The New Career Map for Software Engineers

Things are changing so rapidly right now. What's your advice for folks that are early career and trying to get a foothold on things? Number one, use every product. There's just so much alpha in using the products and having your own intuition. You're 22, 23. How are you interacting with the senior vice president of a major multinational corporation? I came up to him after his all-hands presentation and told him everything that was wrong. Companies like Salesforce are plummeting right now. If you were an allocator of capital, which I guess you are, like where should we put it? What's a good bet? Software is oversold. I just don't think people are going to vibe code their ERP, their CRM. The upside is so low and the downside is so high. It doesn't make sense on any level. And look, again, if you're somebody who wants to get promoted in one of these organizations, the way you typically get promoted is by. Today we're speaking with Anish Acharya. Anish, we were both at Amazon as SDEs in the mid 2000s, but went on completely different career paths. I stuck around for the corporate thing, but your career trajectory went from builder, operator to betting on founders and their companies. Walk me through some of the critical moments and junctures in your career and what your thought process was. Well, great to be here. Thanks for having me. It's so hard to look at your career from the outside in, you know, it's sort of like looking at your own LinkedIn page, because it appears to be a set of very deliberate, sequentially improving steps. And it's anything but that, you know, of course, it's chaos on the inside. So I don't know that I've got a grand unifying theory, but I will tell you the sort of two things that I've paid attention to. You know, one is trying to be at the edge of where at least the most interesting people, most ambitious people I know are working. And, you know, when I was growing up in a small town in Canada and Kingston, Ontario, every interesting person who is sort of a software curious or computer curious person in the late eighties was talking about Waterloo, the University of Waterloo. So I made my way there in the late nineties when I graduated high school. And, you know, after that, the most interesting people I knew were all going to Silicon Valley or to Seattle to work at Amazon Microsoft. And actually Amazon and Microsoft were both very focused on recruiting out of Waterloo. They were two of the earliest. So there's strangely a big contingent of Waterloo folks in Seattle. And when I got the Amazon opportunity to go there full time, even though most of my friends, my co-founder ended up coming with me, most of my friends stayed home in Toronto for very good reasons, you know, family and girlfriends and boyfriends and all the, you know, very reasonable jobs with sort of reasonable career paths in Toronto. I instead decided to be more singularly focused on like, hey, let me go try to be at the edge of this thing. And as you remember, those were really interesting years at Amazon. You know, it's funny because my parents never quite got it. They used to tell people I worked at an online bookstore, which was technically true. I think at the time we had books and maybe CDs and a few other things, obviously turned into something much more substantial than that. But what I remember, I'd love to hear what you remember is just how serious minded and technical everyone was. I remember both like serious and then not serious stuff. The dogs roaming the hallways, the whiteboard conversations and staying up all night, you know, like the big launches when we're all in like the room stuck together. Yeah. No matter what you're working on. Yes. Yeah. But also, you know, I think we were at the forefront of like service oriented architectures, distributed computing, high volume transaction processing, serving a ton of requests to end users. And so it did feel like a really special time. Yeah. Yeah. And actually it's a fun topic. I never get to talk about this, but I remember there was two types of, two shapes of problems at Amazon. There was the very high volume request reply synchronous systems, which is what I worked on in the, you know, software platform team. And then there was the workflow systems that were all built on Tibco Rendezvous. If you remember that, that was everything in sort of supply chain and order management. And they were just, I mean, I don't know where else at that time in your career, you would have gotten exposure to those two problems at that scale. It was fascinating. So after Amazon, you said maybe your career is defined by going to the edge. Yeah. And so why did Amazon stop being the edge? And, you know, what were the other opportunities there? Well, a lot of it, I mean, I sort of always, and it's an interesting moment in time to rewind it back because I think that, you know, given where coding agents are today, there's a set of people whose identity is a programmer and a different set of people whose identity is a builder. And I think the builders have never been more excited than the kind of programmers who live for the output of the programming are perhaps a little bit sad. You know, so for me, I was always interested in putting things out into the world, sometimes more so than the how we actually do it. And there was a entire community forming around the Facebook platform, which was very early and interesting in 2006, because you had these new primitives around social graph and identity. And there was just no concept of that prior to Facebook, not at the scale that they actually were alive with. So that just felt like a lot of, you know, where a lot of the best builders were gravitating. And then with the release of the iPhone, that then in 2007, it was very clear that the combination of, you know, what the iPhone was doing from a hardware perspective and Facebook was doing from a social platform perspective was where the most interesting people would be thinking and working. So I quit my job to go work there. In fact, I'll tell you