You've never seen a company grow 10x in GAAP revenue and run rate year on year for 3 years. So you're leaning into the singularity here. Now it's worth pointing out. It is my favorite show of the week. Rory O'Driscoll, Jason Lemkin analyzing the biggest news in tech. Anthropic raises $30 billion. OpenClaw creator joins OpenAI. Thrive closes on a mega $10 billion fund. Wall Street has decided this is the bet they want to make. Capital has decided this is the bet they want to make. People are going to try and make this bet. Wall Street fell in love with AI and to do that had to fall out of love with SaaS, right? Who wants to get back to early stage? I want to do the Anthropic round. Give me a break. I don't want to like have to pick which accounting software in four years might break out for AI. No one wants to do that. Just show me the Gary. Ready to go.
Boys, it is so good to be back. Now, we might as well rename this show for the first segment at least this week in Anthropic because my god, the news announcements that came out. I want to start with the fund raise itself. Anthropic raises 30 billion at a $380 billion post money. Now, it was originally 10 billion and it upscaled to a $30 billion round. I'm gonna hand the mic over to you. How did we think about this?
I think I think if you're not that, you're not of interest to 98% of venture. I mean, literally, I I had people I mean, I'm not even in the loop. I don't have an allocation to Anthropic or a pro rata, but people I saw on the periphery of this investment that I didn't even know would be on the periphery investment shocked me. So Rory can talk about we can talk about whether it's cheap or not on a forward multiple basis. There's probably arguments each way. There's arguments it's cheaper than the last round, right? Just on a forward revenue basis, but not to be uh an echo or captain obvious this is just, you know, it's the same day Thrive or the same week Thrive raises 10 billion. This is what everybody wants to do. This is the only play in venture. Um when you know the public markets for soft for software stocks are down 20 some odd percent this year. This is the play. It it is that simple, right? There's almost no other play other than showing up to demo day.
I think actually Jason nailed the comment that this is the play. Look, the last three, four, five rounds have worked. The the two rounds in '25 have worked really quickly. You know, the 60 billion rounds at the start of the year, the 160 billion round at the end of the year. Remember, we talked about we said, 'Oh, that's not actually crazy.' And here you are getting a 2x again. You've now clearly reached a point where, as Jason says, if you have a multi-stage fund, you just got to put some money in and get your 0.1% ownership and add value because otherwise you're not showing up in the only thing that's working. And it really is one of the only thing that's working. I mean, if you eyeball down the unicorn list, what's amazing is, and I was just doing it, yeah, reflecting on this, even two rounds ago, it was the second to third largest unicorn, and it's probably outperformed most of the unicorns that are one-tenth its size in terms of share price appreciation in the last 12 months. This thing is a juggernaut with momentum. And you know, in the short term, in the short term, momentum massively outperforms value. So, everyone is wisely getting in on the momentum play. The thing that I think too is just like the elasticity of capital supply. Like, you know, I'm doing rounds today where even at the early stage, a $30 million Series A round has $1.2 billion of demand. And I've never seen so much money chase seemingly such a small concentrated number of companies.
But they are hoping there'll be another Anthropic. I mean, here's the thing. Here's the thing about Anthropic. I was thinking about about this with the SaaS crash is just, you know, gravity's almost gone up to Jupiter levels in tech. Everything's being pulled down, right? 56 billion is not enough for Atlassian. You've got to be 6 billion like Databricks growing 60%. Okay? Gravity is just pulling everything down. And so you've got to invest in these handful of folks that can achieve escape velocity not just on Earth but on I mean I'm not a I'm no Elon Musk but on Jupiter or something because the the flip side is everything else is getting pulled down by gravity down a gravity well right they're they're unsellable. I mean you look at AppLovin, 70% growth isn't good enough. Okay, Non coming down, look at Non for the gravity well coming down to a $2 billion market cap. Okay, that is a gravity well sucking everything down when the ship can't leave the planet. And so with Anthropic, it's left the planet. I mean, you know, uh um uh Claude Code from 1 to 2.5 billion at the end of the year to today, right? That's escape velocity. And so I, you know, the early stage ones that that's why it's a head scratcher for folks that have been around for a while because the bet is that you'll that you'll escape this this massive gravity well. But if you do, there's, you know, there's there's 10 billion from Thrive last week and 30 billion from Andreessen. But man, the gravity is is it's a [__] right now. It's a [__] gravity.
And when you say, what you and let's be precise in what you cuz I I get the analogy. It's like it's almost like it's a black hole in the sense of these two or three companies are sucking everything else in and spitting them out. Exactly what you said. So I mean now over the medium term, is that massive over extrapolation? I think so. I I think there's an element of projection going on here and you know a fair amount of discarding babies with bathwater but short term I mean short-term we are narrative creatures right and you're right Jason the narrative right now is AI is going to eat everything within AI the narrative is arguably less clearly um but nonetheless they the models are going to eat everything and and Anthropic looks like for enterprise it's clearly the best model you take all that together and this is the thing that's going to eat everything and once you have that you can justify any price that that that and conversely at the same time you see everything else just trading down to points where you'd sit there and you go some of these SaaS companies are trading at eight or nine times cash flow right and are they really going to go away in that period of time when they're growing 10 15 20% at the moment so my guess is look in the end it will equilibrate because that's how price works so but you find yourself trying to articulate what's the forcing function that will kind of break this over investments break the spell, right? And that's actually a hard question because we're talking a lot about the value of the company, but really you should also be talking about the performance of the company. I mean, this stuff is working. Um, and that's the aha, the revenue acceleration we've seen. I mean, the sound bite is going to be 3 years of 10x revenue growth, right? Technically, ARR growth, but yeah, revenue growth as well. And that's not only unusual. I went back and plowed through the early Microsoft, the early Google, the early Compaq. It's unprecedented, right? I even started to try and adjust for GDP growth, try and adjust for inflation. The truth is, you've never seen a company grow 10x in GAAP revenue and run rate year on year for 3 years, right? At this scale, just hasn't happened. So, you're leaning into the singularity here. Now, it's worth pointing out Microsoft, Google, many of these other companies that had 3x, 4x year growth for three or four years were also wildly profitable while they were doing that. With this company still losing a ton of money, but from a growth rate perspective, this simply has never been seen before. And at the margin, we're all, you know, growth chasers. It seems like what other $380 billion company is there a CEO where they say, but if we misspend on compute for a year, we're bankrupt. They're like, seems to be this absence of fragility not being priced in when he openly says it.
I think it's I first of all you're right and this is a reference to Dario's long and darkish podcast where one of the very sensible and I always find him frankly way more grounded than some of the other leaders in the space um where he basically said this is tricky because if you underinvest and you miss a cycle um the other guys pull ahead and then you're then your growth dies and we know what that's like if you overinvest because the money is so much if you overinvest and you know you hit the point where you don't get the returns to scale then you are really long compute and I think actually he I mean it was obviously the element of and then one could go bust but I think part of the message was we are trying to be a little more circumspect than the other major player OpenAI in terms of our future commitments to make sure that doesn't happen but yeah no it it's again it is back to the same thing these companies aren't like you know aren't like the explosive go internet companies or software companies like Microsoft because they are wildly capital intensive. They are much closer to semiconductor companies in terms of structure. You have to just, you know, put in 10. I mean, even Dario talked about the amount of capex he's looking at over the next 3 to four years. It's an astonishing amount of money. It's, you know, tens and hundreds of billions of dollars, right? These are not software companies with free cash flow. And that's probably one of the big risks. At some point, people go, hm, it's just better to own companies where there's free cash flow than companies where there's not. Right? And intuitively, you know that you can get to free cash flow by slowing down. And that's actually the getting to free cash flow probably also assumes slowing down growth. And once you slow down growth, you wake up and realize well on a GAAP revenue basis. You did 4.5 billion last year, maybe 15 plus or minus this year and GAAP basis. You know, maybe 30 40 times GAAP revenue is a little pricey. So it's going to be hard to manage that.
