Skip to content

Podcast

From SaaS to AI-First- How Companies Are Reshaping Innovation

Technology | Startups

Source ↗ ← All highlights
  • SaaS Fear Is Over-Extrapolation
    • The current “SaaSpocalypse” narrative over-extrapolates small startup behavior to large enterprises.
    • Many durable SaaS businesses (e.g., Samsara) are unlikely to be replaced by quick internal “vibe-coding.” Transcript: Sarah Guo Hi, listeners. Welcome back to No Priors. Markets are melting down about the end of software. Today, Alad and I are hanging out and asking, is SaaS actually dying or are people just projecting five-person startup behavior onto the Fortune 100? We’ll talk about what’s real, incredible revenue growth, collapsing token costs, and faster turnover of vendors. What’s just hype and how to size the opportunity. We also discuss the changing bottlenecks in building a software company and some parallels to the internet and cloud eras. Let’s get into it. It’s good to hang. The market is freaking out around us. So in all that noise, what are you thinking about? Elad Gil Oh, you mean the SaaSpocalypse? The SaaSpocalypse. Sarah Guo The end of software. Elad Gil Yeah, that’s kind of interesting. I feel like there’s some meta trends that people are getting right. And then a lot of specific companies that people are getting wrong. And so, you know, I think, I guess the basic premise is that SaaS software and first-seed software will no longer exist and everything is going to be replaced by AI and everything’s just Going to get vibe-coded. So why would you pay X dollars for a Salesforce instance when you can just vibe-coded internally? And all that stuff strikes me as incredibly short-sighted in the near term. Over the long run, who knows what happens in 20 years or whatever, but there’s lots and lots of companies that are quite durable. I think an interesting example of that where I’m still a shareholder is Samsara, where nobody’s going to vibe code a fleet management app that will then be distributed through like What, vibe sales? Enterprise sales or something. And you’re going to build a vibe like in cab camera sensor that everybody will install in these fleets. And then you’re going to support them using vibe agents or something. It’s just, it’s just very overstated. So I feel like it’s one of those things where there’s a massive market correction around something that in the long run has a lot of truth to it. And maybe in the short run for certain types of companies has a lot of truth, right? Ultimately, I think that Gagan and Sierra are examples of companies where you’re moving from perceived software to basically utilization-based customer support related agents, Right? That is a real shift. That may impact some of the prior wave of sort of perceived software companies, but this isn’t going to be every single SaaS company. So I view it as very short-term overstated in the long run. Who knows? How about you? How do you think (Time 0:00:38)
  • Change Management Is The Real Bottleneck
    • Production-level change management, security, and scale remain major barriers to internal replacements.
    • Engineers who dislike change management won’t rebuild enterprise workflows despite better coding tools. Transcript: Sarah Guo Think the idea of Vibe Enterprise Sales is hilarious because we have portfolio companies with hundreds of millions of dollars of revenue who are very committed to as much token usage As we can, as few great people as we can have. And today, you know, they have less than 50 engineers. And they went from zero to like, let’s say close to a hundred salespeople very quickly, right? And so it’s just a view from the growing AI natives that like Vibe Sales is not happening, right? Elad Gil Oh yeah, Vibe Sales is definitely never going. It’s not happening anytime soon. And so it’s just, again, all this it just seems like a very strong market reaction and market correction and it seems like it’s very overstated especially relative to a handful of companies That you’re just like why like how will you displace this company with coding and you know in the fleet example you’re not going to have the fleet managers like writing their own apps To do all this giant surface area of stuff it just doesn’t it’s just not going to happen in the short run i think a lot of it’s actually driven by um some assumptions that you know persona Sarah Guo Close to my heart but engineers and builders are making about like the rest of the world right because there’s this there’s this implied belief that like everyone will want to make their Software. And I think it’s like software is eating the world. Elad Gil Is that what you’re trying to say? I am not. Sarah Guo I think like we’re we’re still time to build Sarah. I don’t think that everybody wants to make their own software. I think some set of people want to make it and others will want other people to do it for them. And like, sometimes, like, what’s a, what’s a, like, if you think about a good example of this, engineers sometimes have a, like my personal labor focused picture of the world. So if you, like, should you build JIRA in most engineering organizations? Like, is that a… Elad Gil Yeah, it’s not the best use of your time if you’re focused on product. I mean, the other piece of it is the examples that people use. Oh, my five-person startup built her own CRM, vibe-coded it, blah, blah, blah. Yeah, of course. I mean, before that, you just did it all on a spreadsheet and that was