The Chopping Block: RWA Perps Go Parabolic, ClawdBot, & Superstate's $82M Raise

Summary

---

Unchained – The Chopping Block: RWA Perps Go Parabolic, ClawdBot, & Superstate's $82M Raise

Podcast Host: Laura Shin
Panelists: Haseeb Qureshi, Tom Schmidt, Robert Leshner, Tarun Chitra
Date: January 29, 2026

---

Episode Overview

This episode features the regular Chopping Block panel discussing the most compelling trends at the intersection of crypto and traditional finance. Key themes include the explosive rise of real-world asset (RWA) perpetual markets, the innovations and risks around AI assistant bots (especially ClawdBot/Moltbot), and Superstate’s headline-making $82 million raise. Using a blend of insider knowledge and sharp humor, the crew unpacks both the latest on-chain developments and the broader shifts being driven by AI and tokenization.

---

1. [00:29] Superstate’s $82M Fundraise: Bringing Wall Street on Chain

Key Points

• Superstate’s Mission: To tokenize ‘official’ company equity shares directly—enabling the same stock that trades on NYSE/Nasdaq to also exist as a token on Solana or Ethereum, with seamless transferability between blockchains and brokerage apps.
  • “The unique thing about the Superstate approach to tokenization is it's the company's official shares tokenized … they can do official company things with those shares, like sell them for money, which is the point of being public in the first place.” —Robert [02:28]
• Investment and Growth: Raised $82.5M from leading crypto VCs (Bain Capital, Galaxy Digital, Brevan Howard, more) to expand engineering and execute ambitious tokenization pilots.
• Types of Tokenization: The team discusses naming conventions for the various models:
  • Bootleg Tokenization: Third parties wrap stocks as tokens without company endorsement (e.g., XStocks) — minimal KYC, focused on offshore use.
  • Back Office Tokenization: NYSE/Nasdaq settlement improvements using tokens for efficiency (no DeFi integration).
  • Issuer-Led Tokenization: Core Superstate approach, whereby tokens directly represent officially issued company shares.
  • “There's going to be like 90 competing bootleg standards.” —Robert [07:11]
• Geographical and Regulatory Nuance: Bootleg models are often unavailable for U.S. users for regulatory reasons, but are popular in other jurisdictions.

---

2. [10:18] RWA Perpetuals and the Trade XYZ Boom

Key Points

• Massive Growth on HyperLiquid and Trade XYZ:
  • Open interest in RWA perps (commodities, stock indices) has surged, with Trade XYZ becoming the 7th-largest perp DEX by open interest.
  • Gold and silver perps lead the growth, mirroring surges in “real world” commodity prices.
• Liquidity Fragmentation Shift:
  • Brief period of fragmentation across competing RWA venues, but Trade XYZ quickly established near-monopoly in volume and users.
  • “There are tons of memes of like, I searched for Tesla and I got four different Teslas. Which one do I trade? Then this current boom completely swung everything back market share wise to Trade XYZ—it's like over 90% now.” —Tarun [12:34]
• User Demographics:
  • Notable share of trading concentrated among mobile and Asian users, reflecting broader trends in Asian crypto engagement.
  • “In Asia they're mobile dominant. It's not the ultimate signal, but it’s clear things are different than the pure crypto side.” —Haseeb [14:45]
• Crypto Cohort Maturing:
  • Crypto traders increasingly chase volatility across asset classes, not strictly crypto; they are “loyal to the game, not the asset.” [22:33]
  • As spot crypto activity stagnates, the same crowd drives commodity perps and newer RWA on-chain derivatives.
  • “If you're there for speculation...they're not loyal to one asset to trade…They’re loyal to the game and they'll go where the action is, just like poker.” —Robert [22:33]
• Implications for Exchanges:
  • Exchanges (like Coinbase, Robinhood) now seek to retain this aging, sophisticated cohort by adding more "everything" trading: stocks, metals, even prediction markets.
  • “They're trying to hold on to the crypto trading cohort and keep them all within the same app experience.” —Haseeb [24:31]
  • As platforms age, most fail to capture new generations—suggesting new entrants will emerge, likely with entirely new interfaces or AI-driven UX.

---

3. [28:11] ClawdBot/Moltbot, AI Agent Coding & Crypto’s Meme Coin Reflex

Key Points

• ClawdBot (Moltbot) Phenomenon:
  • ClawdBot, rebranded Maltbot after a naming tangle, is an open-source AI agent (based on Anthropic's Claude) that can automate workflows across email, calendar, chat, and more.
  • Twittersphere is obsessed: some tout it as a personal assistant revolution, but panelists are skeptical about widespread real-world utility.
  • “There’s so much…classic in AI. There’s a Twitter demo of some AI influencer who talks about, 'I now have 20 employees of Cloudbots running on my desk'...what are they doing, exactly?” —Haseeb [35:49]
  • Security risk: Many instances exposed online without authentication, making them “very easy to prompt inject…to steal keys, to go in and just take control over someone's Gmail.” —Haseeb [31:20]
• Crypto’s Reaction: Meme Coins and Beyond:
  • Multiple meme coins sprung up tied to ClawdBot and similar AI projects, rapidly speculating on the hype.
  • “Crypto has decided to get involved in the way that crypto knows how, which is to just launch tokens about claudebot...Once again ruined a perfectly good thing.” —Haseeb [32:13]
  • Pattern repeating from earlier chatbot hype cycles (virtuals, AI meme coins).
• Why Bots Like Maltbot Are Popular:
  • Brings AI tools closer to non-technical users, with simple, familiar ways to talk to models (text, messaging).
  • Most users, though, struggle to get practical value beyond what standard chatbots already provide.
  • “People are just bored and want something to tinker with.” —Tom [35:44]

---

4. [41:10] Agentic/Vibe Coding: How Claude Code is Transforming Software Engineering (and Crypto Startups)

Key Points

• Rise of Agentic Coding Tools:
  • Claude Code makes it possible for non-developers to build working software from natural language prompts.
  • “I really do think that now the speech that I gave a year ago is now correct, which is that it actually does feel true now that you can just tell Claude a verbal description of what you want to build and it will just build it.” —Haseeb [47:04]
• Inside the Panelist Companies:
  • Nearly all technical staff at Gauntlet, and portions of Superstate and Dragonfly, now use LLMs or agentic tools for almost all significant tasks—especially documentation and boilerplate code.
  • “Almost everyone, including myself, is literally looking at a cloud code terminal every time I look to the left on this podcast.” —Tarun [44:06]
• Shift in Engineering Skills:
  • Writing code by hand is rapidly being replaced by reviewing/curating model output. The “individual contributor” engineer role is waning in favor of software orchestrators/“engineering managers".
  • “You have to basically unlearn…everything you learned in some ways to use these things properly.” —Tarun [51:13]
  • “It's a total devaluation of the skills that previously mattered as a software engineer.” —Haseeb [51:30]
• Implications for Startups & Investment:
  • The startup landscape bifurcating: those built “before” wide AI agent adoption will have slower iteration, higher costs; founders and CTO choices must now factor for strategic leverage of AI coding.
  • “Clearly it's going to be a huge difference in the startups that are hiring people based on this…how they architect their engineering orgs.” —Haseeb [52:45]

---

Notable Quotes & Memorable Moments

• On Tokenization Models:

  • “Back office tokens...these are businesses that approach things from the complexities of settlement.” —Robert [07:47]
  • “Bootleg because there's going to be like 90 competing bootleg standards.” —Robert [07:11]

• On Shifting Trader Behavior:

  • “The only time people are happy with bitcoin is when it goes up. If it doesn't go up, it's never quite behaving the way people want it to.” —Haseeb [19:48]
  • “They've completely forgotten about crypto...the toy story meme, they just like threw away the doll.” —Tarun [20:20]

• On Aging Crypto & Trading Cohort:

  • “They're just tracking with this cohort as they get older and older. And like, I think, you know, crypto exchanges are probably going through something similar.” —Tom [25:05]
  • “Soon they're going to have a target date fund per ETFs or perps. And so it's almost all the most boring assets.” —Tom [26:28]

• On ClawdBot Meme Coin Madness:

  • “Once again ruined a perfectly good thing because he's now said…the reason I brought up the truth terminal thing…someone who’s in tech who’s never touched crypto, who suddenly was like, I made X million dollars…I have zero.” —Tarun [33:01]

• On Agentic Coding’s Impact:

  • “If you're still writing code by hand, you are kind of holding on to this previous generation of coding.” —Haseeb [50:38]
  • “Everyone is a manager, engineering manager, no one is an ic—that’s what’s happening, right?” —Tarun [51:40]

---

Timestamps for Important Segments

• [01:42]: Superstate’s Series A, expansion, and capital strategy
• [06:47]: Differentiating tokenization models (bootleg, back office, issuer-led)
• [10:18]: Surge in RWA perps and commodities trading on Trade XYZ/HyperLiquid
• [20:20]: Crypto traders moving on from crypto to volatile global markets
• [24:31]: Exchanges attempting to retain users by becoming “everything” apps
• [28:11]: Introduction to ClawdBot/Moltbot and the AI tool hype wave
• [33:01]: Meme coin frenzy around AI bots and the dark side of tech-crypto crossovers
• [44:06]: Agentic coding tools at work in panelists' organizations
• [51:13]: Panel reflections on the need to “unlearn” old software habits
• [52:45]: Impact of the AI coding revolution on the future of startups

---

Takeaways

• Crypto and TradFi boundaries are blurring, with tokenization increasingly mainstream but models and regulatory realities diverging.
• A new era of on-chain trading has arrived, as derivatives on real-world assets gain market share and reshape trading behavior.
• AI agentic coding is rapidly democratizing software creation, fundamentally altering the cost, cadence, and structure of startups.
• Crypto’s meme reflex will find a narrative—no matter the innovation, the sector will trade and speculate on it, especially when novelty meets hype.
• The future of crypto interfaces and communities may depend on who captures the next young cohort—with AI likely to be at the center.

---

Transcript

[Tarun]

They've completely forgotten about crypto. Or like not forgotten, but they're just like that toy. They've, you know, the toy story meme. They just like threw away the doll. It feels like I don't want to.

[Tom]

Play with you anymore.

[Tarun]

Yeah, exactly. It feels like that a little bit.

[Robert]

Not a dividend.

[Tarun]

It's a tale of two Kwan.

[Haseeb]

Now your losses are on someone else's balance sheet.

[Tom]

Generally speaking, airdrops are kind of pointless anyways.

[Haseeb]

I named trading firms who are very involved.

[Tarun]

Alec Eth is the ultimate pun.

[Robert]

Defi protocols are the antidote to this problem.

[Haseeb]

Hello, everybody. Welcome to the chopping block. Every couple weeks, the four of us get together and give the industry insider's perspective on the crypto topics of the day. So quick intro is first. You got Tom the defi maven and master of memes.

[Tom]

Hello, everyone.

[Haseeb]

Next we've got Robert the crypto connoisseur and czar of Superstate. Good morning. Next we've got Tarun the Giga brain and Grand Poobah at Gauntlet.

[Robert]

Yo.

[Haseeb]

And finally, I'm Haseeb, the head hype man at Dragonfly. We're early stage investors in crypto, but I want to caveat that nothing we say here is investment advice, legal advice, or even life advice. Lisa, chopping blocks at XYZ for more disclosures. So, Mr. Czar of Superstate, congratulations. You've got some big news. Superstate just announced that they raise 82 and a half million to quote unquote, bring Wall street on chain led by Bain Capital, Crypto Distributed Global, Galaxy Digital, Han Ventures, Brevin Howard, huge round. Absolutely massive. Talk to us at sofa. Congratulations. Give a little golf clap. Golf clap, there you go. Wait, no, you don't clap yourself. Oh, you don't. In golf, when you do the thing, it's a big. You just present yourself to the world. So yeah, talk to us about the raise, talk to us about what's going on in Super State land. And obviously congratulations from us.

[Robert]

Thank you. Well, Super State last really raised a series A in August of 2023. So it's actually been quite some time since we raised capital. Since then as a company, we were able to prove out the sort of like early primitives of tokenization of securities in an issuer first non wrapper model. We launched two funds today. USTB is a T bill fund. USCC is a high yield basis fund. Those are over a billion dollars of aum. And we've begun to prove out the tokenization of equities Last year we brought Galaxy Digital and sharply Gaming and Forward Industries their shares onto Solana and Ethereum really as a test case to prove out the mechanics of what does it look like to tokenize the security and what does it look like when it's the same stock that trades on the Nasdaq or the New York Stock Exchange as a token on the blockchain. The basic mechanics of this are live in production, but the really cool thing is that you can actually transfer shares between, you know, brokerage accounts and apps on your phone and tokens on a blockchain and back and forth again. And it's an early example of where tokenization is headed, but you know, sets the stage for a lot of things to come. And so, you know, heading into this year we have big ambitions at Superstate. You know, we're going to tokenize a lot of stocks and hopefully we're going to be able to plug them into defi and integrate them into the sort of capital formation process. Because the unique thing about the Superstate approach to tokenization is it's the company's official shares tokenized and so they can do a official company things with those shares, like they can sell them for money, which is the point of being public in the first place. And so, you know, we're going to try to pilot and demonstrate a lot of the use cases of tokenized securities this year and how it differentiates from other models that are out there. And so, you know, this was a raise from some of my favorite investors in crypto to give us more than enough Runway to execute on this. You know what I like about crypto is that unpredictable, it's volatile. You never know whether the good times will never end or you don't know what's around the corner. And so as a founder, my strategy has always been raise less frequently, but raise a lot to always have at least a full cycle of capital in the bank just so you're never backed into a corner and you're always approaching the market from a position of strength.

[Tom]

Have a lot of money. It's a good tip.

[Haseeb]

You have a lot. Yeah, $82 million in the bank. I have to imagine you guys be throwing crazy parties. Now tell us where, what's the theme going to be?

[Robert]

No, you know, I'm not a party.

[Haseeb]

Rent out nightclubs.

[Tarun]

It's a theme theme. We're all going to have to wear bald caps. There's going to be $82 million of bald caps that we have.

[Haseeb]

Like. Yeah, I don't think I'm out at.

[Robert]

The door as a founder. I don't think I've ever thrown a party, frankly. Yes, like yes, I'm sure you never throw a party. Dude, I'm not cool. Like I don't know how to like throw.