a fun story. You remember Rick Dalzell? Yes, of course. So Rick was incredible. When I first got to Amazon, I remember Rick, he's an SVP. He did this all hands for our platform team and maybe it was the entire sort of infra team. And I came to him after, of course, being 23 or 24, you know, and told him how everything's broken and I had all these great ideas and how to fix things. And he said, all right, come to my office and we'll hang out and we'll meet, you know, once in a while. And, you know, you and he can help me stay up to speed on new emerging technology and I'll help you navigate Amazon. So we built this great relationship over the years. And in my last year there, I came to him one day and I think we were talking about YouTube, which was a new thing back then. And he said, you know, Anish, you come in here every month and you talk to me about all these incredible startups. Why haven't you quit to start building one? And I said, well, Rick, I don't know any investors and I think you have to know investors to start a company. So he said, Anish, I'm gonna introduce you to three investors after this meeting. And if the next time you see me, it isn't to quit, I'm gonna fire you on the spot. So that was sort of the moment he gave me that push. And yeah, I left to start a company in 2008 and, you know, it really changed my life. Tell me about this, like you're 22, 23. How were you interacting with the senior vice president of a major multinational corporation? I mean, that's the egalitarian nature of tech, right? I think when our community, when our industry is at its best, you have this fluid interaction between the most senior executives and the sort of youngest engineers. And I don't think that's a note around sort of ethics. I think that's a note around survival because, you know, being a new graduate, you're the closest to the technology and emerging consumer trends. And these are things you just don't know if you're, you know, as Rick was then, as I am now in your 40s and 50s. And you've just gotta be able to kind of interplay between the two. And he was very cool. You know, he was happy to meet me once a month and hear all of my crazy ideas. Well, isn't there a cardinality issue? Like, he has this gigantic organization underneath him. There are only so many junior and mid-level engineers that he can interact with. Were you lucky? Was there something that you did to catch his eye and to have that conversation? I mean, I just, I came up to him after his all-hands presentation and told him everything that was wrong. And I had a lot of other people who were my peers at the time, but nobody else did. And I didn't do it because it was some sort of career move. I did it because I felt strongly and I didn't have the probably appropriate level of fear of executives that I should, you know? And he took it in stride. Oh, that's amazing. Do you recommend that people do that? That they speak their mind and they just kind of go for it and send it? I think the most powerful way to show up at your job in most organizations is be the founder in the room. This is what I've always tried to do. You know, if you assume that you have that level of agency, if you assume that you're allowed to weigh in on those things, and also if you take on the sort of responsibility of having a point of view, I think it just makes you a really unique individual. That doesn't mean you've got to be, you know, outspoken to a degree that's unhelpful, but also think that sort of timidity never serves you early in your career. You know, you've got to say yes. You've got to be willing to speak the truth. You've got to do the extra thing. Sometimes that means leading. A lot of times it means following, but I just think that founder mindset can show up in so many positive ways when you're in a bigger organization. If you've been coding with AI agents, you've probably noticed that they're only as good as the context that you provide. They can create high quality outputs, but only after you do the hard work of crafting stellar context as input. Without the full picture, you get the speed, but the output's not quite right. This is where today's sponsor comes in. Linear is more than an issue tracker. It's the ultimate context layer for your entire product system. It connects to your whole stack, like GitHub, Slack, support tools, and your customer data. And because AI agents integrate directly with Linear, when one picks up a task, it can see all of the context that you've accumulated, which customers were impacted, ongoing projects, historical decisions, and so on, without you having to spend the time to craft that context. All that matters because high quality inputs lead to high quality outputs. If you're gonna rely on AI, you need a system that feeds it context automatically rather than having to bring it yourself manually every time. That's why more teams are switching from Jira to Linear and using agents to resolve issues over three times faster. Check them out at linear.app slash A-L-E. So let's shift over to the elephant in the room, which is AI. Oh, great, great, okay, yes. So discourse on AI is polarizing, and it's difficult to find level-headed takes. You have a thesis that the world is short on software or short software. Can you explain this thought? If you look at software, software for so many years has been this incredibly precious resource. And even with all of the incredible technology that we consume on a day-to-day basis, the number of people that can create technology is very, very small. Maybe we have a million programmers, at least order of magnitude. So the impact in terms of the amount of software created to just making it easier for people to create software I think is dramatic. And then if you look at the areas in which there are demand for software, even if you look at the most sophisticated sort of cutting-edge technology company of the, I don't know, let's take a big tech, Google or Meta, the question I'd ask is how many roles are information age roles versus industrial age roles, and how many tasks are information