fine too. You’d have to vibe-code anything. And so for very limited niche applications where it’s a technical team doing something really quick because it’s useful and custom and bespoke. Amazing. Of course, that’s going to happen. Does that mean that a Fortune 100 company is going to displace their CRM with some internal thing they got bi-coded over the weekend? Probably not. And so I think it’s also extrapolating or projecting behavior of very small technical startups onto the world’s biggest enterprises. And that’s the second thing people are getting wrong is they’re misunderstanding the moment. And I think the internal software stuff that people are building is amazing, right? It’s not, like, it isn’t impressive that you can do that. It’s incredibly impressive. It’s just extrapolating that behavior so aggressively, so early just doesn’t make that much sense right now. Sarah Guo I think to your point of like the five-person company versus the very large enterprise, if you ask that same engineer who’s like pissed about paying $10 a seat for Jira, like if you asked Him or her, like, do you want to do the change management in Bank of America of getting everybody to do this the way you think is right? And then dealing with all the security considerations and managing other people’s opinions about potential changes to the story management workflow and then maintaining the system, The answer is like probably not, you know. And so I think it is focused on, I actually think the idea that actual production of code becomes not the bottleneck for, if you know what the spec is, not the bottleneck is like incredibly Interesting. I do think it overstates like how much of (Time 0:02:50)
  • Engineer Reactions Will Vary Widely
    • Engineers will react differently: some lose joy as tooling commoditizes craftsmanship while others gain productivity.
    • Identity and enjoyability, not just skill, shape career reactions to AI tooling shifts. Transcript: Sarah Guo I think to your point of like the five-person company versus the very large enterprise, if you ask that same engineer who’s like pissed about paying $10 a seat for Jira, like if you asked Him or her, like, do you want to do the change management in Bank of America of getting everybody to do this the way you think is right? And then dealing with all the security considerations and managing other people’s opinions about potential changes to the story management workflow and then maintaining the system, The answer is like probably not, you know. And so I think it is focused on, I actually think the idea that actual production of code becomes not the bottleneck for, if you know what the spec is, not the bottleneck is like incredibly Interesting. I do think it overstates like how much of the overall software vendor problem that is. Elad Gil Yeah, I think people also misunderstand how much demand exists for software products. And by software products, I mean everything. I mean, AI, I mean… Sarah Guo Is software eating the world? Is AI eating the world? Elad Gil AI is eating the world. So I think that is actually true. And I think Mark’s post on that was really thoughtful and poor thinking on it all. And there’s so little supply of engineering in reality relative to that demand, that as you add this enormous boost of productivity to software engineers, it just gets soaked up, right? Because there’s so much more stuff to build and to do. And I don’t see teams, you know, startup teams continue to hire engineers for a reason, you know? I think the nature of the work is shifting, and I think some people are going to have real issues with that shift. Because fundamentally, you’re shifting from, in some cases, there’s a few different types of mindsets around engineers. And one of the mindsets is the really bespoke craftsmanship. I’m going to do the aesthetics of the thing that I’m doing really well and I care about the code quality and the artisanal version of what I’m doing. And then there’s people who write code because it’s a utility that allows them to build product. There’s some people who really like aspects of the math. There’s lots of different motivators for people to write code. And I think a subset of those people are going to be less happy in the new world. It’s kind of like the indie game developers who make these handcrafted individual games for themselves and then for their friends. And then they launch them on the Apple store or whatever versus the people who’d work at EA. And they each had their own version of craftsmanship, but it was just a different type of thing. I think we’re going to see a lot of these really great engineers who care about the bespoke craftsmanship of everything they do, they’re going to be unhappy working at larger companies As these coding tools get even more accelerant, because it goes against their approach of how they like working and what they enjoy out of the work. And for other people who are really focused on the utility of just building product, it’s going to be freeing in some ways. So I think there’s also like a variance in terms of the reactions to this stuff, depending on the type of utility function that you have relative to the work you’re doing. Yeah. Sarah Guo I think related to that, the one thing I’ve seen is that if you have an engineering identity that’s based on like a value-based ranking of difficulty or skill, like the specific types Of engineering that are considered, you know, impressive or high status can actually