What percent of the 1 to 4 billion in revenue do you think is customer acquisition from OpenAI customers versus net new customer acquisition?
Never thought about it but doubt it's doubt it's a huge amount. I mean we're seeing some of the lots of enterprise having you know relationships with both which totally makes sense. So maybe at the customer count at the logo count level probably not a lot but at the individual token usage basis I actually don't have insight to that. Someone like OpenRouter might have that and you know that's where it would show up. But I mean to be fair stepping back a million miles most of chat most of OpenAI's revenues comes from ChatGPT on the consumer side and Claude less than 20% of the revenue is consumer. They really are slightly separate companies with an overlapping Venn diagram. Same raw technology. One very much gone enterprise coding the other mainly consumer but with some enterprise coding. So I don't think the narrative is I don't think the narrative is at a direct level Anthropic winning so OpenAI must be losing. I think the narrative is you know venture and soon Wall Street just loves AI and doesn't love anything else. Still it is crazy. I don't think um I don't think it's too early for Anthropic to be stealing enterprise deals for OpenAI because even those deals are one year and they've just been signed, right? Anthropic said it's its 100k customers are up 7x in the last year. So there's not enough time to steal deals. Having said that, going from 5% of OpenAI's revenue to 64% in 14 months, that is stealing budget. You can't argue with it. I mean, it's as a even though the budget is accelerating in ways we've never seen before, you know, global software spend up 14% this year, which is unprecedented, you still it's still a fixed pie even if it's growing. So I don't think anyone thought when ChatGPT broke out that Anthropic would have 64% of its revenue at this point. And uh that's a lot of momentum, right? That is Code Red. You thought Code Red was about Gemini. I guess it was, but maybe it's like Missile Command. They're coming from everywhere.
I think you're right, Jason. That's a great point. I mean, you know, there's a reason when we were talking about CEO of the year that was clearly Dario because if you look at where the story was a year and a year and a half ago versus where it is now, I mean, Anthropic just ran in and scooped the money and can't be great if you're the other guy.
Software spend is at 14% today, as you said, the highest higher than it's ever been. What do you think it is in 2030?
Look, software has grown pretty steadily two or 300 basis points, 2-3% above GDP growth for a couple of decades and that's kind of the the low rent the kind of the low drama version of the Andreessen software will eat the world. It's kind of growing now, but it's still, you know, three or 3% plus or minus the GDP, right? So, the probable answer is it gets back to the discussion we always have. It probably regresses to something like that unless you unless you really see that AI unlocks productivity and labor spend, in which case you probably could see, you know, five or 6%. My guess is the swing on whether it goes back to its GDP plus 200 basis points or stays at GDP and 400 or 500 is all about massive labor and efficiency and productivity savings have to come from the AI and if they don't you know growth will decline. You struggle to articulate a lot of bear cases for most of these AI companies which is why we're all investing in them. Everything's working. Yeah, competition is tough but the markets are huge. So you have to but there has to be some bear case. But probably one embedded one is are you really getting the ROI from this you know 10 20 30 billion dollars of spend that you're making in enterprises if you are then you'll probably do more it'll go to 60 and everything will be fine right the canary in the coal mine would be if a year or two you know enterprises are saying that was great but we're going to slow down on spend for a while because we need to get the ROI we didn't make the labor savings that's probably the thing to watch to the downside.
Here here's what I think though this has changed since we did this show I I think it's even changed in 90 days or 60 days which is that I think more and more enterprises are going to will this into existence. What I mean is you can make a decision. Do I want to invest like no matter what anybody says for a large enterprise AI does not miraculously replace 10,000 employees in an hour. It doesn't work that way. So you have to make a one you have to make a bet just like we've always made in enterprise software. But then you have to decide also I just don't want a bigger company. I also want to do layoffs. I also want to be smaller. I want to be leaner. I want to do it. And so I'm going to lean into this bet. And not only am I going to do it, but my friend Rory over at Nabisco is going to do it. And my friend Harry at Walmart. And I I see this. It's not just all the the the Dario and waving his hands. I mean, he's great. And and all these podcasters. I think enterprises are going to will us into existence. They want it to be true. And with AI, you can build almost anything you want now as we're doing this. You literally can build almost anything you want in software. So I generally believe even if it didn't have to be true, even if this was less a little bit less revolutionary and people might back off in 2 or 3 years, they might be like, you know, I didn't see this is classic enterprise. I didn't see this the ROI I thought from that Salesforce module or Workday financials. I'm going to I'm going to slow it down for a while. I I think you can will these things into existence. I I literally think you can make a decision and I think by the end of this year that that train's going to be so far out of the station that that these num these growth numbers will be jaw-dropping because we will decide enterprises will the Fortune 500 or Global 2000 will decide we are replacing humans with AI even if it's not the right decision because it's not the wrong decision it's doable.
I think actually Jason you're right and I think that's the single big picture statement here. Corporate America has decided they're going to make this bet. The Zeitgeist is making this bet. It's unstoppable now. I mean, even what you're seeing on the Wall Street has decided this is the bet they want to make. Capital has decided this is the bet they want to make. People are going to try and make this bet, which is very different than saying when they made the bet, they like the result. But that's two or three years out, right? My gut would be people will overb, overinvest, and you will have a retrenchment period, yeah, two plus years from now. But I think right now you're just looking at two years where, you know, just as the hyperscalers said two years ago, we're going to do this and we're not going to blink. You're right. Corporate America is now going to say we're going to do this and we're not going to blink. Which means you're looking at, you know, one to two years of mega AI budget, not regardless of AI of ROI, but on the presumption of innocence, which is it's the presumption of success, maybe is your word, J. I like that. That's what they're going to do. They're just going to say it's going to work. This is the thing you got to do. And you know CEOs generally have only two or three agenda items at any one time. You know it was a whole period in the '90s Jason when it was all about right sizing and efficiency. Then it became about getting on the internet then getting on the cloud. And you're right the number one thing now is you know make your big AI play. So everyone's going to make it. Everyone's going to spend there's going to be a couple more years of great spending and then we'll see. That's the movie which means that the next two years would be a good time to access the capital markets.