versus industrial age? Even if you're an engineer working at Meta, you're getting together in a room like this, you're discussing trade-offs, you've got all of this sort of rote industrial age work to do. So even if you're at the edge of technology, so much of your work is not actually on a technology platform. So I think that the opportunity to make it much, much easier to create software means this sort of diffusion of software across our work and across our lives. I also think it's an output and a kind of channel for creators and consumers. It's sort of like generative music. I think that music is incredible, but there is not that many people who make music that's disposable for good reason, because a lot of music is high cost to create and you have to create it very carefully. But now we start to see sort of mimetic music where people are creating a track on Suno or on Udio just for a joke or for a meme or for a bachelor party weekend or whatever else. So you reduce the sort of complexity of creation and you've found all these new needs and sort of demands for music. I think in the same way, if you reduce the cost of creating software, you make it less specialized. All of a sudden, when it comes to personal software, when it comes to software that's disposable, software that's only relevant for a moment in time, you're talking about being at the Super Bowl. You know, there should have been a mini app experience just for you and the people that were sitting around you. Why not? And that software would have had no value the next day or even when the game ended. And because of the sort of trade-offs that were implied, you never would have created that prior. You make an analogy between YouTube in its early days, I think circa 2005, 2006, and the moment that we're experiencing right now within AI. Where does that analogy really fit and where does it end and tail off? That's a good question. If you look at where we were from video creation and sort of just video and media in 2005, 2006, it felt like we had enough. You know, if you had one of those big satellite dishes in the back of your house, you had 200 channels, but either way, the average consumer probably had access to 30, 40, 50 channels. And you know, there's 24-hour content on all those channels. So I think a very natural question 20 years ago was, hey, do we really need more TV, more video, more media? And if we do have it, what form will it take? And in a sense, that question, people tried to answer that question on the early days of YouTube. When you saw a lot of, you know, home camera footage, when you saw people pirating shows from traditional linear television to YouTube, over the years, what you saw was all of these emerging sort of YouTube native formats, everything from unboxing toys to get ready with me. These are things that just were inconceivable 20 years ago. And now those things that I'm discussing are bigger and more substantial than linear TV. And in fact, funny enough, so much of traditional media is actually just commentary on emerging in new media. So I think the same thing is true of software. You know, today our NBC and CBS are sort of like ERP and CRM. They're big and they're serious. And like the trade-offs make sense. What is the equivalent of get ready with me or unboxing in software? I don't know, but I think we're gonna find out. You know, if I were to channel my own experience in the mid 2000s, it's like, oh, cool. You can post a YouTube video, but like, you're not seriously thinking that you could create your own Star Wars. You know what I mean? So I'm like, oh, this must be throwaway because, you know, there's this high art or the high production value is the thing that makes the video good. But it turns out I had a demand for things that I didn't know existed that was actually at lower production values. Absolutely. I mean, look at Twitch. Twitch is such an interesting example. It's like what YouTube, it's YouTube's YouTube in a sense. Because when Twitch was started and Justin TV, you could have easily looked at YouTube and said, okay, we've got old media, we've got new media, like we're good. And then also do people really wanna watch other people play, you know, AAA video games? And it turned out that not only do people want to watch other people play those games, but it was enough to catalyze unique live streaming video platform that of course was bought by Amazon many years later. So I just think that these things continue to surprise us to the upside and software will follow a similar trajectory. Let's talk about advice for, you know, the next generation or maybe people that are currently employed and maybe mid-career. Things are changing so rapidly right now. What's your advice for folks that are maybe early career and trying to get a foothold on things? Yeah, and a lot of this advice is out there, but I think it's good advice, which is number one, use every product. You just have to use every product. I really learned this from my co-founder. My co-founder of both my companies was this, he's just a technology enthusiast. So he was the first person I knew that had an iPod, first person I knew who had a mini disc player, the first weirdo who moved from PC to Mac, the first person who had an iPhone when, if you remember the iPhone mail app used to say send from iPhone, you know? So it just, he was this enthusiast. And as a result, he always had intuition that was three, six, nine months ahead of everyone else. He was bullish where others were bearish and the inverse. And I think that that is so important for all of us to really internalize. There's just so much alpha in using the products and having your own intuition. Really, it's about primary sources. And look, there's a lot of products coming out, but not so many that you can't try most of the big ones when there are new models. I try to hold myself to a standard of trying them all, using them all. It's great to have something to work on. You know, I've got a bunch of products and apps I've been working on, mostly so I have this kind of chassis for exploring