[Haseeb]

Okay, you don't like.

[Robert]

What do we do?

[Haseeb]

What's this money for? What are you doing with this money?

[Robert]

Hiring great engineers in New York. If you're an engineer in New York, hit me up that.

[Haseeb]

Very disappointing. Very disappointing.

[Robert]

Okay, listen, I'm not going to do kwan this one, okay?

[Haseeb]

You're not going to do kwan. Okay, fine. So there is a lot of talk these days about tokenization. We talked I think a few weeks ago about the NASDAQ or sorry nyse moving to 247 through tokenized stocks. Of course there's the also story of Robinhood tokenizing these pre IPO companies. I think this is different than what supersede is doing. Right. This is more. I don't know if. Is there a terminology for this? Is that like bootleg tokenization?

[Robert]

There's a lot of different models of tokenization. Yeah. So I mean Robinhood wanted to tokenize private companies. They still do as far as I understand.

[Haseeb]

Yeah. You know like they were third party.

[Robert]

As a third party. Right. The nice and NASDAQ slash DTC plans to be announced. That's really for back end tokenization for settlement.

[Haseeb]

Right.

[Robert]

Right now the stock market settles T&1. They have 2 trillion trades a day. At the end of every day they add them all up, they compare notes and then they send out wire statements. Oh, you have to send this much, you have to receive this much. Tokenization has a huge opportunity there to say okay, can these things settle as soon as they occur and try to offer some capital efficiency there and reduce back office costs. And then there's the crack an X stocks model which is go out, you know, buy 100 stocks and issue tokens against them. Like stock stable coins where it's not exactly the same stock. It doesn't, you know, have the same qip. It's not the same, you know, registration statements. It's not like the same security per se. But it's a great tracking vehicle especially for offshore investors. And so there's a lot of different models being tried this year. You know the big thing that differentiate it's our model is that it's the official stock of the company. It's not a derivative. It's not some back end settlement tool. It's the company stock. And when you combine that with the capital formation process. We think it's going to lead to hopefully, you know, better outcomes for the CFO and the CEO of the company whose stocks we tokenize.

[Haseeb]

You got to come up with names for each of these models. I feel like that's a big thing that feels. Bootleg tokenization. Yeah, yeah, we'll come up with bootleg tokens. Yeah. Okay.

[Tarun]

No, this is TRADFI shit. So it's not going to have a name like that. It's going to have like a six letter acronym for everything.

[Haseeb]

But that's the thing we got to term the names that like people on the street use.

[Robert]

Okay, so here, here's, here's this, here's the cool crypto slang names for these.

[Haseeb]

Yeah. What's the street name? Give us a street. Okay.

[Robert]

Yeah, so we have the bootleg ones, which is the third party, go out and buy stocks and issue a bunch of tokens against it. It's bootleg because there's going to be like 90 competing bootleg standards. There's the back office tokens or the.

[Haseeb]

Boomer Back office tokens.

[Robert]

Back office tokens.

[Haseeb]

Nyse.

[Robert]

This is the NYSE and NASDAQ and dtc.

[Haseeb]

NYSE is a back office token. Okay, what does that mean? What's the difference? Back office token means that they're using.

[Robert]

It to accelerate settlement.

[Haseeb]

Yeah.

[Robert]

It's not about like taking those tokens and doing defi or doing interesting things with it. It's about like how do we process trade still.

[Haseeb]

And they're just a backend detail of their database.

[Robert]

Yeah, that's primarily the goal.

[Tarun]

Right.

[Robert]

Because these are businesses that approach things from the complexities of settlement. You know, the use case really is back office.

[Tarun]

Right.

[Robert]

So back office tokens. And then the superstate and securitized model is really like issuer led tokenization where it's the issuer saying this is my stock. Put it on Solana and Ethereum as a token.

[Haseeb]

Okay. And then XStock is bootleg or what is X stocks?

[Robert]

Yeah, bootleg.

[Haseeb]

X stocks is bootleg. Okay, I see, I see. So there. So, so one of the. Okay, I have one, one question for you. I was very confused because I learned that on X Stocks you don't have to be like so for, for a superstate you have to be KYC in order to get access token because it's security. With xdox, apparently you don't have to be KYC'd to buy an XDOCS token.

[Robert]

You just have to technically haseeb, you're not allowed to buy One, technically you are not allowed.

[Haseeb]

Okay, so Americans are not allowed. Americans are not allowed. But if you're not American, you can just be anything.

[Robert]

Yeah, there's no safety rails on this. I mean like, okay, like we can get into the sort of gray area of the security law of this. Like I'm probably not the best, best spokesperson for it from you know, competitive dynamic reasons. But at some point have like a security lawyer on the show to talk about the different models but they're not available in the US you know whether.

[Haseeb]

Or not that's not actually true. Yeah, yeah. That's actually never true. No, no.

[Tarun]

If we do that, I'm not coming, I'm not coming on the show. I'm going to fall asleep.

[Haseeb]

I'm going to fall asleep.

[Tarun]

I'm just going to like be in the background falling asleep.

[Robert]

Yeah, Tarun doesn't like the lawyer show. We know this.

[Haseeb]

That's true.

[Robert]

You know they're not available in the US like whatever air quotes. But yes, I mean they're permissionless and like there's a lot of questions around can a security be permissionless? Like how? Why? It's the bootleg mod and like honestly I, I truly believe there is a place in the world for the bootleg model. Right. Are U.S. institutions going to be able to use it? Probably not. But like if you live in China, is this a great product? Yes, it's a great product.

[Haseeb]

Okay. For all our Chinese listeners. There you go, Robert. Recommending the bootlegged tokens. Okay.

[Tarun]

Actually we should talk about the other bootleg tokens which did really well.

[Tom]

Flawed you mean.

[Tarun]

Trademark is 4.5.

[Tom]

People are just chewing through the tokens.

[Haseeb]

We got a lot of, we got a lot of work through this week. Okay.

[Tarun]

No, no, no, I was just saying a natural topic to talk about is like the Trade XYZ billion dollar plus volume days like today, yesterday, et cetera. Because that, that's like the next extension of this.

[Haseeb]

Okay. Very, very different. But yes, we should talk about it. So one of the big stories this week has been the rise of these HIP3 hyper liquid markets, particularly through an application called Trade XYZ. So this is for those of you who don't understand what anything I just said. So Hyper Liquid, of course the largest perp Dex that's on chain today, hyper liquid, you know, by, by open interest as well as increasingly now by trading volume hyperliquid. They have these markets that other people can create on top of hyperliquid that use the hyper liquid engine in order to execute trades and also do, you know, basically do state management, all the other stuff that hyperliquid does. So the One of these HIP3 markets is Trade XYZ which focuses on real world assets. So these are, you know, stock indices, individual name stocks as well as commodities. They recently launched a number of commodities products. And for those of you been following the news, commodities have been ripping, particularly precious metals. So gold just hit an all time high. I think it's at like 5100. Silver has been super volatile going up and down alongside gold. And HIP3 in terms of the overall demand for HIP3 markets reached an open interest all time high of 800 million driven by this surge in commodities trading. So apparently the just the total amount of perps that are now traded by XYZ, 10 of sorry that are traded by trade XYZ on hyperliquid. 8 of the top 10 markets are hip 3 perps now, basically meaning they're real world asset perps and just trade XYZ alone is now the 7th largest perp Dex by open interest. So big change that we've seen in the overall dynamics in the market of what people are trading now. Some people are attributing this to well, crypto is kind of in the doldrums, volatility is low and so maybe people are just bored and metals are ripping and so everyone's trading metals right now. Another version of the story is that, well this is just a better substrate in period to trade these assets for anyone. So it's not maybe just the crypto native story that crypto natives are getting bored, but rather that this actually is a better way to trade metals or commodities or equities or whatever. So quick reactions from the crew here. What do you guys think about the rise here of the HIP3 markets and the metals trading that we're seeing on hyper liquid?