It's also why I hate to say I'm just not I want to be really bullish and say that a lot of these public software stocks are are oversold and and maybe it's true but you know when I mean Harry just had Sebastian from Clari right and he likes to be a rabble rouser but he's like I went from 6,000 to 3,000 employees we know that but then he said on Harry maybe he said it otherwise he said on 20VC and in two years I want to be at 2,000 now he's trying to be a rabble rouser and he was trying to say I I I you know I I fired Zendesk 2 years ago and AI is magically perfect. But I think that's how everybody's thinking that I talk to. And that is great for Anthropic because we're going to replace those humans with Anthropic spend. And it is terrible for for not every software company. Maybe Shopify is oversold. Maybe Non is oversold. But overall, it's not even for Mike Cannon-Brooks. It's not great for Atlassian if we're all going to shrink our teams from 6,000 to 2,000. It certainly is. And and maybe HubSpot and Monday. And no pe we're like why the hell are these horizontal apps oversold? It doesn't seem fair because the numbers are great. But if this is the mode we're in um and nobody wants people uh it's hard for me to feel there's a bottom in this and it's not valuations, right? It's hard for me to feel that they're that enterprises are saying, you know what, I want to buy more seats next year, guys. I've changed my mind. I'm tripling my seat count at HubSpot next year. I've changed my mind, guys. We're going all in. We're going to hire 20,000 more people next year. Humans um who complain and whine and quit every 3 months and we're going to get them all seats of software. I just nobody I talk to has that feeling today. Nobody wants to do that.
I broadly agree but I on the kind of sentiment and I disagree on the statement that there is a bottom. I mean in the end price clears all markets right and what happens what's happened on these on the public SaaS is you've gone from as I said the presumption of success which is what AI now enjoys to the presumption of failure right but that doesn't mean you fail right it simply means that you just look cold-eyed at it and you say what's the growth rate what's the free cash flow how do you value that there's going to be a price at which these things stop declining and compound at you know a normal 15 12 to 15% return a year, right? The you know, markets work. The other side will overshoot. This will be doable, but it's just not a number you like. But here's the thing, and of course, you're right. If if it's just if if you have stable or growing free cash flow, there's a number, right? But but here's what makes me pessimistic, and I don't want to be pessimistic. I want to be the guy saying we're never going to remove our systems of record. I want to be that guy. But as a cohort, public software stocks are are approaching have fallen to almost 10% growth annualized. Okay. Now, if AI just reduces that another 30%. We're in essentially the dead zone, right? We're we're in that area in the deep dark water where nothing grows because even the plant the dead bodies don't fall. That you can't like 5% 4%. It's it's it's it's even worse. It's it's all just price increases and suing customers. There's nothing. If if AI was bringing us from 30% to 20%, you could make a whole bunch of arguments. You could say, 'Listen, we're we're we're we're J, you know, we got a little SaaS got a little older. We've slowed down. We're in senior marathon, but things are still good at 20% growth. At 6% growth, I don't see any future other than the next 5 years of free cash flow.'
I think it's a valid worry if because things were bad before AI. Like let's be honest, they things had already fallen into the mid to low teens before the AI wave. It's not like SaaS was healthy. Public SaaS companies, these were not thriving. Dropbox growing minus 1% a year is not crushing it, is it? That's why I think the market isn't overblown because we were already at risk going into 2026. We're already weak and anemic. The overall blended growth rates and Rory made the point with Mike last week part of it is because we haven't put new names in when we drop Anthropic, Databricks and Anthropic and it's all going to and if we if we allow them to remain of a pure basket it's going to look great but the existing group is not looking healthy it it's not if if we if we look at like a Asana, a Monday, a Non these businesses they're all doing okay to decent actually but the price is through the floor and we keep on saying oh my god it can't go lower and it does genuinely. What what happens?
Well, I I look I think the markets, you know, you know, it's like the old cliche, in the short term, it's a voting machine. In the long term, it's a weighing machine. We're going through the voting stage now, right? My big aha is, you know, is and Jason mentioned earlier, I was thinking a lot about the SaaS apocalypse and, you know, did AI kill SaaS? And I think what really happened is Wall Street fell in love with AI and it fell promptly and and you know and and had to and to do that had to fall out of love with SaaS, right? The burden of proof is the other way, right? But you you got to shake out between them. If you take something like Shopify, I'm going to say it. I don't think there is a credible near-term story that replaces a shopping a a website for shopping and a payments mechanism which is you know 60 70% of their business with some kind of AI arm wave in a way that there is for many of the workflow automation software companies. I'm just picking on Shopify for in a positive fashion.
I'm super positive on Shopify.
Exactly. So super positive. Right. But it is in the basket. It is cousins of the rest. Right. It it is probably over we can argue it's oversold and if we were running a hedged a long short hedge fund maybe we put 5% of the fund into Shopify right it would make sense there is even though here just to be clear I I do think Shopify is oversold it's easy for us to say on a podcast right I do think though there is a bear case for Shopify what's the commonality between ServiceNow, Salesforce and Shopify it seems like nothing but actually they all could be abstracted away into a database even Shopify can't if I am shopping on ChatGPT right uh I I may not go to that merchant store ever. And in the short term, it doesn't hurt Shopify because it is the plumbing of the store, right? And it takes a piece of the GMV. But ultimately, if the future of e-commerce is conversational commerce and it does not happen on the Shopify platform, that is not a net positive. And Shopify and Tobi says the same thing like that's why he's done more code commits in the last 60 days than the rest of the history is because the Shopify software, even if it's only 25% of the revenue today, that software in 2 years may be obsolete. If Tobi thinks it and this is the most resilient AI and Tobi thinks it, what what hope is there for mere mortals?
I I but that assumes the UI of the future is conversational commerce.
It is the UI of today, Harry. If you're deep in e-commerce, you talk to anyone.
It's the UI. It's the UI of today because it's what the most productive people in the world find the most optimal UI which is Sam, Elon, Dario. Actually, when you look at consumer most often, they actually like browser based UIs. They like discovery. They like an option set. It's I I I agree. I think that again don't over steer. I think look I definitely think you know chat is wonderful for research which is why as despite your opposition in the past I've loved the you know the geo marketplace. I think chat is the place you go for considered purchase and look doing research on you know whatever you're buying right but then you go to the store you look at the thing. I don't. We're a long way from I'll just press the magic button on ChatGPT and have it come.
No, but it might not be on Shopify's website. It might be on an agentic flight. It doesn't have to be ChatGPT. It could be Harry and Rory's shopping site that that is better than the built-in conversational commerce and Shopify. Everything's at risk to an AI believe every platform that is open is at risk that an agent is better than the native platform. and every platform that is closed is is at risk of of of being at least slightly bypassed. And that's the risk is that I don't go like I haven't I haven't logged into Salesforce in 6 months, but we use it every every minute. Our agents use it every minute. I think it would be disingenuous to say that couldn't happen to Shopify that agentic commerce agents forget about part of discovery is going to happen on my sofa, don't get me wrong, but all the stuff that really matters may bypass the user interface of Shopify. I just think it's a threat. I'm not whether the size of a threat is as large as it is to others. Probably not. But it's it's not to be dismissed given the incredible pace of of progress we're making that it's like I I can't keep I I'm struggling to keep up.
What I don't like about this discussion and what makes it hard is and actually it was one of the most insightful things I saw about in the last couple of weeks about this last discussion is it's really hard to disprove a negative, right? I think your stock might go down. How do I disprove that? I think you might be displaced by AI once the I mean my big aha is in the I'm going to repeat it again. In the short term we're narrative and momentum creatures and all that's happened now is the narrative and momentum has shifted to AI is going to do everything. So the people who are left trying to prove that that's not the case are left as you say trying you can always articulate a scenario right and the question is on a balance of probability basis is that going to happen right now. So, and the thing is smart traders don't fight the tape, right? Right now, the momentum basically in the short term the momentum is all around this narrative of AI replaces everything, right? That's the way the momentum trade's going. At some point, I don't believe it will. I think it'll do. I think it'll be amazing. Just like software was amazing. Let me repeat, software is the best industry of the last 20, 30 years. At the end of 20, 30 years, it accounts for 4% of the US GDP. We'll still want to eat. We'll still want to drink. We'll still want to drive cars. Everybody should get a little bit of a grip and not extrapolate to the end. But whatever, right? That's the movie right now. So, while that is happening, you would be a buffoon to get in the way of a momentum trade. At some time, the momentum trade will dissipate and people will go, 'Ooh, this is really amazing, but it's just not quite the universal everything.' And that's the point at which you see a correction. But until then, narrative dominates valuation in the short term. Not.