be like less hard for agents. Right. So I think there’s an enjoyability like element and then an identity element. And actually one of your founders from Applied Intuition wrote a good blog post where there is an essay where he says, like, keep your identity small. I think that’s like wonderful overall advice for this period of time. Right. You’re like more adaptable if it’s true. But I think your overall view of there are a lot of unsolved problems and like making an abundance of software can better address that. I strongly agree with. And one thing that actually is near and dear to the audience that is really unsolved is like, we’ve broadly been thinking about what happens if you have abundant code generation. And in like, I think in all of our teams, agent first engineering management and thinking about code quality is an unsolved problem. Elad Gil Yeah, and we’ll get there. It’ll be your homework and we’ll get there. What do you view as the major problems? Sarah Guo Well, the anxiety that I see is like, if you can generate an enormous amount 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’s like the slop problem. But instead of it being like vibe coding slop for random websites for non-technical people, it’s vibe coding slop in my actual production code base for every lazy engineer, which is Every engineer. I think people are like looking at some problems of actually do think ticketing, ticketing systems are like at risk. But I think the broader problem that Jira could go solve or a new company could go solve is like nobody knows how to manage that issue of human attention to engineering. (Time 0:05:39)
  • Manage Generated Code Quality
    • Build tooling and processes to manage abundant AI-generated code quality and human attention.
    • Focus on testing, smart review, and formal verification to prevent fragility from generated code. Transcript: Sarah Guo Well, the anxiety that I see is like, if you can generate an enormous amount 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’s like the slop problem. But instead of it being like vibe coding slop for random websites for non-technical people, it’s vibe coding slop in my actual production code base for every lazy engineer, which is Every engineer. I think people are like looking at some problems of actually do think ticketing, ticketing systems are like at risk. But I think the broader problem that Jira could go solve or a new company could go solve is like nobody knows how to manage that issue of human attention to engineering. And there’s a bunch of ideas like testing and like, you know, smart review, just let agents do it, formal verification. (Time 0:10:08)
  • Revenue And Token Costs Are Accelerating
    • AI labs compress historical revenue timelines, hitting massive scale much faster than past software companies.
    • Token inference costs have collapsed even as usage and revenue ramp dramatically. Transcript: Elad Gil Yeah. One thing that Jared on my team put together that I thought was super interesting was he pulled data from Capital IQ where they just predicted some projections on OpenAI and Anthropic. And they looked at and then he sort of graphed out. And maybe we can share these graphs as part of this episode. He graphed out how long it took different companies in years to go from a billion in revenue to $10 billion in revenue. So for example, ADP took 20 something years to grow from a billion to $10 billion in revenue. And then the next wave of companies like Adobe took about 20 years to go from one to 10. And then you fast forward in time and you have things like Salesforce or SAPs for an even more modern cohort. And they took eight or nine years. Microsoft took, you know, seven-ish, eight years. Google and Meta and AWS took a couple years, you know, three, four, five years. But the AI labs did it in roughly a year, right? And then if you look at the projections- It’s a wild chart, yeah. It’s a wild chart. And so we should add it, right? But you just see it go from like 20 something years with Adobe to like a year for the AI labs. And then if you look at the projections that are sort of the public projections, they aren’t necessarily the company driven data, but the public projections on where the labs will end Up or how long it’ll take them to go from 10 to a hundred billion in revenue. For Microsoft, that was something like 27 years. For Google, it was over a decade, same with AWS, roughly the same for Meta. And then for the AI labs, it’s like three, four, five years. It’s very fast. And so we’re seeing the fastest time to real massive revenue that we’ve ever seen in the history of software. It’s just these insane curbs. And again, we should expose them. Part of that, I think, is just the internet has created this global pool of liquidity and it’s helping your customers online. It’s much easier to distribute than it’s ever been. So that’s one piece of it. There’s more people with access. There’s higher GDP. There’s lots of drivers for that. But then simultaneously, you’re just creating enormous business and user value at massive scale simultaneously. And these capabilities are so rich that you’re seeing this take off in terms of revenue. And so it’s unprecedented. It’s really impressive. And I think people are ignoring the revenue and usage side of the equation. The other thing that we actually put together was the collapse in token pricing for equivalent models. I think this was done initially by David, who worked for me, and