the new technology. So I'd say that's like, if you can only do one thing, it's just obsess about the new products and try to use them, even if you feel like your use of them aren't yet very sophisticated. Yeah, I think about the fact that with all of these agentic coding things that are out there, essentially mandates from leaderships that they need to get on this AI train, that the AI stuff is really, it's highlighting who's actually a builder and who has, it's kind of like trying to hide and hide from the fact that like, they're not really contributing and building because organizations are moving so slowly and code is being written so slowly. Yeah, I do think that it is a, it's a renaissance for people that are really, that are really into building. There's so many CEOs and CTOs coming out of the woodwork and, you know, staying up all night. It's this, the analogy is like the slot machine. You're just like, wow, my prompt actually like made something and you feel really good. And then you put it in and then you're like, oh, my prompt didn't actually, that's not what I meant, right? And then you're like, I'm gonna pull it again. Well, do you remember this concept from the sort of social era of variable scheduled rewards? Yeah. That was like the reason, the big brain. The Farmville sort of thing. That's right, exactly. So randomly dispensing awards or paying, rewards or paying in varying amounts at varying times. And that really is what coding agents are today. You know, once in a while, there's this tremendous outcome. A lot of times there's a marginal improvement and sometimes it's a disaster. So I do think there's a real, you know, sort of exciting casino element. I also think, by the way, maybe most people are builders. Like, I don't think anybody started their career aspiring to be some Dilbert-esque middle manager. I mean, I think it would free people from having to do that work. You'll be surprised how many people are gonna make things. You know, I think I remember being a teenager and just being into games and just being like, I wanna make a game. And then, you know, I made like a Tetris clone. And it was like, oh, that's actually pretty sweet. And then you go get a job in corporate. And then you're like, I'm trying to make sure that this toothpaste like gets onto their doorstep in, you know, a few more hours. And I do think that the magic is lost a little bit when you sort of plug into these larger organizations. You know, let me challenge you on that though. Cause I think they're actually, I think you can get passionate about delivering the toothpaste to someone's doorstep, you know, in 30 minutes or less. I think where people start to get drained is when they are low autonomy. Yes. When there's like sort of very high coordination cost, which just monotonically increases that scale. I think this thing of like having people be higher autonomy, lower coordination costs, even if they're working on problems that feel perhaps incremental, will be a greater driver of human NPS than a lot of other things. I completely agree with you. I think when you tend to take that reductionist view about the fact that you're delivering toothpaste, when you're like, I'm trying to do something, but product is blocking me, design won't get their design in on time, middle management doesn't know what's going on. They don't know what's going on on the ground floor. I think Marc Andreessen talked about this sort of like Mexican standoff that's going on, right? Product engine design. Yeah. So the idea is that engineers are like, well, why do I need product and designers? I have got this AI. I can just go and jam on my thing. And the product people and the design people are essentially doing the same thing as well. Do you see the role of maybe a early career or mid-career person turning into more of a, like a fully featured, like well-rounded sort of individual? I mean, Marc's also said that, you know, high IQ specialists tend to work for lower IQ generalists, which is a funny observation. It's a very funny observation. So yes, I do think the future belongs to some extent to the generalists. Maybe that'll be the kind of torso. I do think there's an interesting unbundling that's happening though. If you look at product management, like there was a part of product management that was very technical, always. There was a part of product management that was sort of pointing people in the direction of a market need. And I do think that if you look at the big labs, you'll see the product managers, they are very close to research. Like they're more technical than PMs have been in a long time, maybe in a generation since the Ben Horowitz era. If you look at the other side, you see these PMs who've got, you know, multi-agent systems and 10 instances of cloud code or codex. And so I think the unbundling and specialization is gonna be how it shows up rather than being the sort of zero-sum Mexican standoff. Yeah, I don't understand the tension. I think there's a sort of like, yeah, people are becoming more well-rounded, but like, where's the tension and the conflict, right? I do think, you know, if you're, I put a meme up about this that was pretty funny a month ago, which is like, if you're an eng reviewing your 57th PR from product management, you know, it's driving you a little crazy, you know? And if you're a design looking at yet another sort of bit of interface slop, tailwind and whatever else, it is driving you a little crazy, but that's just, it's the moment of transition as we go from the old definition of our professions to the new ones. If you're watching this conversation, you're probably thinking about your next career move. Maybe it's a promotion or finally getting the performance rating you deserve. Maybe you're just tired of feeling stuck. I built Top Tier because career growth doesn't have to be trial and error. I've worked with hundreds of people at this point and we all get stuck in the same places, making the same mistakes. You can figure it out all on your own and waste years, or you can get guidance from people