[Tarun]

I think one interesting thing was that prior to this current rally you were starting to see some liquidity fragmentation in HIP3 where Kinetic launched their stock tokens. And then if you look on, if you end the Felix and stuff and if you looked on Twitter, there are tons of memes of like I searched for Tesla and I got four different Teslas. Which one do I trade? Then this current boom completely swung everything back market share wise to trade out XYZ like it's like over 90% now. And I think that's actually kind of an interesting development because I do think there have always been these kind of questions of like, is this market structure good for equities to have like 5 million venues that people might fragment over, use different stablecoins, etc. And it does seem like it's clear that there is UX and order flow benefits that seem to be compounding for certain exchanges over others which obviously for trade clearly it's crushing it. I mean if you compare that volume to the real spot volume, it's a meaningful percentage at this point.

[Haseeb]

Do you know what percentage it is?

[Tarun]

Well, I think spot silver was like all time highest was like 30 billion. I don't know, I didn't look at futures. But so from that even that is like again it's not like to like because it's the future of spot but it's not like out of the ballpark, it's like same order of magnitude, like within the order of magnitude. So clearly like that is working. I think the other thing that's interesting is it does seem like a lot of the demand has been. You can kind of guess this from. Get some data on this from the builder co side but does seem to be coming from the mobile traders. So that the mobile usage versus non mobile usage I think is the best in my mind and it's not at all a perfect proxy, but it's like sort of the best proxy you can have for existing crypto users versus new crypto users. I feel like existing crypto users are not really like using mobile trading apps anywhere near as much as like new Net new users who are like mobile first always. So in some ways it does feel like.

[Haseeb]

I don't know if that's. I don't know if that. I don't know if you can make that inference. I think you can make the inference that it's more Asian traders because. But that's always been true crypto is that in Asia they're mobile dominant.

[Tarun]

Yeah. Like I agree mobile's not like the ultimate signal, but it's very clear. It's like different than the crypto side, the pure crypto side, like the ratio of mobile to non mobile. And I think there's something to that that must probably indicate something.

[Haseeb]

Yeah.

[Tarun]

Again the proxy for mobile is sort of depends on how you want to measure the order flow stuff. So it's not like perfect but it is quite different. So to me that was an interesting observation about the gold stuff and gold and silver stuff.

[Haseeb]

Dom, what's your take?

[Tom]

I was actually just talking with someone today who was kind of asking what's going on with this? First of all, I will say it is slightly early in the year, but I will say this was One of my 2026 predictions was that we would have per RWA volume surpass spot for one asset. And so who knows, maybe we're going to get there one of these days. But we're getting closer, Tom.

[Robert]

We're getting closer.

[Haseeb]

I think you said for an equity. I think you said for an equity.

[Tom]

Okay, close enough. Close enough.

[Tarun]

Yeah, I think, I think, I think I'm willing to take silver versus the silver ETF or something. I, I give Tom the credit.

[Tom]

All right, thank you.

[Haseeb]

To be clear, my prediction was that RWA purpose would hit 20 of total D5 volumes. I'd say. I think I'm. I'm looking pretty good right now.

[Tom]

That's pretty good. We can both win.

[Haseeb]

Okay.

[Tom]

I, I think it is, it is like a push and a pull, right? On the one hand, yes. I feel like I've been. That has been one of the themes of the past. Let's say six months has been all these crypto people actually transitioning to trading other assets, including commodities. And there's just been doldrums. They're not new assets people want to trade. And then obviously it is great timing with hip3 launching and obviously other perp Dex's RWA perp Dex is launching or perp Dex is adding RWA perps. So it felt like this was something that couldn't have happened a year and a half ago and wouldn't have happened a year and a half ago because there just wasn't enough pull for it. So it feels like there's kind of this perfect storm right now. So I am more curious, like you said, to kind of see where this flow has been coming from. I was seeing some sort of data floating around earlier that a lot of the volume is concentrated during Asia hours, which might kind of speak to this Asia user base, but really this has been kind of blowing up in the past two weeks. So there's still a lot to kind of comb through.

[Tarun]

Yeah, I mean, the other thing that's interesting is the single name equity volumes haven't really gone up. It's mainly the index, like the NASDAQ approximate index and the commodities that are just blowing up. So I also do think there's something interesting about that. Like, you know, crypto kind of trades more like commodities. Maybe that's a, that's, that's. We can go through a bunch of these different signals and be like, is it crypto native or not? And I bet you you would still have a lot of variance in that.

[Haseeb]

Well, it is, it is very crypto to be focused on commodities and particularly gold and silver. Right. Because of course, there's been a long standing connection, not just between bitcoin and gold, but the gold bugs and the people who are worried about the debasement trade and all this sort of stuff. So it all very much flows together. So I would guess that actually you're likely to see that a lot of volatility in the NASDAQ does not necessarily result in increased trading. Whereas this broader macro stuff happening with precious metals and other like copper and things like this. For whatever reason, I see the crypto people are more engaged with these particular markets. And I think it's idiosyncratic to crypto. And I think that's a sign that this is not just people moving away from the CFD platforms where mostly people are trading equities. This is really the crypto natives extending into commodities trading. That's what this smells like to me.

[Robert]

Well, the crypto natives, I think are already hardwired to go wherever the volatility is. And the excitement is. And for a very long time, crypto was the most volatile and the most exciting. And this is a moment where like, you know, precious metals are the most exciting thing. People are like levering up copper and silver and going crazy. And it's like so much more fun than crypto right now. Right. I'm in a lot of group chats. Goes up, yeah, it's going up in a parabolic move. And so, you know, everyone's chasing it. And I have all these group chats where people are like, I'm not trading crypto right now. I'm trading silver. And I'm like, what?

[Tom]

I think the PM thing is very, very. I mean, I was talking like a cab driver the other day who was like trading silver. And I was just looking today, they're no longer trading. The most traded ETF in the US today by volume was the ultra short silver ETF by. By a few fold. And like all the other ones are like either semiconductor or precious metal. And so it is just like high volume, a lot of interest. There's a narrative around it. And so I think that just gets. Crypto is not isolated in that way.

[Tarun]

Yeah, I mean, another, an interesting fact about this whole metals thing is like how everyone is just so negative on bitcoin from a macro. It's like it's lost its macro shine. And it's like, I feel unclear what will fix bitcoin.

[Haseeb]

Bitcoin has not had a macro shine for like three years. All I hear is people complaining like oh, it's not behaving like gold. It's behaving like a risk asset. Now it's not behaving like a risk asset. It was acting like the nasdaq. Now it's not acting like the nasdaq. I feel like the only time when people are happy with bitcoin is when it goes up. If it doesn't go up, it's never quite behaving the way that people want it to. And I don't know that it will. I feel like people will just keep repeating the same thing, like, why is it acting like a precious metal? And it's like, dude, it wasn't acting like a precious metal at all for a year. When was it like a precious metal?