then Shran. And so, for example, we looked at the cost of a GPT-4 level or equivalent model. We looked at that a year or two ago, and basically in 21 months, it went from like 37 bucks for a million tokens to 25 cents. And so, you know, pricing dropped by 150x in 21 months. And then we tried to accelerate that curve, but obviously people aren’t really using GPT-4 level models anymore, even though, you know, they’re two, three years old. And so we looked at O1 equivalent models and the cost of a million tokens on an O1 equivalent model in December of 24 was about 26 bucks. And then in November of 25, it was 30 cents. So we saw another 88x drop, not 88%, or 88 times cheaper in 11 months for that next generation of models. So we’re having pricing collapse on the token side while we’re having revenue ramp insanely on the usage side. And so that’s insane if you think about that, just this pace of shift of cost, of revenue, of utilization of everything. (Time 0:15:26)
  • AI Could Pull More GDP Into Tech
    • Tech’s share of GDP and S&P market cap has surged and AI could push more services into tech spend.
    • This implies much larger terminal values and more trillion-dollar company potential over the next decade. Transcript: Elad Gil We basically asked what proportion of GDP is tech, right? And just the US economy, at least. And how has that grown over time? And also like, what does that meant in terms of market caps right and so if you look back to uh 2005 google was worth 100 billion dollars and exxon was the world’s most valuable company For 100 billion dollars market cap and then um it took until 2018 apple was the first company with a trillion dollar market cap right ever and everybody was shocked that anything could Get to a trillion. And at the time, tech represented about 30% of the S&P. Before that, it was say, you know, 10 %-ish back in 2005. And now the top eight tech companies are about 23 trillion of market cap. And they make up well over 50% of the S&P in terms of value. At the same time, they went from basically 4% of GDP in 2005 to about 12% of GDP today. And so then the question is, what proportion of GDP eventually just becomes tech? And AI is a driver of this, right? Because you’re taking services and you’re taking certain types of jobs and you’re augmenting them with AI and you’re converting them into effectively software spend or tech spend. And you can make different assumptions about growth rates. And then based on that, you know, you can end up with anywhere between 15, 20% of GDP to, you know, 30% of GDP in 2035. But that means that the market caps of these tech companies get even bigger. You know, it’s kind of a metric for how big can these things actually get as they sort of aggregate up portions of GDP. So I think that’s the other lens that people aren’t really thinking enough about in terms of what are some of these terminal values 10 years from now? Like how much more can things grow? And what are your assumptions around that basis for growth? You know, and this is back to like that ramp up into revenue. So it’s a very interesting kind of set of questions that we’ve been asking on my side, just in terms of these meta things. (Time 0:22:27)
  • Plan Exit Discussions Proactively
    • Pre-schedule board meetings to discuss exits regularly and remove emotion from the decision.
    • Sell when an offer exceeds likely future value rather than waiting for a sentimental peak. Transcript: Elad Gil And so as a founder, it’s really useful to be asking about two things. One is what is the durability of your business? And number two is how should you think about when to exit if you’re going to exit? Because often for companies, there’s about a 12-month window where your company is the most valuable it will ever be and then it crashes out. For a very small handful of companies, the answer is you should never, ever, ever sell. For most companies, the answer is you should sell when the timing is right. And the question is, how do you know when the timing is right? Because ultimately, you’re going to hit a point of maximal value, and then it has a real potential to die, even if it got enormous traction. And that was the internet wave of the 90s. So I think too few people are thinking about this. And one tip for founders is from a hygiene perspective, but also just a way to make it a non-emotional discussion is pre-schedule once or twice a year, the board meeting where you talk About exits. And that way it becomes non-emotional. It’s not about we’re going to exit. It’s not like we should exit. This has actually been Horace’s advice, I think, from when he was running Opsware. You just set up a non-emotional meeting once or twice a year. You’re like, nope, still not time to do it. Or you say, oh, you know what? Actually, the competitive dynamic has shifted dramatically. Somebody’s come to us with an offer that’s higher than anything we’ll achieve over the next five years. Now’s the time to do it, right? And I think it’s useful for you to be thoughtful about that. And again, the default for a small number of companies is never, ever do it. For almost everybody else, it’s worth considering at one point or another, because you may otherwise get stuck with something that isn’t working for a long time, or you may get crushed By a competitor. (Time 0:31:02)
  • Lotus Example Of Sudden Displacement
    • Lotus grew fast with a killer app and then collapsed after Microsoft launched Excel and IBM acquired it.