who've already been where you are. Top Tier is group coaching for software engineers and scientists who want to maximize their careers. You get access to me and my team and the opportunity to connect with hundreds of people who've been exactly where you are, all in a confidential setting. We work on visibility, communication, delivery, influence, navigating politics and reorgs, everything that actually impacts your career. If you're serious about growing your career, check out Top Tier. The link is in the description. What's your take on like a similar tension between manager and software developer? If you think about software developers now with agentic coding, they're essentially turning into a manager. It's just like, hey, I have this team of Nobel laureate like intelligence agents that can go off the rails and so I sort of need to babysit them. Maybe I have like an agent of agents, but I'm also sort of like managing that. So now I'm a middle manager. And then also on the side of the SDM as well, the software development manager, they are probably like, well, can I just manage like many, many more people? Like, should I just be managing agents? Like what's the thing there? Do you sort of see a emerging there? I think that a lot of the role of management in the sort of modern economy is to manage what is otherwise a very scarce resource. And in software companies, that is the engineers, right? Like the amount of hand-wringing that happens in executive meetings and cross-functionals and VP fighting and all this stuff over prioritization and communication is enormous. I think what changes with coding agents is maybe you don't have to prioritize. Maybe you just build everything, you know? So how does that sort of decrease the management overhead? So I don't know if it's a tension. I do think that the reason most managers will go back to being hands-on is because the work of allocating the scarce resource is a lot less important than it once was. I had heard the term, maybe we don't have to prioritize in a different context. There was a CTO I was speaking with. He was leaving because the CEO had this idea that he would just assign software devs to each of the C-suite and they wouldn't have to prioritize anymore. And so he's like, time to walk away. But maybe that's the world that we're going to. I don't know. I mean, I do think that there's a set of work that can probably happen relatively autonomously now. And this is where systems and analytics and other points of feedback become very important. Because you can imagine if you're gonna ship every feature, okay, well, how do you test every feature? How do you decide what gets promoted or production? How do you sort of expose the right set of features to the right users? So having a system for sort of running, testing, scaling, or culling experiments is more important than it's ever been before. I think a lot of the work that previously went to, you know, middle management, prioritizing the work of engineers will now go to management building out systems to support the idea that there's no more prioritization. Two thoughts there. One is, is your to-do list ever going to be empty, right? If everything on your to-do list is done, I think there's this expanding frontier of problems. And so you will have to prioritize those things, right? So it's sort of like you're squeezing a balloon from one side, you're just displacing it to the other. Perhaps, but at a higher sort of, a higher level of granularity, you know? So now instead of like talking about what's the ambition of an engineer or a team or a function, you're talking about what's the ambition of us as a company. And like, I think that I'm glad that you mentioned the expanding frontier. I think that's exactly right. Like I've said this before, there's no way what we see around us is the peak expression of our ambition. We've all read enough sci-fi books to know there's like a lot more to build. We don't even have enough skyscrapers, you know? Like, come on. So I think that there is no ceiling on ambition and that's where a lot of this new energy will get pointed. Yeah, I do see that the existing constraint now that code is becoming more and more of a solved problem is code review, observability, making sure that the thing that you've created is doing right by your customers. Do you see that as a temporary limitation, these new bottlenecks? Or do you see that as something more fundamental that, you know, there will always be some sort of bottleneck in the process? We have to have better systems than we've ever had before. So I do think there will be bottlenecks and trade-offs. I think the magic though, for most PMs, like if you look at how most PMs describe themselves or aspire to describe themselves, they wanna be the sort of innovative new product thinker. They wanna be somebody who's imagining the future and then shipping the future. But how many PMs actually get to do that work? Very few. I would argue most companies have zero or maybe one person that does that. That is a question of sort of skill versus desire, but holding that aside, I think that if you now start to say, for example, every A-B test and every sort of experiment and product improvement you'd make based on data is something that is fully automated by a system. The PM is free to do the new thinking and the new sort of product designing and take systems that would otherwise go to a local maxima into a global maxima. That's a much more satisfying product role than they might've had five, seven, 10 years ago. I wonder about the dynamics of so many cooks in the kitchen. You know, I think Thomas, the ex-CEO of GitHub was basically like, if you just go and build every feature, you get the Homer Simpson car from that one episode. It's like- Yeah, yeah, yeah. I think that car was sweet, dude. I think so too, exactly. It was ahead of its time. But I think his point holds, which is essentially maybe in this Mexican standoff, like product is the winner, right? At some, like what is in and what is out, you know, taste, I think the ultimate expression