The AI Code Slop: Risk or Opportunity?

00:00:00The anxiety that I see is if you cangenerate an enormous amount of code andno one is reading it, you don't know thequality of the code. Nobody deeplyunderstands the code base and there'smore fragility, right? It's like theslot problem. Vibe coding slop in myactual production code base, but I thinkthe broader problem that new companycould go solve is like nobody knows howto manage that issue of human attentionto engineering. I think [music] it'slike open season around this reallyreally big problem.

00:00:38Hi listeners, [music] welcome back to NoPriors.Markets are melting down about the endof software. Today Alon and I arehanging out and asking is SaaS actuallydying or are people just projectingfive-person startup behavior onto theFortune 100?We'll talk about what's real, incrediblerevenue growth, collapsing token costs,and faster turnover of vendors.What's just hype and how to size theopportunity?We also discuss the changing bottlenecksin building a software company and someparallels to the internet and clouderas. Let's get into it.>> [music]>> Good to hang. The the market isfreaking out around us, so in all thatnoise, what are you thinking about? Areyou mean the SaaS the SaaS apocalypse?>> SaaS apocalypse. The end of software.>> Yeah.That's kind of interesting. I feel likethere's some meta trends that people aregetting right and then a lot of specificcompanies that people are getting wrong.And so, you know, I think I guess thatbasic premise is that SaaS software andper-seat software will no longer existand everything's going to be replaced byAI and everything's just going to getvibe coded, so why would you pay Xdollars for a Salesforce instance whenyou can just vibe code it internally?And all that stuff strikes me asincredibly short-sighted in the nearterm.I mean the long run, who knows whathappens in 20 years or whatever, butthere's lots and lots of companies thatare quite durable. I think aninteresting example of that where

00:01:54I'mstill a shareholder is Samsara whereyou know, nobody's going to vibe code afleet management app that will then getdistributed through like what? Vibesales? Vibe you know, enterprise salesor something?>> [laughter]>> And you're going to build a vibe likein-cabcamera sensor that everybody willinstall in these fleets?And then you're going to support themusing vibe agents or something. It'sjust it's just very overstated, so Ifeel like it's one of those things wherethere's a massive market correctionaround something thatin the long run has a lot of truth to itand maybe in the short run for certaintypes of companies has a lot of truth,right? Ultimately, I think Decagon andSierra are examples of companies whereyou're moving from per-seat software toutilization-based customersupport-related agents, right? That is areal shift.That may impact some of the prior waveof sort of per-seat software companies,but this isn't going to be every singleSaaS company, soI I I view it as very short-termoverstated. In the long run, who knows?How about you? How do you think aboutit? I mean, I I think the idea of vibeenterprise sales is hilarious um becauseI we have portfolio companieswith you know, hundreds of millions ofdollars of revenue who are verycommitted to as much token usage as wecan, as few great people as we can have.And today, you know, they have less than50 engineers and they went from

00:03:09zero[clears throat] to like let's say closeto 100 sales people very quickly, right?And so it's just a view from the growingAI natives that like vibe sales is nothappening, right? Like a Oh yeah, vibesales is definitely never going it'sit's not happening anytime soon. And soit's just again, all this just seemslike a very strong market reaction andmarket correction.And it it seems like it's veryoverstated especially relative to ahandful of companies that you're justlike, why? Like how will you displacethis company witha coding and you know, in the fleetexample, you're not going to have thefleet managers like writing their ownapps to do all this giant surface areaof stuff. It just doesn'tit's just not going to happen in theshort run.>> think a lot of it is actually driven byumsome assumptions that, you know, personaclose to my heart, but engineers andbuilders are making about like the restof world, right? I think there's thisthere's this implied belief that likeeveryone will want to make their ownsoftware.And I think it's like probably>> meansoftware is eating the world? Is thatwhat you're trying to say?>> I'm not I I think like we're we're stillTime to build Sarah. Time to build. Idon't think that everybody wants to maketheir own software. I think some set ofpeople will want to make it and otherswill want other people to do it for themand like sometimes like what's a what'sa like

00:04:24if you think about a good exampleof this, engineers sometimes have a likemy personal laborfocused picture of the world. So, if youlike should you build Jira in mostengineering organizations? Like is thata Yeah, it's not a it's not the best useof your time if you're focused onproduct. I mean, the other piece of itisum the examples that people use, oh myfive-person startup built our own CRM,vibe coded it, blah blah blah. Yeah, ofcourse. I mean, before that you just didit all on a spreadsheet and that wasfine, too. You didn't have to vibe codeanything. And so for very limited nicheapplications where it's a technical teamdoing something really quick cuz it'suseful and custom and bespoke andamazing, of course that's going tohappen.Does that mean that a Fortune 100company is going to displace their CRMwith some internal thing they got vibecoded over the weekend? Probably not.And so I think it's also extrapolatingor projecting behavior of very smalltechnical startups onto the world'sbiggest enterprises.And that's the second thing people aregetting wrong is they'remisunderstanding thethe moment. And I think the internalsoftware stuff that people are buildingis amazing, right? It's not like itisn't impressive that you can do that.It's incredibly impressive.It's just extrapolating that behavior soaggressively so early just doesn't makethat much sense right now.

