Software Huddle - AGI is Surely Coming with Former Snowflake CEO Bob Muglia

Episode Date: December 5, 2023

Today we have the former CEO of Snowflake, a 23 year veteran of Microsoft, Bob Muglia on the show. In this interview, we discuss Bob's book, Datapreneurs, which takes you on a journey about the people behind the first relational databases in the 1970s and early 80s, to Bob's experience launching Microsoft SQL Server and a ton of other products, developing the Data Cloud at Snowflake, and to the future of data and AI. We cover a lot of ground, including some of his experience working alongside the likes of Bill Gates and Steve Ballmer. Timestamps: 02:24 Introduction 04:53 Relational Databases 18:43 Speed of Innovations 24:30 Keeping the Early Stage Culture 31:04 Most successful leaders are difficult to deal with 34:31 Setting up Cloud Data Center at home 36:25 Joining Snowflake as the CEO 38:54 AWS made Snowflake happen 42:18 Google, AWS Missing the Snowflake Opportunity 46:13 Impact On Jobs 50:48 Existential Risk 52:28 Staying Optimistic Links: The Datapreneurs: The Promise of AI and the Creators Building Our Future https://www.thedatapreneurs.com/ Follow Bob: https://twitter.com/Bob_Muglia Follow Sean: https://twitter.com/seanfalconer Software Huddle ⤵︎ X: https://twitter.com/SoftwareHuddle LinkedIn: https://www.linkedin.com/company/softwarehuddle/ Substack: https://softwarehuddle.substack.com/

No, because we didn't know how to build it. I mean, literally the only way we could build it. Snowflake was only made possible because of AWS at the time. The only cloud that could possibly support Snowflake in 2014 was AWS. I was essentially technically raised at Microsoft in the 1990s, in the heydays of Microsoft. You know, as I've said many times, I saw the good, the bad, and the ugly at Microsoft. I was one of the 12 witnesses that testified during the DOJ trial. And, you know, when you participate

in something like that, it leaves an indelible impression on your mind. So, like, if you go back to some of the early days at Microsoft, and of course, like, other companies, like, leaders, like, I think Bill Gates, Steve Ballmer, Jeff Bezos, the list goes at Microsoft and, of course, other companies like leaders like, I think, Bill Gates, Steve Ballmer, Jeff Bezos, the list goes on and on. They're somewhat notoriously known for being somewhat difficult to deal with, at least back then.

Certainly contemporaries like Elon Musk are no less difficult than those guys were. I think the characteristic is very driven people that have objectives. And in some senses, they're going to focus on achieving those objectives. And in some senses, they're going to focus on achieving those objectives. And that sometimes requires people to make difficult decisions and sometimes be difficult. There's many, many infamous things that Microsoft would do. One of the most interesting of which was the mid-year review process that would go through the sales organization. Sometimes it was rough. I mean, Microsoft culture could be really rough. In those reviews, people were sometimes torn to shreds.

It was not uncommon that general managers were fired after that review. That was not an uncommon outcome. So it was a tough environment. Hey, everyone. Welcome to Software Huddle. I'm Sean Faulkner, and I'm extremely excited to bring you today's episode, because today I have the former CEO of Snowflake and 23-year veteran of Microsoft, Bob Muglia, on the show. In this interview, we discuss Bob's book,

Datapreneurs, which takes you on a journey about the people behind the first relational databases in the 1970s and early 80s to Bob's experience launching Microsoft SQL Server and a ton of other products, developing the data cloud at Snowflake, and to the future of data and AI. Cover a lot of ground, including some of this experience working alongside the likes of Bill Gates and Steve Ballmer. All right, I've done enough setup for this.

I'm not even going to waste time plugging my Twitter today. Let's just take you over the interview with Bob Moglia. Bob, welcome to Software Huddle. Thanks, John. It's great to be here. Yeah, thanks so much for being here. We've got a ton to cover. You've done a lot, so let's just jump into it. I want to start by talking a bit about your book. So in your book, you take the reader through some of the history of data, starting with the introduction of relational databases in the 1970s, eventually talking about AI in the future. And at this point,

especially in cloud computing, we kind of take for granted, I think, relational databases for those that, you know, weren't there during the pioneering days of essentially data. You know, it's a bit like when you fly in an airplane now, which is a pretty incredible experience, but if you do it all the time, it's like, you know, it's not that big a deal, even though you're flying through the air. So can you talk a little bit about the impact relational databases has had on the world through your history with them? Well, you know, the experience that I've had in working with all these different databases and different people along the way was really part of the inspiration behind the book. And the realization that there's been innovation for many, many

years that have led to where we are today with some of the new things that are coming with AI. But relational databases, and in particular, SQL, has been such a core to how business systems are built. And that's been true really since as early as the 1980s, I would say. It goes back to the late 1970s when IBM invented the technology, and then it was popularized by companies like Oracle, Sybase, and others. And now it's pretty ubiquitous in pretty much every electronic business system we use.