is that's the surface area for which your customers are actually using and interacting with the company. Yeah, I think that's potentially right. I mean, I do think that there's gonna be, not every experiment will be shipped to production at 100% forever. How can that be the case? But I do think every experiment and perhaps the sort of combinatorical combinations of those experiments will see the light of day or more of them will see the light of day. So I think you probably move to some sort of an optimal state that is a dramatic subset, but that's because you tried everything, not because a bunch of PMs got a room and argue about everything. I do think it's a more optimal state of the end product probably. But look, to your point, it's a harder systems design problem than ever before. Yeah, just to close the loop on the advice for early and mid-level devs, it sounds like embracing the new technology, embracing the edge, leaning into the tools, becoming a product engineer instead of just some backend nerd in the closet that keeps the thing running. I think that's the durable advice there. I think so, yeah. I also think have more hobbies. Everybody is sort of ambitious in some direction. You talked about you building your Tetris clone. If you were growing up today, I'd say build a AAA gaming studio. I don't know. Again, not because that's gonna be, not every person wants to be an entrepreneur, but there's just a sort of creative outlet you get from making things. It's also the best way to build intuition. So yeah, I think just do more stuff is the frustrating advice I'll give. I think a really good litmus test is when these coding agents come out. Like, do you actually start to see the possibility and get really excited about what could possibly happen? Or do you get this like dread that the tasks that you do are going to get replaced? Yeah, exactly. I mean, spreadsheets didn't come for your job and neither will coding agents. It's just a new tool. And you know, you don't wanna be the person who's obsessed with machine code when assembly is the hottest new programming language. And you know, Karpathy famously said, English is the hottest new programming language. So great. Let's embrace it. Let's get articulate in this new language of sort of technical expression. Nice. So let's talk about how A16Z is sort of thinking about allocating capital. We had just talked a little bit about the SaaSpocalypse that's going on right now. Sassacre. Oh, SaaS. What did you call it, Shin? Sassgate. Sassgate, whew. Political scandal, dramatic. You know, companies like Salesforce are plummeting right now. If you were an allocator of capital, which I guess you are, like where should we put it? What's a good bet? Software is oversold. I just don't think people are going to vibe code their ERP, their CRM. The sort of upside is so low and the downside is so high, it doesn't make sense on any level. You know, if you look at SaaS spend today at the enterprise, it's 10% of overall spend. Okay, so the upside case of vibe coding all of this incredibly critical infra is 10%, you know? And look, again, if you're somebody who wants to get promoted in one of these organizations, you've been at Amazon, you know? The way you typically get promoted is by shipping new things, not by shipping slightly more cost-efficient versions of existing things. Also, I think there are all of these sort of second-order considerations for rebuilding those systems, like third-party ecosystems, professional services work, you know, the engineer in the corner that nobody's talked to for 20 years that knows a part of the code base that has specific runtime behaviors that the sort of enterprise depends on. Very hard to kind of infer all of those things with a coding agent, and why would you do it? So I do think the SaaS is dead story is overtold. The much more interesting story in my mind and the story that's relevant to SaaS is I think switching costs go dramatically down. My partner Alex has sort of got this funny quip that, you know, some companies have got hostages, not customers, and nowhere more so than, you know, some of these ERP systems, some of these sort of core systems, the systems of record, as they like to say. And again, if you actually bought an SAP system or an Oracle system, you're effectively a hostage of that company. And now with coding agents, the sort of cost and complexity of transitioning is far, far lower. You had some advice around, you know, going and doing a narrow startup where you can charge $200 a month, monthly recurring. Does that advice still hold in this new sort of SaaS is dead world? More than ever, more than ever, absolutely. Yeah, I mean, I do think you can get a small number of people to pay a very high price for your software because the ambition that you can have, the amount that the software can do can just be tremendous. You know, if you look at, I don't know, let's look some traditional market. What is the software spend in an individual dental office or dental practice or sort of group of those? You know, it's just, there's not enough software spend for you to be that narrow. I think today, if you said that, well, 90%, not the 12% that's being spent on SaaS is my opportunity. You can specialize at maybe two orders of magnitude, higher degree and still have the same size business. So I do think this like dramatic specialization opportunity is a really interesting one. Also makes it more fun to build for yourself. So like, look, if I find another dozen or a hundred or a thousand people that have the very specific problem that I do, I can now go and solve not 20% of it, but 90% of it and charge two, three, $500 a month. Yeah, I think that goes back to your, you know, we're short software. Yes, yes. Right, I think this is interesting. You know, we at Amazon, there was a multi-year, very painful, we need to get off of Oracle initiative. I remember, yeah. We were a hostage. A hundred percent, yes, yes. We were also the poster boy for Oracle. It's like, hey, you know, it works for Amazon scale. It should work for