[Tarun]

To Robert's point, though, like a lot of people who would never have traded gold and always traded bitcoin, at least anecdotally. My experience also is that they're all only talking about. They've completely forgotten about crypto or, like, not forgotten, but they're just like that toy. They've, you know, the toy story meme. They just like, throw away the doll.

[Tom]

It feels, I don't want to play with you anymore.

[Tarun]

Yeah, exactly. It feels like that a little bit.

[Haseeb]

So one thing, one thing that this leads me to think about is like, how to think about crypto traders as crypto matures. So I've had in my mind for a long time that crypto traders are like a distinct category of people and they trade crypto. That's like their thing. Their thing is that they trade crypto and they're distinct from, like the wallstreetbets guys. They're distinct from the Robinhood YOLO ers, They're distinct from the options punters. They're just like a distinct group of people. And what may happen as the industry matures is that we start to stop thinking about them that way. But instead, it's almost like a generation, and this generation, they got into crypto trading at a particular time when crypto trading was particularly volatile and there was some alpha to be had in front running retail investors. And as crypto becomes more stable and more institutional and there's less turnover in the top 20 assets, that maybe these people just splinter. And we start to think of them more as, like, I don't know, now we talk about zoomers, right? As like a shorthand for Gen Z, but really the idea was that, oh, these are all people who took zoom school and it was like a behavior that tied them together as opposed to a distinct generation. Maybe that's the way that we will think about crypto traders in five to 10 years, is that, you know, they're like. I think about this because I think about like poker players. I used to be a professional poker player and there was a generation of people who played online poker around, you know, a 10 year period between, you know, 2010 and 2020 or like 2017. And those people, they don't necessarily have that much in common anymore, right? They all kind of went into different things, but they all had this period in their life where they did, where they played online poker. That might be the way that we think about crypto trading in five to 10 years. So I'm curious how that lands for you guys.

[Robert]

I mean, yes, like people were attracted to crypto for two reasons, right? One is long term belief and one is speculation. And I think the long term belief piece is still there, but on a day to day basis. If you, if you're a believer in Bitcoin, if you're a believer in Ethereum, if you're a believer in Solana, there's not that much for you to do. You're not trading actively. You're hodling, right, Per se. You're not a daily active participant. If you're there for speculation, which has always been the history of this. Yes, you were a daily active trader, you were trading on Binance, you were trading on Dexs, you were trading on Coinbase, you were trading on everything you could get your hands on, perps, whatever. And I think the speculative people are not speculative, only around crypto. They're not loyal to one asset to trade. They're loyal to the game and they'll go where the action is, just like poker. And if that evolves, they'll go there too. You know, people are not, you know, on the speculation side and the trading side, they're not like maximalist for crypto. They're there because it was a great volatile asset with lots of things happening at all times with rumors and with hacks and with launches and with new memes and new metas and it was the best place to trade, period. Like, I can't fathom, you know, a market and an asset class that was more fun than crypto for that very extended period of time. But just like poker, the meta evolves a little bit and I think those people might be moving on as crypto grows up a bit and it's no longer about relentless, ruthless speculation. It's more about there are certain things that are working. It's growing up, we're finding the directions that are hardening and working, and it's no longer as speculative. And so it makes sense to me that they would move on a bit.

[Haseeb]

Because if we believe that that sounds very bearish exchanges, that's bearish exchanges.

[Robert]

They all become the everything exchange, unless they all go into metals.

[Haseeb]

So maybe that's a better way to understand this market. They're trying to hold on to the crypto trading cohort and keep them all within the same app experience.

[Robert]

Right.

[Haseeb]

While they're turning away from just crypto trading. Right.

[Robert]

If you use Coinbase as the sort of, you know, example of this, they have a crypto user base, and now they're saying very vocally, you can trade stocks, you can trade metals, you can trade prediction markets, you can trade all the new things you would want to trade from the user experience you're used to with the account history that you don't have to export if you just stay here.

[Tom]

I think that's right. I think I've seen some data around the Robinhood user base cohort over time, and basically we think of Robinhood users being very young and wanting to trade a lot. But actually, if you look at it, it was this base of millennials to late zoomers that got swallowed into Robinhood. And they're kind of slowly aging through the system like a lump. But it's not like there's this huge pipeline of new people where people, like, age out of Robinhood. They're just like. And you see that in the products, like, I think, what, a year ago now it's like, you can get, like, a better mortgage through Robinhood. And like, they're just sort of. You can. If you bring your IRA over to Robinhood, we'll give you a bonus. Like, they're just tracking with this cohort as they get older and older. And like, I think, you know, crypto exchanges are probably going through something similar. I think even within crypto, there are different sub assets that get really popular, you know, in a moment. And I in some ways also just feel aged out of those. I used to trade NFTs back when NFTs were popular. And then I feel like Meme Coins came around, and I definitely was not trading Meme Coins nearly as much as I was trading NFTs. And I think there's always going to be that kind of component. And maybe the answer is, hey, maybe some of these HP3 RWA markets are kind of the thing people want to trade now. I don't quite know what the next thing is going to Be. But that is maybe more. My model is like, it's a substrate.

[Haseeb]

It's a sign that the cohort is growing up. And in order to sell to older people, you sell them metals instead of trading meme coins.

[Tom]

Yeah. Soon they're going to have a target date fund per ETFs or perps. And so it's almost all the most boring assets.

[Robert]

Can you imagine a target date retirement perf?

[Tom]

List it. List it.

[Haseeb]

All right.

[Tarun]

I would make May maybe I. I don't disagree with what you're saying, but I would add something different, which is the history of financial technology firms does show the tom thesis of like, usually they get one cohort, but then they like slowly start aging out or like they don't get like the new. It's a little like social media. Right. It's like the reason there's always a new one is like it's hard to keep the 17 to 20 year olds interested in your thing where like everyone looks older than them and seems kind of shitty or whatever. And I kind of think financial stuff has the same thing, except a longer half life. Instead of it being like three years, it's like 10 years where these cohorts. But that's also kind of the opportunity from an investment standpoint of like, what's the new interface for like 17 to 25 year olds? It's probably not like a Robinhood Coinbase thing. It's probably like some some sort of agent. Honestly, is Claude bots closer to probably something that someone 17 would not roll their eyes at, Whereas like something that shows you confetti when you buy a stock probably is like, oh, that's so millennial trash. Like, I don't want to use that.

[Haseeb]

Cloud.

[Tarun]

Yeah, there's like, there's like some new interface that hasn't probably hasn't been discovered, but is like that will be the everything exchange for everyone under 25. Not. Not none of these ex month. That's like my personal feeling. Gut feeling.