I think it's right. I just had one thought as investor, public SaaS CEO who as I say trying to you know prove a negative I let me prove to you why I'm fine it's just too hard.
Just one thing the tough thing as an investor and and and maybe is you have to you know you you do also have this um this lab it's called your portfolio.
And and your portfolio one thing you could say today is hey my early stage companies aren't truly at at scale scale is north of 500 million or a billion in revenue so they don't matter but on the other hand you may be seeing the future in your portfolio and publics aside when I look at my last fund right you know and I see what's happening and it the fund's in good shape whatever 4.5x but honestly like two companies are like massively benefiting from the current world in B2B massively benefiting accelerating one I would say is a push like it's benefiting um north of in in nine figures of revenue but also being hurt and everything else is struggling because of AI all the rest of that last fund and and and you can and and the you know it's not all growing zero. I'm not saying that. But man like and every and and it was so similar in 2023. So is that the past or the future? I get confused. But if you have a portfolio of companies exposed to all these things, they they can't hide in high NRR like ServiceNow. They can't hide in price increases like Salesforce. They're getting whiplash at a pace much faster than the public. And so when you see this on the first of every month when you get updates, it's hard to be optimistic in 2026 about anything not already getting a boost from AI because you see the dramatic effects.
I think that's correct. And it's just hard to to keep grinning and smiling and saying go guys when when you don't see any AI boost in your February or March investor updates. It's it's hard to remain optimistic.
I think that on the other hand is probably true, right? I think and and so I just want to one comment on definitions. Everyone says the SaaS is dead and it's such a useless word because we have to define our terms just a little bit more carefully. If you if you're if you think software as a service means selling software on a monthly basis for a you know fixed or variable price then the the thing is Claude is SaaS, Harvey is SaaS and Salesforce is SaaS right so when people say SaaS is dead are they saying there there's kind of two separate threads here this within the new world of AI are we talking about how much the model will get versus the next kind of I call them new school SaaS like the Harveys of this world the next generation companies that pejoratively are called wrappers and um optimistically are called you know standalone companies that can you leverage on top of ChatGPT then separate from that there's old school SaaS so I think what people are people most of the time people are saying old school SaaS in other words stuff that was built pre-2022 pre-GPT is what we're really talking about here right most of the time then sometimes people go go all the way to the you know all software is dead and the model's going to eat everything first of all you have to just distinguish between those two scenarios, right? What you're talking about Jason and I agree is let's for lack of a better word old school SaaS built pre-2022 right and what you're saying which I think is also true is that if you're public already you're now living in a lower in one of those companies Salesforce, ServiceNow, on you're now living in a lower growth world where you got to make the model work financially while at the same time investing in the future because if you just make the model work financially as Mike Cannon-Brooks said last week the good news is you have 5 years of 30% free cash flow and the bad news is in year six your revenue goes to zero right so you got to do the two you got to be financially disciplined while adding AI to grow it right and I think what so I that's that's again I'm going to distinguish public versus private that's the public SaaS company movie right and you're right then the la the tough category is the small private pre-2022 you if you're not getting SaaS lift you really got to worry about what you're doing right especially if there is a credible SaaS story in your space I do believe again just to say There's a lot of markets even pre-22 that are software plus payments, software plus right that aren't all AI. It's not a baby bathwater thing. I mean, but definitely the sweet spot of your argument is workflow horizontal workflow related software built pre-2022 that's not public with critical mass. That's probably the hardest place to be.
I just see a realization in my portfolio now of companies that are in the 8 to 20 million range, not benefiting massively, but growing 80 to 100% that are basically like, okay, we've raised our last venture dollars and we're going to have a good business, but we are moving off the venture train because we can't align to what is needed today to make that next round work. And that realization I'm seeing set in.
Yeah, that's probably fair. Uh, is Figma a baby getting thrown out with the bath water?
No, because I mean, so there's two questions. Is it and should it? Right. The is it question, it's still valued at 12 billion plus or minus, right? It's still an amazing outcome. It's still I think roughly 10 times plus or minus revenues, right? And I remember I used to know the growth rate and now I don't. Yeah, it's it's yeah, 30 40 I I don't swear to this in that kind of it's not 10x-ing. It's 30 or 40. It's not down to let's call it miserable single-digit SaaS. It's a perfectly great public company 30 40% maybe higher growth rate story. So yeah 10 times revenues plus or minus it's absolutely not getting shown out with the bath. Well it's probably it's probably just getting valued correctly in the absence of narrative lift. Right? Whereas two years ago it would have been valued at its value plus narrative lift. Right? It's it's gone to to use the other analog it's gone from the voting to the weighing part of the capital markets. It's now right. So from that's from a value perspective. And then the other part of it is you know where does what's driving that is the where does it go comment and I'll refer to Jason on that.
Well here's the bull case on Figma. You can't argue like here's the existential problem with public. You can't argue with the current numbers right? You can't argue with the current growth north of a billion. Great. Right. The other thing you can say is, okay, let's
Go back to your favorite rep level. Number one use case for both is product teams building prototypes. Okay, Anton said this last week. Number one use case at scale, not when Harry invests. Number one use case for both. Okay, that is all revenue Figma should have. Figma Make is a failure. And I don't mean to be dramatic, but I think folks at Figma, if they're honest, would agree. Like, they thought this was revolutionary. But that is 500, 400, 300, 400 million of ARR Figma should own. They should own this. Lovable and Replit should not be owning product prototyping and development for product people. Figma should own this. They don't. They missed a whole generation here. And that ties into all of this. Figma can't argue with Figma today. That checks every single box on growth, market share, everything. But good god, they missed 400, 300 million of growth in their core for AI. I'd be pretty critical, right?
It's actually double that. Both Replit and Lovable are 350 independently.
Yeah, but not all of it is pro. It's just the number one use case for their enterprise. I'm just guessing it's for both. It's their number one use case for their larger customers. That's the majority of the revenue today versus the minority. But it's not all of it, right? So I'm assuming that morphs to 300 million of bookings. But Figma should have gotten all of it. If Figma Make was better, why do I want to buy another tool to do this? Especially because these tools weren't even really good until the fall. As you guys remember when we started this, these tools weren't that great in the beginning. Okay, Figma should have owned this. This is just the bare case, right? And my god, now we're in 2026. Why is... and the earnings will come out I think before this pod comes out, but why isn't Make a 300 million, 400, 500 million ARR business? Like, this is much easier to sell than these damn seat licenses where we have to bang our heads against the desk to get 20 bucks or 50 bucks for Figma. I can just sell multi-million dollar deals of Lovable and Replit just because my product guys can now ship functional prototypes. Like, they missed this huge change. Base 44 have also absolutely crushed within Wix and their growth. They have... I don't know that they sell to product teams like this for Figma, though. I think what... here's just my criticism of Figma. And again, it's easy for me to say, 'Good god, I didn't build a company as good as Figma.' Okay, but the honest criticism behind closed doors is Replit and Lovable shouldn't have gotten here, guys. We own this space. We let it go. And we're just seeing... and this may happen to everybody, guys. Everyone that's too slow. If Figma was too slow, what hope is there for some of these other public companies?