everybody. And that was so painful. And so I think in this new world where you have to migrate, you know, 150,000 databases, I don't know exactly how many there were and make sure that production operations are fine. We didn't lose any data. Everything works correctly. What a wonderful use case for agent code. That's right. A couple of things here. One, like, yeah, you can maybe slide across like, hey, let's build our own compiler and it would be taken seriously, you know, I think within the next year. But this like, this concept of switching costs is the thing that's rubbing me right now, which is like, okay, cool. Maybe we can get off Salesforce. Maybe we can get off Oracle, move to a different vendor. But does that mean that there might be a race to zero in terms of like the problem that you're solving, especially if there's a lot of prior art out there? I don't know. I mean, I don't know that pricing has been what has held back SaaS traditionally because there's often, you know, you've got Salesforce, but you've also got SugarCRM. So there's always been low cost or open source alternatives to a lot of these products. I think it's a combination of sort of go-to-market muscle when you have an enterprise with reference selling, you know, everybody wants to buy IBM or whatever, buy Oracle. The other thing is how good is the product? You know, and I think that the nice thing about reducing switching costs is that it forces the playing field to be dependent on product quality instead of your ability to take a hostage. And I think that's a very good thing, a very capitalistic force through the entire SaaS ecosystem. Got it. Okay, so your argument is like competition is good. It's not necessarily about pricing, pricing pressure and that sort of thing. That's right. Yeah. It's just like increasing the quality there. That's awesome. I think so. I mean, hey, guess what? Like last night HubSpot had their earnings and they totally crushed it. HubSpot has been like predicted to be one of these losers. Of course there will be winners and losers, but like so far we're not seeing any earnings reports. Yeah. Well, I also think, you know, Dharmesh and his sort of side projects of like having a director of agents is like turning out to be not that bad of an idea. Of course. Are there companies that you are invested in right now that maybe aren't so well positioned for this AI agent era? Hmm, that's an interesting question. I don't know. I mean, I think that there are new categories that are being created and for those categories, I think you're definitionally better set up if you're native to the product cycle. But I think for existing categories, companies that had scale and relevance, incorporating AI is a lot of the way that it'll show up. You know, I think about a company like Clutch, which is SaaS for credit unions, community banks, and sort of torso and tail banks. Really, really powerful product offering. You know, it does all the sort of interactions with the customer from deposit taking to sort of lending, a lot of the back office services as well. In terms of how you do loan processing. And they're doing a lot of really interesting work with AI, pushing it into customer support, pushing it into collections. To me, the way the credit union industry is gonna consume AI is probably gonna be with a trusted intermediary like Clutch versus a net new company. So I also think for the net new companies, they're best positioned to go after a market that didn't exist prior because there is no sort of distribution advantage the incumbents have. If software is oversold, and you know, I'll buy that thesis as well. Right now, companies like Google and Amazon, I think Google put out, hey, they're gonna do an outlay of 100 billion in AI. Amazon, 180 billion. Turns out to be radioactive for the markets. Where should they be deploying that money? Where's the right place to do it? It's like, is buying more hardware the smart move? It's very funny, because I feel like these companies are damned if they do, damned if they don't. You know, five years ago they were doing buybacks and the market was saying, hey, please put your capital to productive use. Now they put it to use and it's like, whoa, whoa, whoa, whoa. Not that way. Not like that. Exactly, not like that. I'm not an expert in data center sort of build out. What I will say is that so far, 100% of the new supply, inference supply that gets brought online is allocated to demand. And anytime you see these dislocations that are not durable, it's because you dramatically build out supply ahead of demand. I just don't think that's what's happening. So I think those are wise investments today. What is the sort of latent demand for intelligence? And even with the CapEx that you're describing, like, does that get to 10% of it, 1% of it? I don't think it's 100. Yeah, absolutely. So if you are a bull on software and you're a general partner, and like, where are you looking for like alpha? Like what companies are on your radar? Like how are you thinking about the future? I think one idea that's really interesting is the idea that coding agents are not just about creating software, sort of software as we traditionally think of it, but rather they're upstream of knowledge work. I think there's a bigger vision for coding agents. And I think, honestly, this is a little bit I'm cribbing from Anthropic. I think I'm inferring this is sort of their strategy, which is if we have the best coding model, that will then be the best model for generating sort of cashflow projections and even writing essays and doing a lot of downstream knowledge work. So I think that's one very important idea. And I think the way that shows up, do you remember a Fourier transforms? Of course. So you know that, of course, a Fourier transform, a Laplace transform, these are both tools for taking a problem that's hard in the case of Fourier time domain, moving into frequency domain. And it's a much easier math problem to solve, right? It's a mathematical tool versus a mathematical concept. Again, the same way if you can take a