There's almost certainly, when you're interacting with just about anything, there's probably a relational database behind that storing some part of what you're doing. If you buy something online, it's probably stored in a SQL database, and the transaction is stored. The shipment of goods and all those things are also tracked that way. These things have become ubiquitous. When I joined Microsoft in the late 1980s and in 88, you know, my first job was working on SQL Server. And the thing, you know, I probably feel best about that we did there in the 1980s and 1990s was really democratize the technology and make it available to companies of

all sizes. You know, go back to 1990, 1992, most business most small businesses were using pencil and paper to track what they were doing. And that, of course, has all changed. And again, it's relational databases that have changed most of that. What were some of the big innovations that have happened in the relational database world? The technology was introduced in the late 70s. It's been a long time since then. What are some of the step functions in the way that we've grown to essentially manage data in a relational database and some of the innovations that happen around SQL? Well, I mean, one of the interesting things, SQL has been evolving constantly. Let's just start with the fact

that SQL has evolved as a standard, an industry standard, with multiple implementations built to it. And that standard has been updated a number of times. And in particular, it has been updated to make analytics much more powerful, to be able to work with data in much more interesting ways. One of the things, when SQL was introduced, it had very strong transactional semantics built into it. The basic idea of a debit credit or the basic idea of a business transaction has been well supported in SQL from the very beginning. One of the big changes happened in the late 1980s or mid-1990s when we began to see analytic databases, data warehouses appear that are optimized for queries and searching for information and separating that from what you might call an operational database that's running those business transactions.

Back when I started working with SQL in the 1980s, you would have one database system and and you would use that for running your business, and you would also use it for reporting. It was called reporting back then. And everything, you know, most things tended to be batch-oriented in what they were doing, and, you know, reports would run at night. And over time, it became apparent that the systems could not handle both of those tasks simultaneously. And in fact, they could be optimized differently to do different things. So we saw data warehouses, specialized data warehouses emerge in the 1990. Teradata led that. And it was a pretty major shift in terms of the kinds of systems that were built.

Later on, some ways the data was stored was changed. You know, instead of storing data as individual records, most analytic databases, in fact, pretty much all of them today, use what's called columnar storage. So it stores data very efficiently to scan columns of information. Again, the difference between optimizing for a business transaction versus optimizing for really understanding what's happening in a historical system of record. So those are early changes. Now, what happened was these databases were all overloaded. Even though they were separated into analytic and operational systems. They just didn't scale as well as they could scale conceptually, and they weren't able

to store all the data and be able to deal with all of the users that wanted to work with them simultaneously. That was the status of these systems in the 2010s, in the 2000s up until 2012, 2015. And you had data scattered in lots of different places because although they were interrelational databases, they had to be put in multiple databases to handle the capacity. And one of the things that changed with the advent of the cloud was the ability to build systems that could scale essentially to any size. And that's much of what Snowflake was all about,

was building analytic systems that could run and work to support an entire company with any amount of data in it and any number of users. So a lot of changes happened. The interface has evolved. The technology behind it has evolved. But many of the applications that were built 30 years ago or more are still running on top of these same systems. Yeah, and you mentioned essentially how the split happened between sort of transactional and operational or analytical database, the data warehouse versus sort of the application database. And then I think also as a perhaps response to scale issues,

there's been more and more specialization in the database world as well. You have like time series databases. Databases are designed to handle specific types of problems that basically companies run into when they're dealing with certain types of applications at really, really high scale or they need really, really high throughput for something that sort of goes beyond just the standard off-the-shelf database, even if that's running in the cloud. Yeah, but most of them are still SQL. I mean, if you look at most of this

today, the most popular database that's being used for operational systems is Postgres. New systems are being built, and it's a SQL database. And it's probably been a SQL, you know, if it was SQL, if it was MySQL or SQL Server or Oracle or whatever, it's always been a history of those things. Yes, there are specialized databases like Time Series, but they're really used for, you know, as you say, for specialized uses. And they aren't, they are not nearly as broadly applicable as these databases are. One sort of characterization that I would say is departed from that has been the NoSQL databases that is really focused on allowing you to work with data of a different shape than just a table. SQL is really oriented to working with data in tables,

and that's great for a lot of transactions and things. But if the environment is dynamic, if you don't know what data you're going to see ahead of time, SQL is not as well suited to that because it really wants to structure information. So when you're working in a dynamic world, say you're doing a chat application or something like that, SQL does not turn out to be an appropriate solution.