00:05:39I think toyour point of like the five-personcompany versus the very largeenterprise. If you ask that sameengineer who's like pissed about paying$10 a seat for Jira,like if you asked him or her, like doyou want to do the change management inBank of America of getting everybody todo this the way you think is right?And then dealing with all the securityconsiderations and managing otherpeople's opinions about potentialchanges to the story management workflowand then maintaining the system, theanswer is like probably not, you know.Um and so I I think it's it is focusedon um I actually think the idea thatactual production of code becomes notthe bottleneck forum if you know what the spec is, not thebottleneck is like incrediblyinteresting, but I I I do think itoverstates like how muchuh of the overall software vendorproblem that is. Yeah, I think peoplealso misunderstandhow much demand exists for softwareproducts. And by software products, Imean everything. I mean AI, I mean>> Is software eating the world? Is AIeating the world?AI is AI is eating the world, so I thinkthat that is actually true.And I think Mark's post on that wasreally um thoughtful andforward-thinking on it all. I think thatfundamentally

00:06:54umyou know, there's there's so much demandfor softwareand there's so little supply ofengineering in reality relative to thatdemandthat as you add this enormous boost ofproductivity to software engineersum it just gets sucked up,right? Because there's so much morestuff to build and to do.And I don't see teams, you know, startupteams continue to hire engineers for areason, you know. I think the nature ofthe work is shifting and I think somepeople are going to have real issueswith that shift.Because fundamentally, you're shiftingfrom you know, in some cases, you know,there's the there's a few differenttypes of of mindsets around engineersand one of the mindsets is the reallybespoke craftsmanship,you know, I'm going to make I'm going todo the aesthetics of the thing that I'mdoing really well and I care about thecode quality and you know,and umthe the artisanal version of what I'mdoing.And then there's people who write codebecause it's a utility that allows themto build product.There's some people who really likeaspects of the math or you know, there'slots of different motivatorsfor people to write code and I think asubset of those people are going to beuh less happy in the new world. It'skind of like the indie game developerswho make these handcrafted individualgames for themselves and then for theirfriends and then they launch them on the

00:08:10Apple Store or whatever um versus thepeople who'd work at EA. And they eachhad their own version of craftsmanship,but it was just a different type ofthing. I think we're going to see a lotof these really great engineers who careabout the thebespoke craftsmanship of everything theydo, they're going to be unhappy workingat larger companies as these codingtools get even moreaccelerant because it goes against theirapproach of of how they like working andwhat they enjoy out of the work. And forother people who are really focused onutility of just building product, it'sgoing to be freeing in some ways. So, Ithink there's also like a variance interms of the reactions to this stuffdepending on the type of uhutility function that you have relativeto the work you're doing.Yeah, I I think related to that, the umone thing I've seen is that if you havean engineering identity that's based onthe like a value-based ranking ofdifficulty or skill, like the thespecific types of engineering that areconsidered, you know, impressive or highstatus can actually be like less hardfor agents, right? So, I think there'san enjoyabilitylike element and then an identityelement. Um and actually one of yourfounders um from Applied Intuition wrotea good blog post where there is a uman essay where he says like keep youridentity

00:09:25small. I think that's likewonderful overall advice for this periodof time, you're like more adaptable ifit's true.But I I think your overall view of thereare a lot of unsolved problems and likemaking an abundance of software canbetter address that, I I strongly agreewith and what one thing that umactually is near and dear to theaudience that is really unsolved is likewe've broadly been thinkingabout what happens if you have abundantcode generation and in like I think inall of our teamsagent first engineering management andthink about code quality is an unsolvedproblem. Yeah, and we'll get there. Iknow they are going to work and we'llget there.What do you view as the major problems?Um well, the the anxiety that I see umislike if you can generate an enormousamount of code and no one is reading it,you don't know the quality of the code,nobody deeply understands the code base,and there's more fragility, right? It'slike the slop problem, but instead of itbeing like vibe coding slop for randomwebsites for non-technical people, it'svibe coding slop in my actual productioncode base forevery lazy engineer, which is everyengineer. I think people are likelooking at some problems