[Haseeb]

Okay, so that's a good transition. Let's talk about claudebot, which is taking the Internet by storm. So so far you have maybe noticed if you've been on the show, we have not talked very much about what's happening in AI land. That's been a very intentional choice. But we now can no longer remain silent. Unfortunately, we also have to now talk about AI and we also have to talk about Claud Code. So those who've been living under rock. Claude code is basically taken the Internet by storm. Particularly since December time period when I think Everybody on break started trying out the new set of agentic tools launched by Anthropic, particularly with Opus 4.5, which is their newest model. And Claude has basically broken everybody's brain about what is now possible with what's called Vibe coding, which is basically allowing an AI agent to do all your coding for you with less and less human intervention than what previously these tools demanded. So you now have people believing that essentially the job of software engineering is permanently changed and that perhaps this is the threshold by which it's no longer an intellectual question. It's now demonstrably true that engineers are just writing fewer and fewer code and more and more non technical people can basically just let it rip with respect to coding. And this has really culminated in a lot of hype around one particular bot called Claudbot. Okay, now this is Claude C L A W D as opposed to C L A U D E which is how the Anthropic model is named.

[Robert]

Well now that name is also outdated because they had to change the name today.

[Haseeb]

Yes, they just changed their names. Exactly. So Claude, which is the product launch by Anthropic formerly had asked. No, sorry, actually still Claude real Claude asked what's now formally Claude asked formerly Claude Bottom to change their name because of the obvious name collision and confusion. So it's now called Maltbot. It's themed after lobsters and of course lobsters malt their shells. So I guess that's the connection there. And so anyway, what does claudebot do? Why is it such a big deal? So claudebot is basically a bot that you can install on any device or on a virtual server. And essentially what it does is it can take control of your machine. It has complete, you know, access over everything on a particular device. It can connect up to your email, your, you know, your calendar, your telegram, your discord. It can be accessible via WhatsApp or just any way you want to contact it. And basically you can automate almost anything and it uses some of these very high end models underneath the hood in order to make decisions and people are talking about automating their entire lives. They're talking about, okay, my email inbox is now totally run by Claude. They're talking about it's booking flights for me, it's doing this for me, it's doing that for me. Now, all that being said, enormous amount of hype around this product. There's also clearly a lot of overrepresentation of what it can actually do. There's also a lot of very non technical people doing very Stupid things with this bot, as well as there were some recent reports of thousands of claudebot gateways that are just open on the publicly accessible Internet with zero authentication. It's very easy to prompt, inject these things, to steal keys to go in and just take control over someone's Gmail, because it just has complete write access to everything in their life. So want to caveat for anybody who's listening? Be very careful with using these tools because these are not generally recommended for the public to just give control over their entire lives, these things. And they're not that good yet. It's still kind of Twitter demo vibes, I think, for a lot of these things.

[Tarun]

Well, it reminds me of like, the. What was the bot that you could send Bitcoin to that everyone that. Like Marc Andreessen.

[Robert]

Yeah, the Mark 1.

[Tarun]

Yeah, the Mark Andreessen.

[Tom]

Oh, Zerebro.

[Tarun]

No, no, way before that. Way before that, this was like.

[Haseeb]

Yeah, the goats.

[Tarun]

The goat C1.

[Tom]

Yeah, yeah.

[Tarun]

True terminal. This, this. True terminal. This, this, this. Fear the current moment with Maltbot. Wow. It's first. I've.

[Haseeb]

Well, so this is broader than just crypto. This is like.

[Tarun]

Yeah, I know, I know, I know.

[Haseeb]

Tech phenomenon, but for sure. But crypto has. It's.

[Tarun]

Yeah.

[Haseeb]

Yes. So crypto has decided to get involved in the way that crypto knows how, which is to just launch tokens about claudebot. So this has now become an increasing trend, crypto meme coins. Yeah, so they keep launching meme coins named after AI SDKs. And they do the thing where they basically say, oh, this is like the official one because we're donating the, you know, 20% of the token supply to the devs, and if it takes, then oftentimes this meme coin will run up. So apparently the founder of this protocol has just been harassed by all these various meme coins that are throwing things at him and saying, oh, this is your. This is the official cloud bot meme coin. And so, yeah, naturally we have once again ruined a perfectly good thing because he's now said that the reason I.

[Tarun]

Brought up the truth terminal thing was like there was another AI tool which was like a kind of like proto Claude bot called Gastown, where it would like manage quad sessions for you. And then people launched a meme coin for it. And then that guy had the same experience that the truth terminal guy did when the meme coin happened. You know, like the up, the up and down from like someone who's in tech, who's never touched crypto, who suddenly was like, I made X million dollars. I have zero. And I thought it was just like this is like. I feel like we're reliving that that era again. Like that's what.

[Tom]

Well, there's also the Ralph Wiggum Luke guy, who's another sort of AI agentic library author who. There was also a Ralph Token, but then he endorsed it and he was having fun in public and he's like, ralph Token's going up and check this out. So I don't blame people for thinking that maybe the Multbot guy would also want to celebrate his token. I mean, the whole thing is kind of silly, I guess, but yeah, I think we've been talking about just being surprised by the level of excitement around Multbot, given that, hey, a lot of these things you can already do and have been able to do with existing mainstream models. So what is it about Moltbot that has gotten people excited? And I think there's maybe a few different things. I think one is a very simple thesis. It is just skeuomorphic with how maybe people think about computers. It's like it's a box that lives somewhere and it can run things for me versus it lives in some amorphous server somewhere and I don't know what it's going to do. And it has my data, which obviously doesn't really fit with the model. But it is less scary to think that if you don't like it, you can unplug it and it's not like someone is going to go steal it from a server or someone else has your data. 2 I think very, very simple thing is obviously in the US especially people or on Twitter, Apple products are extremely popular. And Apple products have historically been very difficult for agents to access because they don't really have nice accessible APIs like automating iMessage stuff or Apple Notes stuff is just really kind of a pain in the ass. And so hey, if you can run this thing locally on your computer, of course it can just go and automate some of these things. And so I think that is also probably something that sounds, I mean it's classically like the blue bubble kind of premium where it sounds so stupid, but it is something that people really actually I think value and like. And so I think that is probably another component of it. But maybe people are just bored and want something to tinker with. I don't know, it is maybe a bit confusing still.

[Robert]

Is anyone here running cloudbot Malt?

[Tarun]

I ran it on a laptop.

[Haseeb]

Yeah, yeah, yeah, I tried over the week. I Didn't do it locally. I did it on a vps, but I found it completely underwhelming. Yeah, there's a few blog posts that talked about this that, like, there's so much. I mean, this is classic in AI. Is that like, okay, there's a Twitter demo of like, some AI influencer who talks about, oh, I, like, built a. I have now 20 employees of Cloudbots that are running on my desk and they're solving all these great things. And it's like, okay, what are they doing exactly? Writing your AI newsletter? You know, like, what amazing business things are they solving for you because you're like an AI influencer? So there is a. Here's what I would argue is the reason why claudebot is taking off. Is that Claude code, which I don't think we've even talked about on the show. So maybe it's worth also just doing a little bit about Claude code and AI and how it interacts with crypto. But broadly, Claude code people realize that it's a really big deal, but I think it still feels like a software engineer thing. And I think for most people, even if they've heard this is a big deal, they're like, well, that's not really. I have to open a terminal and learn how to go to a directory in order to turn on Claude code. And this is all very scary and confusing and maybe someday it'll be so easy that I can do it too. But claudebot is like, oh, you install a bot and connect your email and talk to it on iMessage. And it's like, that does sound like something that I'm supposed to be doing. And now, that being said, I think when people actually try it, I think they will find that, like, okay, this took hours to set up and it's like, super finicky. And okay, now it's kind of like I'm talking to a model out of the box on iMessage. That's not really that much better than talking to it over just my ChatGPT app. So I think there is a little bit of, you know, and once you get there and you're like, okay, do I really want my AI answering emails for me, I'd say if you are, you probably have not done a very good job managing your email. If you want an AI to answer emails for you, either one, you don't have very valuable emails, or two, your email people are talking about. Because of Quadbot, I was able to process 10,000 emails and I'm like, wait, why do you have 10,000 emails to process. You fucked up something way earlier. If this is the solution to the problem that you have.