    • Historical winners can vanish quickly when platform shifts realign power and distribution. Transcript: Elad Gil Yeah. And if you even go back to the 80s, you know, you had Lotus. I don’t know if you remember this company. Sarah Guo I have implemented Lotus 1-2 at an enterprise business as an intern. Elad Gil Yeah. So, wow. So Lotus built one of the first spreadsheet products and it grew explosively. It got into the hundreds of millions of revenue, like really, really fast. And this was the 80s. Yeah. Right. And then a couple of years later, it basically collapses into the arms of IBM and Microsoft launches Excel and takes the whole market roughly. Right. And so, again, it looked like a very durable business. Was the killer app on computers, you know, for its era. And then it just died. It didn’t die. It ended up with a great exit to IBM, but still it is, it no longer exists, right? In reality. And so I think the same thing is going to happen for a number of companies of this era. And the question is, which companies? That’s a really hard question, right? Who knows? But for some companies, you’re starting to see cracks, right? And so for the companies with these cracks, as the market structure shifts, as you see shifts in what the labs are doing, as you see shift in usage, as you see shift in differentiation And defensibility and all the rest, it’s a good time to ask, hey, is this my moment? Are these next six months when I’m going to be the most valuable I’ll ever be? And then I’m at real risk. And if so, you know, you should think seriously about what to do with that. And I view this not just, I mean, right now, I mean, every six months, there’s going to be these shifts that are worth considering. And that’s why it’s like pre-schedule the board meeting. (Time 0:33:52)
  • Bundle To Defend Market Position
    • Defend against fast AI displacement by building multi-product bundles and broader surface area.
    • Become part of multiple workflow touchpoints to increase stickiness and reduce clone risk. Transcript: Elad Gil Best way to defend against this is to build a bundle. So it’s to build a multi-product surface area for your company so that you cross all multiple things into the same organization and you become a default part of the workflow. And that’s the best way to defend against this because then you’re being used for five or 10 different aspects of that vertical that you’re in or that application that you’re in versus Here’s my singular thing that’s easy to clone or copy or for people to kind of displace. So I think the sort of defensive advice on that is do that. Bundles are often seen as offensive, but I actually think they’re amazing for defense, you know. And so I think that’s the other thing that people are underdoing a little bit for some of these vertical applications. (Time 0:37:08)
  • Velocity Compresses Competitive Cycles
    • AI’s rate of change compresses decade-long displacement cycles into years, increasing turbulence.
    • Founders must treat every two years as if it were a decade and iterate strategy faster. Transcript: Elad Gil That was kind of SaaS era. And so the difference with AI is the velocity of change is so high that what normally would have taken a decade and you’d have a normal decade-long displacement cycle is now happening In a year or two. And that’s really the reason that these things are so turbulent. It’s because the technology is shifting so dramatically so quickly. And that’s just part of scaling laws and that’s part of reasoning and that’s part of all these things that are, you know, all the post-training stuff that’s been rolled out. So there’s just been so much innovation in such a compressed period of time that that’s the reason things are turning over and things that normally would have taken a decade are happening In a year or two. And that’s why we’re seeing these displacement or potential displacement cycles. But that also means as a founder, your mindset should shift into this new world framework. You should say, okay, if every two years is 10 years, I need to think really quickly on changes that are happening. I need to react to them in all sorts of ways. Yeah. And so it’s just back to, you know, it’s a fun and interesting and exciting time. (Time 0:38:30)