00:10:40of actually dothink ticketing ticketing systems areare like at risk, but I think thebroader problem that Jira could go solveor new company goods could go solve islike nobody knows how to manage thatissue of human attention to engineering.And there's a bunch of ideas, liketesting and like, you know, smartreview, just let agents do it, formalverification. But I think it's like openseason around this really really bigproblem. I think the one other thingpeople are bringing up that I don'tquite buy is that um agents are alreadymaking likebig decisions for vendor purchases andthings like that. I think somebody nearand dear to your heart could say aboutthat and umuh I think that uh there the thestatement was oh agents are increasinglymaking decisions about what softwarepeople are using.And really what that is is while youhave a partnership, your Cognition oryour Claude or whoever, and you have apartnership and it's part of thatpartnership, you spin up a super basicinstance and you use very specific toolsbecause you have a partnership to dothat. And that's always happened, right?If you're using Airtable and they're onAWS, like you're spinning up an AWS AWSinstance without knowing about it,right? In the background. So, I alsothink that whole notion that in theshort run agents are making thesechoices is alsooverstated. I think in the long run it'strue, but then you get into all sorts ofagent

00:11:56commerce decisions and do theyunderstand your persona and what youactually want and need and all thisstuff. So, I just feel like we're in alittle bit of a noisy moment wherepeople are kind of potentially and I'msomebody who's very pro AI progress anda believer in all the changes that havehappened and are coming, but I thinkwe're having a lot of overstatement nowof what's actually happening in theworld and part of that is assessapocalypticand this giant re-correctionand part of it is umyou know, extrapolating that the futureis here already when in many cases let'sjust say we did a BDDL or whatever. So,I just think people kind of need to oryou know, the multbook stuff whereyou're like, "Yeah, well, that seemshuman generated." You know,in terms of the emergent behavior. So,I don't know. We're we're we're in thisodd moment where I feel like this wasthe month of hypein a way that we haven't seen in a whilewhere a a bunch of stuff got overstatedin all sorts of ways and people believedit. And by people I mean like mainstreammedia and others are like, "Oh my gosh,look at this behavior ofyou know, these agents trying to cut outhumans from their forum where it's kindof like and blah blah and you're like,"Okay, like maybe you should see wherethe posts are coming from in somecases." And it's exciting, by the way,don't get me wrong. I think there's veryexciting behavior that's happening. Ijust think, you know, a subset of it wasplanted for marketing purposes.>> Yes, certainly. I think people are alsofiguring out like there there are thingsthat tap into

00:13:11um deep emotionalreactions that people have to their viewof likethings that feel very human, right? Umfrom a marketing perspective and likethat's [clears throat] clearly one ofthe things that's happened around them,the multbook stuff. I also think thatlike one of the things I actually thinkhappened was likethe idea that demos are different fromthe reality of the full software thatyou need like has not quite arrived inmany of the equity research people'sdesks, right? And so, like I'm like,"Guys, like your whole job was to thinkabout these like the structuraladvantage of your businesses and what isgoing to compound and the theory ofcompetitive advantage didn't just likepoof disappear, right? Likesoftware markets have been a fight abouthow to do things and how to distributeto customersas well as a battle of how to producecode for a long time. So, I umI feel like that has been missed alittle bit, but I I do think long runthe the fundamental thing that thebottleneck on production ofyou know, expensive to produce softwareuhbeing loosened is really cool, right? Itjust means like if you think of there'sa lot of embedded points of view insoftware

00:14:27on how to solve a problem,right? You know, if it's engineering oruh enterprise sales, not very software-yproblem, or or general productivity,right? Like Notion is a way to dothings. It's building block system, butit's definitely got a a point of view.And so, if youreduce the cost to express that point ofview in software, I think it's cool thatwe're going to like see a lot moreideas.>> That's amazing. And again, I think it'sa revolution. So, I don't get me wrong.I'm I'mI've been involved with coding companiesreally really on andI'm very excited about everything that'shappening. And I think it'stransformational and I think it'srevolutionary and I think it's reallyimportant.I just think we had a month of kind of hype.Okay, so if we ignore the noise of thelast monthwhere people got a little like frantic,what do you think is a signal thatpeople are not paying attention toenough in such anoisy landscape? You were telling methat like growth growth pace is like ofthe of the biggest companies is is stillunder underpriced. Yeah, one thing thatum Jared on my team put together that Ithought was super interesting was um hepulled data from uh Capital IQ wherethey just like predicted someprojections on OpenAI and Anthropic andthey looked at um and then he sort ofgraphed