[Tarun]

On the other hand there's like examples where it's like Joe Wiesenthal like from Bloomberg wrote this like sentiment analysis thing and like he's clear he was not a developer and I think there's been.

[Haseeb]

Like examples where it's like that's Claude code.

[Tom]

It is Claude code.

[Tarun]

That's just Claude code.

[Robert]

My point is UX of malt makes it easy enough for him to finally use Claude code.

[Tom]

No, this was, this was pre molt this. He, he used Claude code.

[Tarun]

No, no, no. Sorry, sorry, sorry. I'm just saying like we went from that last week to this, this week. That's like the, the speed of the shift, that sort of.

[Haseeb]

What, what I would claim though is that like the stories about cloudbot are going a lot wider than the actual usage of cloudbot. Obviously. I mean there are, to be clear, there's a huge amount of use of cloudbot but I think the under reported part of it is just how, how few people are actually getting that much value out of it right now because it's actually just, you know, I mean maybe some people are like yes, automate my life. And for a lot of people I think it's just I want to Talk to my AI on WhatsApp and that's cool, that's totally cool. That's a reasonable thing to want. Very easy to just build that yourself in like 10 minutes on, you know, telegram. Like I've literally done that. I did that like months ago. It's just I have a proxy to Claude on a telegram bot. It takes honestly like 10 minutes to set that up.

[Tarun]

I think the more funny thing to me is like all the people who are like, oh yes, I'm like genius poly market trader now because I made claudebot trade for me and it's always like the first, first, first hour is positive P L and then it's the bloodbath.

[Haseeb]

That's what convinces you to load up more money into the.

[Tom]

Yeah, yeah, yeah.

[Tarun]

It definitely, there's definitely this, this like casino like thing where they give you the, the, the joy up front before the loss later with these things because everyone who's doing this is crowding at the same time so it does push the price up. So there's definitely some like hurting behavior like meme coins going on also that I've observed. Like that's like a, a meta thing about this phenomenon. But I think it's. This is the first of many of this type of thing this year.

[Haseeb]

And this is the thing that I've always said in the past is that as AI, so you know, last year we had the rise of the chatbots and as a result crypto created a bunch of chatbot meme coins and we got obsessed with virtuals and all this other stuff and that kind of died off when the novelty wore off. Now we've got claudebot, slash Maltbot and we've got all these other things and that's creating this new set of novelty about Ralph Wiggum loops and all this other stuff that people are like, great, let's create a meme coin around that and let's tell some story about how it's going to interact with crypto. I think this will keep happening as we get new capabilities. Crypto will find some way to get excited about some narrative and slap a token on it and find a way that people can trade it or even trade the sentiment around it.

[Tarun]

Well, I think the prediction market one seems more popular than the meme coin, which is a little different than the last AI trading bot. That I think is an active difference this time.

[Haseeb]

That may be true. That may be true. But I want to shift the conversation a little bit to just talking about the rise of agent decoding and how that is going to affect crypto at large. So first thing I'll ask just the panel is, are you guys using Claude code slash agentic coding tools like beyond like IDE based ones? And second question is, what's happening within your pure terminal base? Pure terminal base, where it's like you're no longer looking at the code.

[Robert]

I am not.

[Haseeb]

And what are your companies doing? Why don't you start? You just raised 80 million. Did you buy cloud code subscriptions for everyone on the team?

[Robert]

Yeah, we love cloud code at Superstate. Am I personally doing it? No, because I'm not shipping PRs and merging them in. But are other people. Absolutely.

[Haseeb]

Have you done any. Have you done any agentic anything not coding?

[Robert]

I've not vibe coded anything. I'm like the late Cod.

[Haseeb]

Anything like vive coding. Okay, got it.

[Robert]

I'm not a vibe coding.

[Haseeb]

So the people within, the people within Superstate, how many of them are writing code by hand versus entirely now on agent decoding?

[Robert]

I don't think anything's entirely agentic.

[Haseeb]

Right.

[Robert]

I think like, and I don't think anything is entirely by hand. Right. I think like, we're at the point where, you know, there's a lot of things where it's like the documentation is all clot, right? You know, you do like, you do a PR and it's like oh, write the coverage and like write the like documentation and it goes into like a PR and it's like, oh, this is a beautiful pr. This makes a lot of sense. Like this is way better than if you took the time to like type up a bad pr, bad documentation and bad coverage. You know, it's like Tarun, what does.

[Haseeb]

It look like at Gauntlet?

[Tarun]

I think so it's something like 95% of people have at least one LLM subscription where like 40% anthropic, probably 30% cursor. So like multi LLM and then the rest codecs. Basically we have a lot of internal Claude skills. So skills are just like pre made definitions of like how to use the repo, etc. Obviously all the CI check in stuff, but been doing that for like a year. And then the public stuff. We have a lot of documentations for how to use our vaults SDK for like how to like you want to build a strategy that borrows from our vaults, like how do you do that? And we are will have some like MCP server and skills for that soon for both Codex and Claude. So we have internal stuff a lot and external stuff soon for other people who are building on top of these. So almost everyone, basically almost everyone including myself, like including myself. I'm like literally looking at a cloud code terminal every time I look to the left on this podcast. Just get off the terminal, Tarun. Yeah, yeah, yeah. No, no, I have like two more screens over here because he has an.

[Tom]

Army of Claude code bots running a gas town.

[Tarun]

1, 2, 3, 4, 5, 6, 7.

[Haseeb]

Yeah, anytime he needs to comment on a story and he's like, Claude, come up with something smart for me to say. You kick off a bunch of stuff.

[Tarun]

No, no, no. It's bad at that type of stuff. Actually I will say I use ChatGPT a lot because I think so I use Gemini for like live research. Like if I'm asking something about companies or real people, it's like a million times better CHAT GPT for any math or research thing. It completely crushes like anthropic. Clearly did not train, did not pay any Merkur people to win IMO prizes until it's like not very good at certain math problems. Whereas like Gemini and ChatGPT are really good. But Gemini or Chat GPT is like you ask an unsolved research question. I mean that's the reason people have been really talking about the solving erdos prizes with GPT5 direct without formal verification kind of stuff. Because it is just like fundamentally few orders of magnitude better than the rest research for things that haven't been solved. And then Claude is obviously the king for long running programming. You can give it a task, it works for an hour. No one is coming that close, close to that. Codex is actually pretty good. But it like you have to give it very clear instructions if you give it an abstract task. I think that's why Claude code became mimetic, because it's very good at like kind of fix. You give it an under specified task like make me a dashboard to like show me how many times Haseeb says fuck on the chopping block. Like that's like an abstract task. You could, you could give Tarun.