domain that's seemingly unrelated to code, you can represent it in code, you then get the direct benefit you're seeing from sort of exponential improvement in coding models. So I know that's a little bit hand wavy. I don't know exactly how it shows up, but I think in 26, we'll see coding agents show up in places that don't at all look like software. We've got a section called bearish or bullish. And so you tell me whether you're bearish or bullish. I mean, I can just tell you I'm gonna be bullish on everything, go ahead. Oh, you are? Okay, then you're just naturally optimistic. Tell me, tell me, tell me, let's go through it. Okay, so open claw or claw bot. I'm bearish on the moment. You're literally bearish on the first thing. Hold on, here we are. Bearish on the moment or the point. Karpathy had this great note about like, people are overestimating the point, but underestimating the slope. I think open claw as a moment in time is probably more discussed than it maybe deserves, but what it points to, what it has started, I think is under discussed. So I'm very, very bullish on sort of implications of open claw and mold book while feeling like everyone's saying that this is AGI right now is maybe overstating the case. Yeah, mold book is hilarious. When I was thinking like, LinkedIn is like mold book, but with humans, like facilitating the AIs Unclear, yes, yeah, exactly, yes, yes. There might be more intelligence on mold book than LinkedIn. You know, I also think about like open claw, like the pattern, there's nobody stopping anybody else from like, you know, implementing the pattern. Everything's like open source. Open source is incredible, yeah. Yeah, so like whether like open claw is the, like the final expression of it, I'm very doubtful of, but this idea of giving it like actual computer use, like play and stuff, I think is great. I think the cool idea there is that instead of having context be opt-in, it's opt-out. Yeah, like we don't drip feed the context. We just let it rip. It's terrifying, but also very powerful, you know? AI companion apps. Incredibly bullish. I think that that is an area in which you need a really thoughtful product primitive and you need to design for human sort of consumer preferences. So I think on product design, a couple of good examples, Tolens is a really interesting product, you know, where they've got this animated alien character that runs around this mini planet and you can FaceTime with it. It's an AI companion, but it just feels more, I don't know, tactile than speaking to a traditional chat-based companion. I think that's the first of many experiments you'll see when they bring companions to life. On the kind of consumer preference bit, I do think you need a level of indirection for companions to work for a lot of people because people don't want to say like, oh, I need an AI friend or think that about themselves. So you almost need a context in which to interact with the companion. Like maybe a new verb, right, or a new noun. So that we don't say like our AI that we need. It's like, hey, I play, you know, I play Minecraft with like, you know, grizzly bear 225 and like, you know, whatever. We're really there for friendship. But like, you know, it's, we're playing Minecraft. CS degrees for people that are going through college. Bullish, yeah. I mean, and actually I think there's a very sad thing that's happening, which is if you're an insider, which means you go to Stanford or Harvard or whatever, you know that having a CS degree is incredibly important. And yet the mainstream media in a lot of the rest of the world is saying it's the end of CS and encouraging kids who are already not inside the bubble to not study, you know, CS. I think that's a mistake. It's just a new level of abstraction for interacting with technology. Going into big tech before starting a company. Oh, that's interesting. I think the best people have done a little bit of both. You know, I was thinking of it as like founder at scale. And I actually, big tech to me is different from hyperscale. I think the best place you can be is hyperscale, be a part of something that's really working with a really good cohort of people. That's the best way. Because I think at big, big tech, you learn a set of skills that are only applicable at big tech. And then like, I mean, being a founder is extraordinary as well. So I think having the combination and the best place to get it in your early career is probably hyperscale. AI token prices, are they going to infinity or are they going to zero? I mean, on any individual model basis, they're going to zero. But I'm sure that as, you know, the models become more powerful, the cutting edge token price. Oh, that's a good question. I think we'll go up or at least we'll not go down. Cause I think what's happening is the models are substitutes for 80% of the queries and they're specialists for the 20%. So in the cost of the 20% where they specialize should go up and the cost of the 80% where they're substitutes should go to zero. Token prices are essentially an indirection to like power and compute, right? And so, yeah, I don't know. It's just one of those things where it's just like, is it going super high or super low? I don't know. Okay. Ads and AI. Well, I think taking a stance against ads and Mark has said this, is taking a stance against broad access. I do think that ads are a way that we, you know, give nice things to people that can't necessarily pull their wallet out. And I don't think any of us wants to live in a world where people that are directly paying have access to more intelligence than everyone else. When ads are well done, they're like Instagram ads. Have you ever bought anything from Instagram ad? Oh, or the Superbowl ads that are there. It's like, you're actually like, those are great. They're product experiences, you know? So I guess I'm bullish. Well, Anish, thanks so much for taking the time to have this conversation. I had a wonderful conversation with you. Super fun. Next time, let's do it over a beer. Yeah, absolutely. Awesome, man. Thank you.