00:15:42out and maybe we can share thesegraphs as part of this episode.He graphed out um how long it tookdifferent companies in years to go froma billion in revenue to 10 billiondollars of revenue. So, for example, ADPtook 20-something years to grow from abillion to 10 billion in revenue. Andthen the next wave of companies likeAdobe took about 20 years to go from oneto 10. And then you fast forward in timeand you have things like Salesforce orSAPs or an even more modern cohortand they took eight or nine years.Microsoft took, you know, seven-isheight years.Google and Meta and AWS took a coupleyears, you know, three, four, fiveyears. But the AI labs did it in roughlya year.Right? And then if you look at theprojections that>> a wild chart.>> are sort of the public>> Yeah, it's a wild chart. And so, weshould we should add it, right? But youjust see it go from like 20-somethingyears with Adobe to like a year for theAI labs. And then if you look at theprojections that are sort of the publicprojections, they aren't necessarily thecompany-driven data, but the publicprojections onwhere the labs will end up or how longit'll take them from to go from 10 to100 billion in revenue, for Microsoftthat was something like uh 27 years.For Google it was over a decade, samewith the AWS, roughly the same for Meta.And then for the AI labs it's likethree, four, five years, you know, it'svery

00:16:57fast.And so, we're seeing the fastest time toreal massive revenue that we've everseen in the history of software. It'sjust these insane curves and again, weshould just post them. Part of that Ithink is just the internet has createdthis global pool of liquidity and yousuddenly have customers online. It'smuch easier to distribute than it's everbeen. So, that's one piece of that.There's more people with access. There'shigher GDP. There's lots of drivers forthat. But then simultaneously you'rejust creating enormous um business anduser value at massive scalesimultaneously and these capabilitiesare so richthat you're seeing this take off interms of revenue. And so, it's it's it'sunprecedented. It's really impressive.And I think people are ignoring therevenue and usage side of the equation.Um the other thing that we actually puttogether was the collapse in tokenpricing for equivalent models. I thinkthis was done initially by David whoworked for me and then Sharan.And so, for example, we looked at thecost of a GPT-4 level or equivalentmodel. Uh we looked at that a year ortwo ago and basically in 21 months itwent from like 37 bucks for a milliontokens to 25 cents.And so, you know, pricing dropped by150X in 21 months.And then we tried to extrapolate thatcurve, but obviously people aren'treally using GPT-4 level models anymoreeven though, you know, they're two,three years old. And so, we looked at 01equivalent models and the cost of amillion

00:18:12tokens on an 01 equivalent modelin December of '24 was about 26 bucks.And then in November of '25 it was 30cents. So, we saw another 88X drop, not88% or 88, you know, 88 88 times cheaperin 11 months for that next generation ofmodels. So, we're having pricingcollapseon the token side while we're havingrevenue ramp insanely on theusage side. And so, that's insane if youthink about that. Just this pace ofshift of cost, of revenue, ofutilization, of everything.And this is back to like I'm incrediblybullish on everything that's happening.Um it's just more dismodulating itagainst this, you know, this oddover-extrapolation of what's actuallyhappening or actual capabilities or, youknow, what these things are reallydoing. Yeah, I I think one thing thatpeople miss in the like bear case andall this stuff is as you said likerevenue numbers, which is hard to miss.Um but but hey and then umuh just like actual um like tokeninference count, right? If you look atone if you look where's the inferencehappening? It's either happening ininference clouds, right? Base 10 modelor fire up, or it's happening at thelike the very large model provider.>> happening up here. And it's happening ina lot's brain, which is still much moretwo magnitudes

00:19:28more efficient, right?>> humanity in general. In humanity ingeneral, yeah. Yeah, that's true. Interms of power utilization, human brainis really impressive. What is it like?Tens of watts? 20 watts? How much likewhat's the power utilization of a humanbrain? I want to look it up right now.It is it is two magnitudes It's like 10or 20 watts, I thought. I think to thepoint of like real datathe inference clouds are growing athousand X in terms of consumption,right? And then they're getting moreefficient, so revenue grows at somelower rate than that, but it's wild.That's 12 to 20 watts of power, which iscomparable to a dim light bulb or acomputer monitor in sleep mode. It's noteven like a computer monitor. It's whenyour monitor is sleeping, that's theamount of energy that your brain isconsuming as it does all these crazycalculations. It's one blade of one GPUfan in one of these data centers. That'swhat I think of it. It's nuts. I feellike Nagesh used brain that was probablyconsuming like a thousand watts. Well, Ithink that's great. I think like we havea lot of efficiency work to go.>> [laughter]>> I I kind of meant it the opposite, youknow, he's so smart he's probablyconsuming more energy, but to your pointmaybe he's more energy efficient. Oh.Maybe he's at like one watt and I'm likeat a thousand watts or something. Imeant for the computer.We're all stuck without the you knowbrain computer interface work improving,but I'm I'm just interested in how