And I think you're exactly right. Well, it's brutal. First of all, I think you are correct. So let's do Figma first and then the other companies. Yeah. And I think back to the valuation. I think the valuation is perched between, you know, if you'd done this and you add another 300, 400 million of revenue, your growth rate will be north of, you know, 120%. You would probably be valued at 20 times revenues, 25 times revenues, right? If you don't get on top of this soon, that 10x revenues is going to go to five and four, and where the other 7% growth are. So markets are doing their job. There's probably... there's an embedded probability that they make a better product and take more share here, and then... they don't, right? And it's kind of in the middle. And you're right, fast forward five years, there's really only one or two stories. They get on top of this, they get a third of market share, you know, valuation continues to grow, they're fine. If they don't get on top of it, you know, you trend down to the horrible multiples, right? But the comment I would make though is I do... and again this is back to the not everything will roll over in one day, right? I do believe ironically, even though Figma I think is one of the most exciting companies of the last five, ten years, I think that space was very vulnerable to disruption because it turned out to be very AI adjacent, right? And I do think, and I'm not trying to sound boring, I do think there's a continuum across all these companies of how adjacent you are to what the models could do. And there are companies that are very close to it and are going to get, as you say, sucked into the gravitational field, and there are companies that are fairly far away from it. Right. I, for example, you know, I don't believe accounting... I believe there will be change in accounting. I believe the next generation of accounting software companies... we've looked at them, they're super interesting, they're adding AI, but fundamentally 80% of the value in an accounting software package is independent of AI. It's just debits and credits and a good UI.
How adjacent is Canva to AI?
Probably pretty close, too. I mean, anything that's creative... individual creativity, individual coding, and individual creativity strike me as extraordinarily close to AI because it's happening right now. You know, conversely, I mean, I'll say it, even Salesforce, we talk a lot about, and Jason, I hear you on your agents, but the truth is the level of disruption in go-to-market software hasn't been as acute as it's been in some of these creative areas or coding or customer support. Now, I'm not saying it won't. I love what you're doing with your agents, but just being objective about where is the eating happening fast, where is it maybe happening slow, maybe where has it not happened at all, right?
Yeah, but... and you're of course right, but I just... listen, I'm an investor in a company called Monaco that Harry knows. It's the next AI SDR for us. We already have four, okay? And they launched last week and we went live last week at the same time. A couple of things that are interesting. First of all, from a technology standpoint, it booked us a six-figure deal the first day. None of our other agents could do that. It booked it on its own, a six-figure deal. Now, that is interesting because that is the pace of progress. Our other products, we had Agentforce, Artisan, Qualified, and a few others. They're all great. We use them every day. They couldn't book a deal on its own the first day. So the rate of progress is accelerating. Like, we couldn't believe it did it on its first day. Okay. And then here's the second point. They have every demo booked up through the summer. If they had enough manpower, they could do two to three million their first month because the demand is so strong. The demand... now there's fragility, there's turnover. We could talk about how prompts are portable. And the fact that this is so good that we would throw out an agent we bought a couple of months ago shows you how fragile everything is. But my point is that the demand is so strong as soon as the products catch up to the demand. Everything... maybe these sales tools are not safe. Like, so much demand in one week, just a torrent of demand. And I see it in all of these products that it is inexorable. And it just got... like, this one is better than the one from two months ago, which is better than the other one. It booked a six-figure deal on its own the first day. Jesus. Holy... people don't like in the pod when I say JFC, but holy cow.
When you say booked, what do you mean?
It reached out to a leading hyperscaler. We've talked about this, that wants to do a six-figure sponsorship at SaaStr on its own. No human required. It decided who the targets were. It loaded them up. It went out to the person and it booked the meeting.
The meeting or the money?
Well, it didn't close it. That's the end of the year. But don't... none of these agents could do this before February. They could do a lot of it. They could send emails. They could do a Marketo or HubSpot 2.0. They could customize it, but they couldn't literally go out to a VP at a very busy tech company and get the meeting entirely booked and loaded and ready to go. Like, it is. And that's just progress in 60 days. What will it be like at the end of the year? It could completely close it, right? Yes, it sent a well-written email and was able to schedule a meeting. Yeah. And you can crap on it if you want, but my point is the products weren't this good 60 days ago. You can say, 'Oh, I have this company in my portfolio that could do it.' Well, show me the money. Like, we've done all the... we have 20 agents. I'm going to tell you this wasn't possible on its own. You could get close, right? You could have folks qualify themselves in and go to Rory's website and say, 'I want to,' and the AI would move it down the funnel. But to do it raw... people talk about it. I don't mean to spend all of our time on it. It's just... this is why I really wonder if anything is safe from disruption. And I actually don't think any of the agent vendors are safe from disruption because we will... even us, like, everyone's promiscuous with agents. We will switch it. Like, this is so much better than what we had two weeks ago. We'll switch to the next one if I... they're too good.
As is often the case, I hear you on the narrative of just pushing on the conclusions because look, I totally get the AI SDR use case. We're in Regie. We've absolutely seen things get better and better. What's been super interesting is the product a year, a year and a half ago had a medium level ROI, low level ROI. It's taken iteration after iteration and now you're starting to see with the reasoning models increased performance, as you say, starting to be able to convert, you know, old leads to active leads. You're starting to be able to predict who you should reach out to. Definitely a lot going on here at the top end of sales tech. But my point is just compare it, for example, with, as you say, to take one of your other two examples, coding. There's a two and a half billion dollar market already taking place. There is not a two and a half... not all the markets are moving at the same speed. If you take, you know, again to your example of... I'm sorry, my brain... Figma, right? The next generation product tools like Lovable and Replit, right? They're doing three, 400 million in the space. My point is merely the pace of adoption is not consistent across the board. There's a sequencing question here for sure. And this we can spend all this time, but at a meta level, like, in some ways I don't think it matters because there'll be spaces that are destroyed, right? Like, and there'll be spaces that are maimed. And I don't know that it matters on a terminal basis. Like, if your space is going to be killed in eight years versus eight months, if you've gone on to hospice care, does it really matter how long you stay in hospice? I mean, to the employees it does, to the family, but everybody pretty much knows when that SaaS company goes into hospice, it ain't coming out. Depends on how long it's going to be.
I mean, look, you say, yeah, I mean, there's a period of time in which you probably haven't got time to adapt, and then the longer you have, yeah, the more chance you have to adapt. Again, as I say, it's utterly boring trying to prove... you know, you will play the game and find out, right? I don't know if everything gets rolled over in two years. I think, look, give an example.