[Robert]

You just sweared in trying to talk.

[Tarun]

About swearing, you know, that's, that's the game to play here.

[Haseeb]

Yeah, that's fair. That's totally fair. Why would you, why would you click on it for that?

[Tarun]

And so like let's say I give that abstract task, right? Like actually realistically you'd have to do it in Gemini because it has YouTube access. But let's pretend Claude had YouTube access. Claude could take that abstract task and run on its own. And it'll make mistakes on the way you don't see them. It's just running in the background. Whereas Codex would like ask you 500 times for like how to correct, you know, like are you sure you mean this Haseeb, not this other Haseeb? Like so there's there is like a lot of like nuance to these things which you only get from playing around with them a lot. Anyway, sorry, that was a long rant.

[Haseeb]

Yeah, yeah, no, that's good. So we, so at Dragonfly, I think it was last year we bought everybody, everybody at Dragonfly, including like non technical people, back office people, we got them cursor subscriptions. And I remember we did this like big rigmarole of like oh, coding is now democratized and anybody can code. And we like had some demos that we did across the org and very naive because almost nobody used it. Everybody was, everybody tried it for like 10 minutes and they were just like, I don't know, I don't know what the fuck I'm doing. I don't know how to do anything.

[Robert]

Tom, did you use it?

[Tom]

Clearly I have a personal sub, so I wasn't really part of the sub.

[Robert]

Do you use it after buying everyone's subscriptions?

[Haseeb]

I was using. Yeah, I used Cursor to like build some internal tools. Yeah, I was the one who like advocated to do this and I think I was a little bit. I gave this big speech about how, oh, the software is going to go, it's going to be so cheap, it's going to be too cheap to meter and everybody will have their own personal software and blah blah, blah. And so I was like, yeah, we should all try this. And there was about, we have about 45 people at Dragonfly and there was about six people who really use it and they were non technical but they really dove in and they really got excited and they found ways to build something valuable for themselves. But the reality is that Cursor a year ago was still like you needed to be a software engineer or you needed to have very good intuitions about software to solve problems to prevent it from getting into loops to debug it. It still required a lot of mental labor to get the right thing to happen. Then about a month ago or about three, four weeks ago, we killed everyone's cursor subs and we got everybody cloud code subs. And I really do think that now the speech that I gave a year ago is now correct, which is that it actually does feel true now that you can just tell Claude a verbal description of what you want to build and it will just build it. Now that being said, there's still some massaging. You still need to build your intuition about what the model can do and can't do. But it does feel like a game changer with respect to totally non technical people being able to solve these kind of software problems for themselves.

[Tarun]

This gets back to my point of it takes very underspecified prompts versus a lot of the other coding agents really require quite specific prompts. And the real question is, is that because they paid Merkor more for really wide ranging workflows or something, or is it a real training advancement? And obviously no one other than some very tiny subset of anthropic employees probably know the real answer to that.

[Haseeb]

Yeah, I mean, clearly they put an enormous amount of work into the harness of quad code. And I was thinking about this the other day, is that there may be just a runaway effect where you can detect that you're being used in the quad code harness and make the model better when it's being used by quad code as opposed to when it's just being invoked from anywhere, such that if you use Claude in cursor, maybe actually you get significantly worse performance because the model can see the format of the Claude code harness and say you're already.

[Tarun]

Seeing that divergence if you can actually do the experiment between the two, which is I think why Cursor went to go make its own model and stuff. Because I think this is how the model creators are locking them out.

[Haseeb]

Right, right, right, right. Which makes perfect sense, to be clear. I mean, that's exactly what they should be doing from a business perspective. But the interesting question now. So I was chatting with somebody at Anthropic a few weeks ago and it was the guy who actually wrote the paper that I think we talked about on the show about smart contract hacks that Anthropic was able to create a smart contract hacking benchmark and get Claude to hack a bunch of smart contracts. And he was telling me that his team at Anthropic is entirely on Claude code. They don't use IDs anymore. Basically. They're not writing code by hand, they're code reviewing. They're reading diffs when they're actually merging stuff, but they're not writing code by hand anymore. That is not happening Anthropic. I was chatting with some other folks I know who work at some other tech companies and that's not happening everywhere. There are some places where people are you guys just talked about. Okay, yeah, there's still people using Cursor and writing code by hand and ides or some continuum between autocomplete and full agent decoding. But it does feel like Anthropic and a certain number of other companies, especially brand new startups, they're kind of living in the future. This feels like this is how the future of software engineering is going to look is that you are not writing code by hand anymore. If you're still writing code by hand, you are kind of holding on to this like previous generation of coding. Unless you're really far out of domain, you're doing something that these models cannot do. Yeah.

[Tarun]

Carpathi tweet right before Christmas that like exemplifies this exact thing, which is like you have to basically unlearn. If you were a software engineer in the last 10 years, you basically have to unlearn everything you learned in some ways to use these things properly.

[Haseeb]

And it's a total devaluation of the skills that previously mattered as a software engineer.

[Tarun]

Yeah, for sure.

[Haseeb]

Knowing libraries deeply. Yeah, like knowing all.

[Tarun]

Everyone is a manager, engineering manager, no one is an ic.

[Haseeb]

Exactly.

[Tarun]

That's like what's happening, right?

[Haseeb]

Exactly. And again, in certain domains that's not quite true yet. So if you're writing VDHL or you're writing. Or even for some of these esoteric.

[Tarun]

Languages, smart contract stuff I think is still a little like that.

[Haseeb]

Yeah, exactly. Smart contract stuff. There's not enough training data, It's a little bit out of domain. But if you're writing web dev, if you're doing Python, if you're doing Typescript.

[Tarun]

No, even performance, even performance stuff is pretty good. Like Rust in C are way better than.

[Haseeb]

Right, right, right. Systems level languages have improved a lot over the last few years. So everything is changing with respect to software. And that really matters for crypto because software is the primary input into our industry. It's the primary cost driver for every company, every startup. Right. I mean, Robert, why you have all this money, hire engineers. It does feel like things. Yes, exactly. It does feel like things are going to change. Everything is going to get a lot cheaper. All these products are going to become built a lot faster and the future software startup is going to look different where they spend their money and how long it takes them to build products. And the iteration cycles are going to change. But there's sort of the before agentic coding startups which are going to be much slower to adapt, and the after agentic coding startups that are going to be built from the ground up for this new world. And so I don't totally know how to think about it yet as an investor of what to look for or even what the questions are to ask. You know, I don't know how you guys are thinking about this, but it clearly is going to be a huge difference in the startups that are hiring people based on this and know what to look for and know how to architect their engineering works around agent decoding. And the ones that don't. I think you can see even hiring a cto. Who do you hire as your cto probably really matters because if you hire a CTO who's like a great CTO for two years ago might be the wrong CTO for a startup starting today.

[Tarun]

Yeah, I basically think you A, have to ask people what their plans are, B, inspect their code base with the LLM and see if it actually understands it well, because honestly that's like, that's the best test of any open source code base is like figuring out the model's interpretation of the code base is fast. If it is, you can tell it was created agentically also. I got to run, but I.

[Haseeb]

Okay, all right, we gotta wrap. We're up on time, but I'm sure we'll have more to talk about. All these topics. Meantime, we'll see you all next week, and happy trading.