00:20:43muchefficiency we can get out of the models.Yeah, it's probably obviously just basedon the human brain there's a lot ofroom. You know, one thing I do thinkabout I I was talking toa friend who leads a bunch of purchasingat a traditional large enterprise thismorning and he was like, oh well thelike incumbents can this whole thing isoverstated. We're so committed to allthese big enterprise vendors, whatever.A lot of things that we've been talkingabout hereand his other view was that theincumbents have the money to buy and golike fight back on these dimensions. Ione thing I immediately thought of wasjust likelikereflexivity in markets is such a goodconcept and here it's like well theythey do unless they don't have themarket cap to do it, right? With thesecompanies that to your point, you know,first the labs, but then a series of thevery best application companies, ifthey're growing toa billion of run rate rapidly andvaluations grow in concert with that,then I do think there's a there's aquestion on whether or not you youhave the currency to compete too. Yeah,I'm already seeing that in the supposingmarket, right? whereSF housing is starting to rise again inpart due toI'm assuming

00:21:58outcomes from the labtenders and things like that cuzsuddenly you have these companies thatare worth hundreds of billions ofdollars out of nowhere in a few yearsand as employees you're selling in thetenders, there's this new sort of influxof cash in the ecosystem.So that's also in video going from youknow, tens of billions or a hundredbillion to trillions in market cap. Likethere's just this shift happening rightnowin terms of scale. There's aninteresting question actually where thisis one other thing that we looked at asa team and maybe I should just publishall these slides.We basically askedwhat proportion of GDP is tech,right? And and just the US economy atleastand how has that grown over timeand also like what does that meant interms of market caps, right? And so ifyou look back to 2005,Google was worth a hundred billiondollars and Exxon was the world's mostvaluable company at 400 billion dollarsmarket capand thenit took until 2018 Apple was the firstcompany with a trillion dollarmarket cap, right? Ever and everyone wasshocked that anything could get to atrillion and at the time techrepresented about 30% of the S&P. Beforethat it was say you know10% ish back in 2005

00:23:14and now the topeight tech companies are about 23trillion of market capand they make up well over 50% of theS&P in terms of value.At the same timethey went from basically 4% of GDPin 2005 to about 12% of GDP today. Andso then the question is how whatproportion of GDP eventually becomestech? And AI is a driver of this, right?Because you're taking services andyou're takingcertain types of jobs and you'reaugmenting them with AI and you'reconverting them into effectivelysoftware spend or tech spend.And you can make different assumptionsabout your the race and then based onthat, you know, you can end up withanywhere between15 20% of GDP to you know 30% of GDP in2035.But that means that the market caps ofthese tech companies get even bigger,you know, it's kind of a metric for howbig can these things actually getas they sort of aggregate up portions ofGDP. So I think that's the other lensthat people aren't really thinkingenough about in terms of what what arewhat are some of these terminal values10 years from now? Like how much morecan things growand what are your assumptions aroundthat basis for growth, you know, andthis is back to like that ramp up intorevenue. So it's a very interesting kindof set of questions that we're we'vebeen asking on my side just in terms oflike these meta things, you know, likewhat are the what are the bigger

00:24:29trendsthat people may not be paying attentionto that may be super interesting. Okay,well then I have a set ofstructural questions about how to investbased on this for for you because I youknow asking for a friend, my funds aresmall.I think there's like good implicationsand bad implications based on what yousaid. Like one might be if everything'sgoing to get a lot bigger,a billion dollars is no longer latestage, right? It's like just you knowtake a marker on valuation that's likethe beginning. Well even now it's notlate stage because people are raising ata billion dollar valuation with two twomillion revenue.Well, you can decide that's I know atleast one company like that. You candecide whether that's a like a smartidea or not, right? Butbut you know the point we wouldabsolutely agree on I think is just youknow the runway for some of thesefoundational companies is just muchlarger, right? Thenthan the conventional wisdom. I thinkwe've already believed that though. LikeI think everybody shifted. I remember Iread a blog post like 15 years ago orsomething 10 years agothat basically talked about how hard itis to get to a sustainable five billiondollar market capbecause at the time there was a smallbasically once every couple years acompany would actually get to that andstick with it because this is back toyou know 10 15 years ago the biggestmarket caps were in the hundreds ofbillions at most and low hundreds ofbillions, right?And