Not everything. Not everything. The closer you are to code and support, the faster the disruption's been. No argument. Let me just rest on one thing. Just this is why I feel anxious about everything. But I don't have to run this. Figma IPO'd in July of 2025, like right when we got this podcast going, right? Freaking rocket ship, right? Actually at the margin seemed to be benefiting from AI at the margin, right? Just since July, Replit and Lovable take 300, 400 million of their market since July, and maybe even since September or October because the product didn't even really work until September, October. Good god, this is the best. I mean, we were saying Figma was the best of the best that there had ever been in July. And now AI has stolen... and yes, product is one derivation away from coding. Like, no argument, but let's not view any islands of stability when in July this was the best thing we'd ever seen.
Yes. Just... we went around the room in July when the stock was at 110 and asked where do you think it'll be in six months? My vote, I think, was 35. So it's 20-some. But I was pretty right at the point in time. So I get it. But on the other... pushing back, another... the SaaS again, the question is the velocity it happens. I was just thinking about kind of the SaaS change and, you know, we both started investing and look at this stuff, early 2000s, right? And, you know, Salesforce went public in 2004 and ServiceTitan, which is also a SaaS company, went public in 2024, right? So it took around 20 years for the whole thing to transition, right? I don't think it'll be that slow here, right? But I also don't think it'll be two years, right? And I think there is... and I think that matters a lot in terms as an investor where you choose to place your bets, right? I mean, I think that... what we find ourselves thinking about is which markets are going to adapt quickly and which markets are going to take more time. Because, for example, two of the most interesting app categories, other than coding and customer support we've talked about, but if you look at two of the most interesting app categories have been the entire law field and the entire healthcare doctor information field. And it's no accident that both of those didn't really have a compelling old-school alternative, right? It was much more greenfield than some of these places where you are competing against existing SaaS companies. Just worth pointing out that the sweet spots of adoption, it's this kind of jagged edge idea. It's not all happening uniformly. Some places it's happening now, and those have been amazing categories for categories that frankly weren't great in SaaS land. I mean, legal was a miserable category in SaaS land. It's been amazing in AI land because LLMs manipulate language and lawyers manipulate language too, and however you want to interpret manipulate, right? Whereas some of the other more structured apps, it's been a lot slower. That's my point.
Well, you know what? Maybe it is... maybe just to wrap up on the point, May's the boss, but maybe it is an interesting investing question, right? Because we started this on topic. You could have two strategies to investing in AI and software. One is let's invest in massive disruption today. That's Anthropic. That's most of capital. Another strategy, and maybe Harry would crap on it, is listen, actually I want to invest in AI. I want to invest in the categories that are changing the slowest but with the best founders. That way there's a little bit of extra time. Some folks may be missing it. It's not that it's not coming. It's going to be as disruptive as Anthropic and the rest. It's just I've got two, three, four years instead of two weeks.
Right. It doesn't work unless you can do lifecycle funding as well, like because with the opportunity cost of cash, you have the concentration going to a certain direction. If you take the second route of the contrarian approach, you have to lifecycle fund that until it becomes a...
Not contrarian, just slower.
Yeah. It has to be more efficient for sure, right? You can't raise four rounds a year in that approach, right? Not for a while.
Which is why I think that, you know, the best place to be is neither of those two alternatives. Actually, the best place to be is about a year before the peak in terms of saying what's going to happen next. And that's obvious when you say it. One of my quotes I often give is the Henry Luce quote of Time. I think I've said this before here. He said, 'My job as the publisher of Time magazine is to be six months ahead of the American public. Not two years, not a week.' Right? And it's the same thing here. I mean, you know, fair credit to the smart folks who looked at Anthropic in May of '23 at Spark Capital and Menlo and said, 'Hey, this is the trend that's going to be.' And then you get the momentum, right? You don't want to be looking at something that's not going to happen for five years because I'm with Harry, right? The trick is to have... that's why I'm saying try to construct some kind of thought process on sequencing, which of the markets are most vulnerable, which of them will happen next, which of them will take more time, in my view is going to be a key part of the app side of picking where to play. You also said about customers willing it into existence. What I see in law, and I don't know if you do, Rory, too, but is just big law firms willing AI into existence in a way that they never did in a SaaS era where your Justin Kans and your Atriums were coming with products. They were not willing in with the same persistence that these customer bases are today.
I agree. You're exactly right. I remember because we, you know, have invested in AI companies for 20 years. I remember for a long time there was deals where you go, you don't emphasize the AI because the customer would get scared, they wouldn't like it. It would slow down the sales cycle pre-2022. And what's happened, and Jason said it earlier, the big aha here is post-2022, if you don't have an AI strategy as a corporate leader, you're a buffoon. And no one wants to be a buffoon. So...
Whoever's going to have an AI strategy, therefore, they're going to spend money. I mean, if you're right, Harry, if you were running an American, an AmLaw 500 or whatever it is, right, you're not going to sit there and say, 'I'm not going to make a play because I don't think it's all going to work,' right? You're going to make damn sure that you do something. And therefore 500 companies went, all of them went to spend, I need to spend a million bucks right now.
Dating a lawyer, I see this every day. They are competing on price just like us as venture investors. If they are not using AI, they are not able to charge a lower price and win that deal. And it is that existential threat that they will lose consistently if they don't. Which is why, yeah, I mean, it gets back to kind of where we started, is there's nothing better than being on board a market where all the participants have decided they want to buy something right now. I mean, that's why Anthropic is working. That's why some of these other markets are working. It's picking the markets that just tip, and you know, it generally lasts only two or three years and everyone's made that decision. Right. And that's what's happening here right now in some of these markets.
You said about 20 minutes ago about like a narrative chasm or a narrative shift with regards to one of the companies we were talking about. When I think about narrative chasm and news cycles today, the most striking is Stripe being worth $140 billion and Adyen being worth a third of that. Can we just try and understand, is Stripe wildly overvalued? Is Adyen wildly undervalued?
Actually, no. I went into this thinking, oh, I want to find some narrative story, right? But in fact, there are perfectly rational reasons, with one caveat, why they are where they are. I mean, for a start, so zooming out, Stripe just raised privately, I think 130 or something like billion. Adyen's publicly traded, oh my god, I looked at it this morning, only my brain has gone here.
Less than that now, right? It's half. So, a couple of things. One is, Stripe's doing 5 billion plus or minus, Adyen's doing two billion, right? So it's half the size, so two and a half times the size. So instantly you have to multiply by two and a half, right? You don't have, Adyen is very, very profitable, right? But the growth slowed, right? Stripe, it's not clear it's profitable, not clear as much, and it's not clear where the growth rate is. So they're a lot closer together than one would think when you allow for size. But what's really, so what's, so then on top of that you have the, it's the classic, I mean, it's a little like the whole SaaS, do you want mid-level growth and massive profitability or do you want more growth even at the expense of profitability? And right now people are opting for the latter, right? I mean, basically instead of saying, you know, the Stripe price is wrong and the Adyen price is right or something, just saying is that it's the eternal venture question: how much extra in revenue multiple do you pay for how many extra points of growth? Right. If Adyen's going 20, and I'm sorry, I knew these numbers but we're starting early in the morning. I looked at them last night and the detail is gone, but I had the sense of the relative order. If Adyen's growing at 15, 20%, how much, and you pay, yeah, 10 times, how much extra revenue multiple do you pay for something growing at 25%? 30%? Right. And I think that those are the contrasting narratives here. Right. I do, you know, I do, I mean, look, Adyen is a wildly profitable company. It's like almost 50% operating margins. Growth slowed.
This is a company also that just wildly miscommunicates.
