about focusing on growth rate means you can focus on startups. You can sell stuff to startup startups for cheap. Instead of having to go and do these big deals with big companies that take a long time and make your products stupider, you can sell things to these quick deciding early adopters and then you just get more and more of them and your company grows by several percent a week. Eventually it's going to be huge.

[John]

Are you still recommending to folks who ask for advice for kids that they should learn to code?

[Paul Graham]

Yeah, oh yeah, yeah, yeah. I still tell people that, or at least learn technology. It doesn't have to be coding specifically. You can learn how to make rockets.

Or drones or work with lasers or gene editing or something like that, but you should do the stuff and not just like play house pretending to start fake startups in some business plan competition.

[Guest]

You know.

I tell everyone who, who says they might want to start a startup someday to learn to code because.

[John]

It'S the most important thing.

[Guest]

You could do that and save your money.

[John]

Yeah, that's really good advice.

[Guest]

And no one likes to hear that, by the way, but I tell them anyway.

[John]

Yeah, no, no, we give a lot of advice.

[Paul Graham]

People come and they want advice. It's like if you went to the doctor and you said, doctor, what can I do to be healthier? And the doctor says eat less and exercise more. And you're like, oh, I was hoping you'd say something else.

[Guest]

Right.

[Paul Graham]

Well, that's what it's like when they come to me, they come to me for advice and I say the startup equivalent of eat less, exercise more, and they're like, oh, isn't there some trick I could use to get virality? Couldn't I get virality instead? Just like, do the startup equivalent of eat less and get more exercise, which is build stuff and talk to users, understand your users and be good at building. That's the recipe it was in 2005, and it's just as much the recipe now.

[John]

Yeah.

[Jordi]

How many startups do you think. How many startups do you think YC will have per batch a decade from now? Because I think in a perfect world, we have a lot more earnest hackers and.

They can apply to yc and if they meet the. I know you're not setting targets and there's not like a specific acceptance rate that you're trying to track. But, but we feel like YC is one of the most important institutions in the world. And ideally it can be bigger, but maybe there's some.

[Paul Graham]

No, no, no, they will be bigger. They will inevitably be bigger because there's this secular trend of more people starting startups.

[Guest]

Yeah.

[Jordi]

Do you think we're early in this trend? I mean, it feels like there's so much. It's. Now you can create a startup, you can create a C corp in a few minutes.

[Guest]

Right.

[Jordi]

It's like there's all this sort of like underlying infrastructure that's been built that is reducing friction to starting companies. You can ask ChatGPT, how do I start a business? And it'll give you a good playbook. And that maybe helps somebody that hasn't found the YC blog yet figure out how to get going.

[Paul Graham]

Where are the training data, even if they don't know it.

[Jordi]

Yeah.

[Paul Graham]

So will more people start startups? Yes. If you talk to like ambitious people, 15 year olds, they all want to go start startups. Nobody wants to go work for some company and work their way up the corporate ladder anymore. The whole idea sounds so like, sounds so like 1980s.

And there's a lot of earnest hackers. The limit and the limit. You think, like, what's the limit? So the limit is what people want.

[Guest]

Right.

[Paul Graham]

That's what startups do. They make something people want. What are, what are people's. People's wants?

[John]

They're limitless.

[Paul Graham]

Not literally limitless because eventually you're out of atoms in the universe, but for all practical purposes, in the near term, people's wants are infinite. And so there's infinite demand for good stuff you could make.

[John]

Well, that's a great place to end it. We have to catch a flight. Thank you so much for taking the time to talk about it.

[Guest]

Yeah.

[Jordi]

Thank you for everything you guys have done for the industry and the world through yc. It's an honor to be here with that one. It's an honor to cover every batch and it's been great having you guys on.

[John]

Yeah.

[Paul Graham]

Nice to meet you.

[Guest]

Thanks for having us. I love you guys.

[John]

Yes, we love you.

[Jordi]

Love you, too.

[John]

Thank you so much.

[Jordi]

Have fun in sf.

[John]

Have a great rest of your trip. We'll talk to you soon.

[Guest]

Thanks.

[John]

Goodbye. We have to hop on flight, but.

[Jordi]

You hear that, John?

[John]

I hear it, yes. I hear the goat noise. The sound cue. That one's a little bit subtle. I think that there's a lot of people that might not pick up on why they're hearing this random goat noise.

[Jordi]

If you know, you know in the back.

[John]

But if you know, know, you know. And also, if you want exceptional sleep, without exception, you go to eightsleep.com, you fall asleep faster, you sleep deeper, you wake up energized. And we should close out. There's a lot of stuff going on. Deal. Book summits going on. There are debates raging on the timeline, but we will have to cover them. Tomorrow we will close out with a congratulations to Ed Elson, the co host of the Prof. G Market podcast. I love his bio because he says he's not Prof. G's son, even though they look somewhat similar.

[Jordi]

Yesterday because he got into Forbes 30 under 30 and he said, I'll see you guys in prison.

[John]

He said, woke up to learn I made Forbes 30 under 30. Congrats to the other winner winners.

[Jordi]

Can we play this before we, can we. Before we jump, can we, can we play this? Oh, Gary Tan's in the chat. Ali's in the chat.

[John]

Gary, we hope you feel better.

[Jordi]

Jana's in the chat. Gary, feel better.

[John]

Thank you so much for making this happen. We're very sorry we couldn't be there in person, but we had a blast. We went on a whirlwind tour, we talked to tons of YC founders, and the state of YC is healthier than ever, stronger than ever.

[Jordi]

Gary's got elementary school or preschool.

[John]

It's so rough. I've been there, man. I've been there. It's daycare virus.

[Jordi]

Yeah.

[John]

Well, we hope you get well soon, team. We need to definitely send some soup or some flowers to Gary Tan as soon as possible. And we will see you all tomorrow.

[Jordi]

Thank you for tuning in. Thank you to Y Combinator for hosting us and all the founders. It was a whirlwind tour and I'm very excited about a lot of these companies.

[Guest]

Yes.

[John]

We will talk to you later.

[Jordi]

Cheers.

[John]

Goodbye.