00:25:44then we saw everything grow 10x overthe last 15 years, right? You suddenlyhave trillion dollar market capsand that means there's a lot morecompanies also worth a hundred billionthan there used to be in tech. So Ithink in general we've seen these shiftshappening already and that the reasonthat we were asking the questioninternally about how much bigger canthese things get is because that hasfurther implications. How many moretrillion dollar companiescan be supported?Is it two? Is it three? Is it a dozen?Is it50, you know?And relatedly like if everything getspulled up, how do you think about howyou invest over the lifetime of acompany in general or how do you thinkabout that as a founder in terms of thethe end state and then also there's arelated question of what's the actualfail rateof startups? Should the fail rate go upor down in that world? And you couldargue it either way. You could arguethat the fail rate should go up becausemore and more value is gettingaggregated in a platform as liketraditionally has happened,right? Every single platform shift hasseen a commensurateforward integration of that platforminto the most important verticalapplication. So as an example,you know, Microsoftvery famously on its OS forwardintegrated into the office suite Exceland PowerPoint and Word, right? Theykilled or bought companies in thosemarket segmentsand that became office and then theyredistributed it alongside

00:27:00the OS orGoogle forward integrated into verticalsearches. They had a platform and thenthey built out travel and they built outlocal and they built out all thesethings. And so it's not surprising thatthe labs will forward integrate into themost interesting applications on top ofthat. You're already seeing thatpartially with code.But what else is coming there and thenwhat implication does that have forpeople running startups, right? Likewhich of those verticals are are durableand defensible and which of those aregoing to get eaten by the labs?And so you know, you can make argumentsin both directions in terms ofwill more of overall GDP aggregate intoa smaller number of companies, which isalready what's happening, right? Justignoring the labs even, right? Thatthat's kind of what happened with Amazonand with Google and all these things.Or do you end up with this broader taileffect as well where things are kind ofhappen simultaneously?We also have a lot more startups thatare worth more because there's just somuch more market cap to go around, butalso the internet continues to providethis global liquidity. To meI think the tail dominates because thesurface area of what you can addresswith technology is just increasing morerapidly. But maybe to add more nuance tolike a billion dollars is But it's nottrue. So if you actually look atmarket cap,it's very much power law, right? It'sthe head and torso aggregate

00:28:15almost allthe value. That's actually true ofcustomers too although that people tendto misunderstand that.Even for things like Google where therethere was I remember that book that waslike the long tail or whatever theinternet and the claim was the long tailreally matters and then you'd add upGoogle's ad revenue and you're likeactually it's all the head and torso,right? And so I feel like there arethese head and torso effects that keepgetting ignored. It's like Paul Graham'spower law on startups, right? Most ofthe value of YC is probably fivecompanies. Like 80% of it is I'm makingit up, right? But it's reallyconcentrated. And so why would thatchange in this era? I don't I don'tthink it changes in this era. I thinkthat it depends what your measure was.If your measure is how many hundredbillion dollar businesses are there, Ithink there's a lot more, right? Like itit doesn't mean there fewer hundredbillion dollar businesses. Actuallythere are more because the surface areais growing. And at the same time likethe distribution of how much is in thehead is probably the same. It those areeven bigger. Yeah, it's possible. Yeah,it's an interesting question. You thinkfor investing like there's a thingthat's good for me and then perhaps likebad for me or just a question for thefor thecontinued growth stage investors, thetime to market leadershipand to revenue scale I think iscompressing. I mean it's not I thinklike this is happening. We havea large handful of companies that havegone zero to a hundred million

00:29:30plus runrate faster thanSAS companies that we'd seen 10 yearsago.And so valuations have grown with that.I think some set of companies that looklike this,they are durable and some likeleadership can still flip, right? Like aquestion might be, you know, is it youor is it Ant or is it Open AI over timeto your point of like actually you couldgrow to a billion dollars of revenue andstill face that question.And and and that is I think a risk thatmaybe some other growth ecosystem wouldfind as a new thing versus like categoryleadership at a certain scale feltunassailable like 10 years ago. Yeah,and I think there's two interestinghistorical precedents to this. One isthe internet wave where, you know, 1999450 companies went public, 2000 andanother 450 went public. And so therewas say one to 2000 companies wentpublic during the internet age.And maybe a dozen to two dozen of themare still relevant, right? Everythingelse roughly died or got bought.And then you fast forward 10 years andyou saw this assumption of things thatpeople thought were unassailable, right?In social networking people thoughtFriendster and then MySpace wereunassailable and then Facebook won.In payments, I remember when I investedin Stripe everybody said that why areyou doing this, you know,