Or communicates very little about how much cash they have, about how they're going to use it. And I think alternatively, Stripe communicates in an incredibly strategic way. They're telling a brilliant narrative, ironically, given the fact that it's private. And there is this, I do think there is this narrative chasm. And I do think you can actually blame the communication of leadership at Adyen directly for a valuation mismatch.
Line to that.
Do they even have a podcast at Adyen management?
Do they? They should give up then. They should, they should give up 40, 50 billion if they don't.
What do the Dutch drink? Not a cheeky pint. They need something to drink. They're the Dutch. I don't know what it is.
21% H2 2025 year-on-year revenue growth. Would you rather buy Stripe at 130 or Adyen at 40 or 50? I would go with Adyen because the likelihood of it being mispriced is lower, to the upside is lower, right? In the sense of...
Yeah, you're trading on value. I mean, look, you have access to the data. You know it's wildly profitable. You can value it, right? On the other case, you don't have access to the data. It's a private market. It's a wonderful company. I admire that company enormously. Isn't that meant? Again, you have to distinguish.
Disclaimer, people, for the Collisons that are listening to this episode, Rory is not in any way being detrimental to Stripe. None of his comments are seen as a criticism, and he is intensely sorry for even talking in any negative light about your company. Please.
I'll tell you the simple reason I would take Stripe.
Well, but that's the role of price is to equilibriate between the other issues.
I'll tell you the simple reason I would take Stripe, and it kind of echoes in my brain from Michael Cannon-Brooks last week, right, is just the flexibility and agility today you have from being private in today's world. It's so stressful being public. There's so much going on. You heard Mike's point, I got to be more profitable and massively invest in AI. I mean, he was great, right? It was such a great one, right? But you could hear that, and he was up for the challenge, like to his kudos, right? He's up for that challenge. But to not have to make that trade-off perfectly and be private. Um, even his sort of cheeky comment of his buddy Cliff didn't have to deal, maybe he'll wait to IPO like that. I mean, I'd rather be Stripe, and so I'm going to bet on Stripe on this one just because I think you have more flexibility today to respond to change as yet. My prediction is back end of this year we'll see the big three guys diving for the public line: Anthropic, OpenAI, and SpaceX. And I don't think it matters when you're profitable like Stripe, you really can stay private forever. But companies that need big capex for the foreseeable future are going to go public because, you know, they've sucked up all the private capital and now they need to go.
Isn't it a funny time where the public markets are both at the same time the most attractive place in the world for a company where it has the ability to be meme'd or consumer loved like never before, but also at the same time be the single most hellish spot on earth? And it's both at the same time for two very different sets of companies.
Yes, it sucks.
It's problematic for the public markets because, yes, you have this thing whereby you can only be in this public market when you're pristine and your narrative value is high. And the problem with that's not all the companies all the time, which is why the number of public companies continues to decline, right? I do think at some point it has to be an easier place to be public while you're dealing with, you know, while you're dealing and managing transitions. If the idea is every time you have to deal with a problem, you have to go private to fix it, that strikes me as a little absurd. And it's a combination of lawsuit avoidance, board exposure, you know, I think indexing and very much shareholder bases that have little patience, activists, there's a whole bunch of reasons why it's shitty to be public. And the only good reason to be public is when you're hot, capital is dirt cheap, and there's lots of it. That's an okay value proposition for the best companies in terms of going public, but it's not a great value proposition overall.
You mentioned OpenAI there as one of the companies diving for the line. Very big news was obviously OpenClaw creator, I'm going to probably mispronounce it, but Peter Steinberger joining OpenAI, the open source bot becoming a foundation. How do we break this down, Jason? I'm so intrigued to hear your thoughts on this because you spoke about how Moltbook was running different agents and how you were experimenting with it. How did you analyze the news of the acquisition by OpenAI and Peter joining OpenAI?
Well, look, I don't, on the one hand, when OpenClaw, whatever it was called back then, Claudebot or whatever it was, at first I was like, and I said this to one of the top CTOs I work with, I said, 'You're going to turn off this because I was already there last summer when an agent deleted my entire database.' And now, basically, what this app does is it's designed to break guardrails. It's designed to allow you to go onto your C drive and onto your desktop and access things that the leaders don't want you to do because they know it's a problem, right? And it's designed to sort of run pseudo, not really, but pseudo autonomously 24/7, which Anthropic or OpenAI could do but chose not to do. So I'm like, you're going to, there's no, there's, you're going to turn off this. It's a, it's a, it's a, it's a dance, right? It's a proof of concept. Um, probably still true. You're not seeing amazing applications in the last month that have come out of it. But one, it has ignited the developer community like something we have not seen in a long time. So buying into that mojo has some value, right? Buying into one of the fastest growing stars on GitHub, fastest growing deployment, the fact that every cool engineer is playing with it, it may be ephemeral value, but it's real, right? The fact that Zuck lost the deal and cloned it on Manus yesterday pretty much shows you the technology itself is not that differentiated. Right now it's available mostly hosted as of last night on Manus. Um, so you know, I, you know, is it worth a hundred million to OpenAI? Maybe if that's what the price, I don't know what the price is, but a hundred...
I think this would be worth a billion plus.
I don't buy that it was sold for a billion. I don't buy it, but maybe it's true. You have the numbers. Um, do you know...
I think, I think it's about 100. Um, that would be my guess based on a number of things. Uh, maybe it's more. Um, obviously he probably turned down more from Zuck, right? Which is a great, which would, which is a great insult. But um, but I think what it opened everyone's eyes to is, and I think this is why Anthropic was so dismissive of it at first, was the safe thing is we don't want semi-autonomous agents running. We don't want agents running rogue 24/7 with no guardrails, okay? Breaking guardrails, saying crazy like this is the AI, today's AI nightmare, this future AI nightmare is coming. But it's too late. I think it's too late now. Like now every developer wants to develop truly autonomous agents. And whether this platform will decline, right, if he gets bored and goes on to the next thing or not, I don't know. But the horse has left the stables for these autonomous agents, um, and the risks associated with them. Um, there's no reason that Anthropic couldn't have built this last year. They just didn't think it was safe.
They just didn't think it was safe, but it's too late now. Just like Elon said whatever six months ago, I wish we could delay AI three to four years, but since we can't, I'm going all in on xAI. I think this is the moment where we shouldn't be doing these semi-autonomous agents, but it's too late. We're doing them anyway. So the risk has been elevated, but the developers have just, everyone's excited. So the amount of innovation that will come, it's hard to predict, right, when all of our apps run 24/7 making their own decisions, deciding whether guardrails are appropriate, working around them, the, you know...
Jason, what does this do for inference requirements?
In theory, it's an untenable thing, right? That's why not only are we buying Mac minis, we're buying Mac Studios that can run a full model on the Mac, the, you know, the $4,000 or $6,000 Mac Studios. Um, but we'll figure it out, right? I mean, you know, listen, there's a limit of my expertise, but you can run 24/7, right? And you can run on mini models and you can run on, um, sorry, I use it, what's the really cheap one from Anthropic? It costs almost nothing. So I think, listen, you can't run Opus 4.6 24/7 and not bankrupt yourself. Maybe you run Haiku 4.5 and you're actually not running it 24/7. You're, you're, you're doing, you're running it every 20 minutes, right? And so you figure out the cost. Haiku is like a twentieth the cost of Opus 4.6. So it's manageable, especially for apps where there's budget. I don't know. I think we'll figure it out.