00:30:46umBraintree existsand PayPal exists and all these thingsexist and so, you know, why would youever invest in another payments company?And of course that ended up being thewinner.Um or one of the winners, right? I meanpayments is so big it's a fragmentedoligopoly. Umbut I just feel we've kind of seen thisstory before and so as a founderit's really useful to be asking abouttwo things. One is what is thedurability of your business?And number two ishow should you think about when to exitif you're going to exit because oftenfor companies there's about a 12-monthwindow where your company is the mostvaluable it'll ever be and then itcrashes out. For a very small handful ofcompanies the answer is you should neverever ever sell. For most companies theanswer is you should sell when thetiming is right and then the question ishow do you know when the timing isright? Because ultimately you're goingto hit a point of of maximal value andthen and then it has a real potential todie even if it got enormous traction andthat was the internet waveof the '90s. And so I think too cheappeople are thinking about this and onetip for foundersis from a hygiene perspective but alsojust a way to make it a non-emotionaldiscussionis pre-schedule once or twice a year theboard meeting where you talk aboutexits.And that way it becomes non-emotional.It's not about we're going to exit. It'snot like we should exit. This actuallybeen Horowitz's advice I think um fromwhen he was running

00:32:01Opsware.You just set up a non-emotional meetingonce or twice a year. You're like,"Nope, still not time to do it." Or yousay, "Oh, you know what? Actually thecompetitor dynamic has shifteddramatically.Somebody's come to us with an offerthat's higher than anything we'llachieve over the next 5 years. Now's thetime to do it." Right? And I think it'suseful for you to be thoughtful aboutthat and again the default for a smallnumber of companies is never ever do it.For almost everybody else it's worthconsidering it at one point or anotherbecause you may otherwise get stuck withsomething that isn't working for a longtime or you may get crushed by acompetitor.And many many years of very hard workcan just go down the drain. I think thisis a an interesting point about thecomparison especially to like theinternet age versus the SaaS I don'tknow what you call the the the likecloud age from the last decade as beingmore similar because there were I wasnot around for this era but from from myum research and from working with abunch of people in that period you'renot old enough for this era either. LikeAOL was the internet for a moment,right? Yahoo was the web's front page,Netscape was the browser, InternetExplorer was the web runtime, eBay wasthe market. Like I I think there are anumber of these [clears throat] And thenAOL exited at the exact right moment toTime Warner, right? At their peak, theirpeak valuation. Right. And I I do Ithink that people founders and investors

00:33:16may um over-rotate on the SaaS era wherelike it did feel like at a certain scaleum like internet era there's a period oftime where like growth was the default,right? Growth at a wild speed that wasnot true in SaaS land. And so it wasmore like you know, incremental andbeyond a certain scale it felt veryprotected but I umI think that this probably does lookmore like the internet era where thequestion is like does that growth likedoes it compound to a control pointwhere you're a very special company orlike do you actually think about exitsin a different way? Yeah, and if youeven go back to the '80s, you know, youhad Lotus. I don't know if you rememberthis company Lotus.>> have implemented Lotus 1-2-3 at anenterprise business as an intern. Yeah,so wow. So Lotus built one of the firstspreadsheet products.And it grew explosively. I it got intothe hundreds of millions of revenue likereally really fast and this was the'80s, right?And then a couple years later itbasically collapses into the arms of IBMand Microsoft launches Excel and takesthe whole market roughly, right?And so again it looked like a verydurable business. It was the the thekiller appon on computers, you know, for its era.And then it just died. It didn't die.

00:34:31Itit ended up with a great exit to IBM butstill it is it no longer exists, right?In reality.And so I think the same thing is goingto happen for a number of companies ofthis era and the question is whichcompanies?That's a really hard question, right?Who knows?But for some companies you're startingto see cracks.Right? And so the com for the companieswith these cracks as the marketstructure shiftsas you see shifts in what the labs aredoing, as you see shifts in usage, asyou see shift in differentiation anddefensibility and all the restit's a good time to ask, "Hey, is thismy moment?Are these next 6 months when I'm goingto be the most valuable I'll ever be andthen I'm at real risk?" And if so, youknow, you should think seriously aboutwhat to do with that. And I view thisnot just I mean right now. I mean every6 months there's going to be theseshifts that are worth considering andthat's why it's like pre-schedule theboard meeting so it's not emotional.You're not putting something on theagenda and everyone's like, "Oh my god,you want exit? What's going on? Are youupset? Are you worried?" It's more like,"Oh yeah, we booked this 6 months agoand we booked it a year ago and webooked it 2 years ago." Whatever it is.And this is just when we talk about thisstuff. So we can just have a verylogicalemotion-drained conversation around thisstuff. And maybe I think, you know,again in comparison to internet era asto like why think about it more now isWell, people in the internet era shouldhave thought about it Sure. Sure. I meanMark Cuban did

00:35:46this. Mark Cuban's claimto fame is he sold a a company that thatyou know, let's let's put this way itwas early in terms of product.And he sold it to Yahoo for a fewbillion dollars and he collared Yahoostock so that as the stock dropped hedidn't lose any money. It was one of thebest all-timefinancial engineering moments in techhistory, right? That's what made MarkCuban a billionaire was he sold atYahoo's high water mark and then he keptall the value as it collapsed in price.That was one of the few people who didthat at uh during that era but peoplewere thinking about it.>> I think what most people missed, right?Um and like in retrospect like thinkingabout the flips that made it happenwhere the ground was moving a lot um isuseful, right? Because you have toanswer the question, am I that companyor not? Um or is my acquirer thatcompany or not? And like in the internetcycle you had new distribution, newperformance, new interfaces changinguser behavior. It was just likeeverything happening all at once and newexploration. Not true in cloud land,right? Just more replacement market andthen like niches that you could cheaplydistribute to. New business model. SaaSis amazing. Um but in AI it's like,"Okay, is is the next major capabilityjump from the labs going to screw me andreset the leaderboard?" Like that is animportant question to ask yourself. Andthen also Yeah. Um like surface areaquestions, right? Like agents versus

00:37:01IDEs, voice as a default. Like therethere are things that change in product[snorts] experience that also couldreallocate power. The best way to defendagainst this is to build a bundle.So it's to build a multi-product surfacearea for your company so that youcross-sell multiple things into the sameorganization and you become a defaultpart of the workflow. And that's that'sthe best way to defend against thisbecause then you're being used for fiveor 10 different aspects of of thatvertical that you're in or thatapplication that you're in versus here'smy singular thing that's easy to cloneor copy or for people to kind of umdisplace. So I think um the the sort ofdefensive advice on that is do that.>> Yeah. Bundles are often seen asoffensive but I actually think they'reamazing for defense, you know? And so Ithink that's the other thing that peopleare under-doing a lot but for some ofthese vertical applications that's goingto be the way to win long term or todefend long term. Well, I actually stillthink I I now I sound like I just hatelike the

00:37:56era and like apply now without thinkingabout it where it was like, you know, doone thing well.Yeah, it was do one thing well and thenpeople buy you and then like don'tcompete with a million things but, youknow, we we think of That was badadvice. That was always bad advicethough. I mean it it substantiated okaySaaS companies was bad advice becausebefore that the prior wave of companieswere very acquisitive and verymulti-product.And it was just the SaaS era where itbecame a singular thing. I think theother piece of it is umthe rate of change and velocity of thetechnology during the SaaS era was justslow. It's just like, "Let's just keepbuilding out the internet."You know, that was kind of SaaS era.Right? And so the the difference with AIis the velocity of change is so highthat what normally would have taken adecade and you'd have a normaldecade-long displacement cycleis now happening in a year or two. Andthat's really the the reason that thesethings are so turbulent.It's because the technology is shiftingso so dramatically so quickly and that'sjust part of scaling laws and that'spart of reasoning and that's part of allthese things that have, you know, allthe post-training stuff that's beenrolled out. Soum there's just been so much innovationin such a compressed period of time thatthat's the reason things are turningover and things that normally would havetaken a decade are happening in a yearor two. And that's why we're seeingthese displacement or potential fordisplacement cycles. But that also meansas a founder your mindset

00:39:11should shiftinto this new world framework. Youshould say, "Okay, if every 2 years is10 yearsI need to think really quickly on uhchanges that are happening. I need toreact to them in all sorts of ways."And so it's just it's just uh back toyou know, it's a it's a fun andinteresting and exciting time and Ithink I think it's going to be anamazing decade of transformation.>> Yeah. I I do think um maybeone way to think about like a lot of thedefenses that people did not in thesoftware era or uh the last software eraor like, "Okay, well, what does notdepend on, you know, my little featureset just incrementally growing, like,platforms, ecosystems, networks,bundles, even hardware, like youdescribed with Samsara, like, that feelslike non-trivial control points. And so,maybe the takeaway for me and a lot ofHangout today is like, hey, don'tover-rotate on the last month, but alsoyou have to think about when, you know,what be intellectually honest about theposition you have in market and then inthe speedofuh change era,actually think about what the controlpoints are.Yeah, let's come back. Let's shift in.It's going to be fun.>> Okay, have fun.Yeah, see you later.>> [music]>> Find us on Twitter at No Priors Pod.Subscribe to

00:40:27our YouTube channel if youwant to see our faces. Follow the showon Apple Podcasts, Spotify, or whereveryou listen. That way you get a newepisode every week. And sign up foremails or find transcripts for everyepisode at no-priors.com.