Well, agreed. That was awesome. I mean, the interesting thing is, is they're going, how does the overlap come between apps for which there's budget and apps, you know, and an agent that has that much power on, you know, your desktop? Because when I mentioned to my IT guy that I'm planning to download and try it, he pretty much had a conniption. So obviously you just have to go and do your own little thing offline, but it will, I agree, it was like, wow, this is the agent on traml with no controls over what it can do on your desktop. I don't think that that's obviously not a sustainable corporate thing, but as you say, Jason, it opened everyone's eyes to what you could do if you had this. So now there's going to be a whole plethora of how do you have it on traml but with some kind of security guard, some kind of controls, and to get, you know, the positives of it without kind of literally having it nuke your entire hard drive or delete everything on your hard drive. But it's the way it's going. It was great.
I mean, I know, like, I mean, one of the things people got excited about in Opus 4.6 was that it was easier to spawn a bunch of agents. You could spawn six or eight agents that would go off and do things at the same time, the human spawn. Um, I know, I'm sure it's true in level two. I know on the next release of Replit, not only does it happen because they've built their own set of guardrails, but the agents do it at night. So in the, I know in the next release of Replit, when you log in the next day, it will have built three to four features for you on its own. That's coming like in the next release in v4 of Replit. So imagine that happens to every app where you go to bed and you wake up in the morning and it's done all your general ledger. It's done all of your whatever, and you can say maybe it'll be slower to accounting this and that. It will have built all of your assets. It will have rebuilt your entire website. That might be a threat to Canva. Well, when you go to bed and you wake up in the morning, it's built four versions of your website. I mean, I know, and I'm not saying others won't do it. I'm saying Vercel and blah blah, but I know this is in the next release of Replit. That's pretty disruptive. And OpenClaw is kind of like the hippie version of that. It's cool. We may not all use this exact product as it is today, but the idea of autonomous agents doing work for us, I mean, it's coming in 2026. And part of it is pretty scary.
In one sense, you're right. Anyone could have done it, but maybe in the big labs they said you shouldn't. And you think this guy just put it out there and everyone...
Yeah. That's why Anthropic threatened to sue them. They're like, 'This isn't safe. This is the last thing we want to do. Get our name off this horrific thing that's going to go and be released.' And just, you know, because when both OpenClaw and Moltbook immediately published everyone's private keys, private passwords, private, they're like, 'This is exactly why we have a safety team. We're going to sue you if you use our name. This is terrible.' But then every single developer thought this was off to, you know, off to the Apple Store or worst case Best Buy. They're buying Mac minis and Mac Studios because no one was more excited to spend nights and weekends building these things. And that's why folks are like, Anthropic fumbled this. I mean, maybe it's just a world like they thought this was the wrong thing to do. Completely unsafe. And Zuck and Sam Altman had another month and a half to think on it and said, this is a movement. Like, this is a freaking movement in AI.
Well, actually, let's, you think Anthropic fumbled the bag by not investing in this, by not trying?
I just think the world, this went from something that is a seemingly goofy guy with the [__] eating grit out of his face and can bench press 400 lb, and it seemed like a goofball guy almost mocking us with this, and your legal team, and it's unsafe, and Dario is about safety, we don't want this on our platform, to like utterly the coolest thing that's been built in six or eight weeks. I just think that everything changed. It's just like investing. I don't think they fumbled it. It was hard to tell at the time. Maybe threatening to sue him wasn't cool. Maybe the legal department could chill on the next one and just instead of the cease and desist, maybe said maybe just chill for another couple weeks. Maybe that was the error. But I think it was too early to know it would be a movement. But it just...
It just—if you talk to CTOs in your portfolio, it just all—whatever it was, it was only like two weeks ago, right? Just all of a sudden everyone was using this two weeks ago. Everybody. And I'm like, "You're not going to like this, man. It's going to delete your database. It's going to steal your credit cards."
And they're like, "But look what it can do."
But it's so cool. But okay, what's the great app that you did? Well, I built a TikTok and it sent an email for me last night. Oh, great.
For what it's worth, I think that's a great take, Jason, because you're right. My initial was, "Oh my god, they at topic funneled it." But you're exactly right. You set yourself up as a safety company. You're extremely careful about what you allow to have happen. That has been your brand and a wildly successful brand. You're right. The idea of some cowboy pretty much half taking your name and doing almost—I can see why they reacted as if they've been stuck, you know, with a stick, right?
The fact that Manis built this in one night said it wasn't about building the technology. You could do it in one day and just—we underestimate how important these guardrails are. They're doing everything from—from, you know, what the agent says to folks that have suicidal thoughts to how they're interacting with your data. You know, the easiest thing in the world is just to take a guardrail off and sell. It's like selling data. Like one of the easiest things to do is sell data you're not supposed to sell, right? Data brokers. Another thing that may be easy to explode is removing a guardrail you're not supposed to remove. You might get a million developers to use that in two weeks.
See Glock for details. Yeah.
Who is the responsible provider of guardrails? Do we think—I don't want to get into like two, but is it, you know, the vendor who's offering the agent? Is it the data holder which holds the data—your Salesforce, your ServiceNow, your—you name it—or is it an independent third party that sits as a layer between?
I think we're learning, right? I think there's a lot of responsibility. I've talked with—I've talked with a lot of the chief AI officers and others at some of those public companies and there is a weight of responsibility on their shoulders for what these agents do that the guy from OpenCloud don't have. Okay, he don't have that weight on his shoulders.
And were they, Jason? Were they tech providers or tech users? Were they customers of these—of the top chief officers at public B2B companies have a lot of weight on their shoulders about—about responsibility for—for their own guardrails which are much more narrower than what OpenAI and Anthropic have.
I mean you take the guardrail off, OpenAI can shoot a gun, you know, hook—a gun up to your LLM and have it—and change the outcome of the LLM and—and say someone's threatening my home and it—and it takes it the other way and it shoots—it could shoot a person. Ask your LLM if it could do that. I can say it's possible, right? You got to have the guardrails, especially if you hook it up to the real world. You—you laugh, but it's not that funny. Take someone—take—take someone that's—that's angry in the world and allow an LLM to control that. Crazy things could happen. So, I think the weight of responsibility is huge for guardrails. It is—it's massive.
Leaving gun comments aside, because I think it's a little farfetched right now. You are right. If I'm a B2B software company and my agent goes from a very constrained agent that I'm selling to third parties, goes from a very constrained agent to something like, you know, the—the—the slightly safer equivalent of, you know, OpenClaw, you're selling a software product to your customers that can exfiltrate all their data. You could make a 24-hour career-ending, company-ending move here.
You're right. So, you're right. I actually, now that I think about it, those guys are bearing it because Harry, to your point, there's two separate questions. Who's—I mean who's—you said who's responsible for this? I actually think it breaks into who's going to be fired if they get it wrong and then what software will there exist to help ensure they don't get it wrong and those are separate questions. My guess the answer to the former is anyone responsible for initiating these, no one's going to care whose fault it is. If you let it into your company and it goes crazy you'll be blamed. The interesting question now this—
I think there's multi-layered to that because lots of—lots of different people can bring in software but CISOs, chief security officers will be blamed also. Like there's multi—
Absolutely. So my guess is even as we speak there's people building really compelling agent-first security products to make darn sure that doesn't happen.