And there are alternatives, Mongo and other document databases being an example of NoSQL that's used for that. So that's probably the biggest area, I would say, of where there's been specialization is when the data doesn't take the shape of a table. Yeah, that makes sense. And then you mentioned how sort of the introduction of the relational database, some of the work going back that you did at Microsoft, helped sort of democratize data. And I think we're seeing a similar trend in the world of AI now, like the GitHub copilots, the chat GPTs, OpenAI, these are helping democratize AI in many ways where people

are able to just hook into an API and do things that would have been magical just a couple years ago. And when you were writing your book about the entrepreneurs of data, I assume that that was in conjunction to this explosion in AI. I think you even mentioned that ChatGPT had only been out like for a month or something like that at the time of writing. So how did some of that influence your, you know, writing as you were starting to put together sort of the, you know, finishing touches on the book? Did you have to adapt in the real time to all this stuff that was sort of happening in Gen I to make some of your, you know, predictions and talk about the future in this space? Well, for sure. Yeah, absolutely. There's no question I did. Because I did not expect to see the advancement, the speed

of the advancement of the large language models and AI like we have seen it over the last year or so. It's, I think, been somewhat breathtaking to all of us and the potential of what it can do. At the same time, we now are a little bit further on, so some of the limitations are also being understood in the technology, as is almost always the case when something new comes out. But, you know, when I looked at the book, when I started the book, this idea of constant technical progress, you know, built by people that are, you know, that have a set of important values that drive them, that was sort of the central theme of the book from the very beginning. I had this idea and had created this arc of data innovation, which is the set of technologies that have been developed over a period of about 40 to 50 years

that have continuously increased the speed of progress that we see in the world around us. Really, data technologies are at the root cause of that. And we are in a world where technology continues to increase progression faster and faster. But I did not see how fast some of the improvements were going to come in artificial intelligence. And this idea that we might achieve artificial general intelligence, say within the next decade, was not on my mind. I always believed it was coming. Now, I want to say that that's important because I, you know, from my teens and on, I've always believed that the world was progressing forward in technology.

And ultimately, we would build machines that had the level of intelligence of people and even beyond that. But I honestly didn't expect it to happen until like 2100 or something like that 2050. I did not think I would see it. And now that you know, now that it's happening around us, it's it really changes things. And it just shows how fast things are progressing. And I think it's only gone faster. I mean, since the book came out, you know, I finished the book in February. It came out in June. But you have to finish writing.

One of the things about a book is it's very much set in a period of time, and it's not really very changeable at that point. And this is a topic which changes literally every day and every week. So it's been exciting to watch it evolve since then. And certainly things have not slowed down. If anything, they've gone faster. Yes, you mentioned this, you know, now you feel convinced that we're going to see AGI and, you know, in our lifetime or like by essentially, you know, 2030 or so. If you look at, you know, the early work by, like, you know, pioneers of AI, like Claude Shannon, Marvin Minsky, other folks, you know, they were predicting that we'd have human-level intelligence in, like, the mid-1960s based on their work in the 1950s.

AI has been predicted, you know, this has been predicted so many times, and it's one of those things that's always pretty much further away than the prediction. But things are different now. I think things are different now than they used to be. Yeah, I mean, there's the famous quote about how we, you know, humans tend to overestimate technology in the short term, but underestimate it in the long term. So there's probably some of that going back in the 1950s, while those guys were, you know, hanging out at MIT thinking about the future. But I guess, like, what has sort of convinced you that this is going to happen, you know, so quickly? Like, what is the kind of, I mean, is it, does essentially all the things that we're seeing in DNA just feel so much different than what we've seen previously in the world of Michigan? I think that this idea of the

models that are being built, these neural network models are technically directionally correct. Now, whether the, you, whether the algorithms inside there will continue to be refined and improved by researchers, but I think that building this thing on a statistical probabilistic model is probably the right way to do it, which is interesting because I'm a database guy and databases are all about symbolic things.

There's always a right and a wrong answer in a database and that's not true with these networks, these neural networks. And it's not true with the way the brain works. We work in a much more dynamic, interpretive sort of way. And we see a lot of these characteristics being derived from the models. I guess what I would say is that the speed of innovation seems so great that it seems almost certain that we're going to see tremendous increases in intelligence on these models over the next few years. Even if you watch, I mean, the biggest thing that's happened since I wrote the book is the real explosion of open source models where there's just been a tremendous amount of effort by researchers all around the world adding value. Doing things that maybe the big guys like OpenAR aren't quite as

focused on about quantizing models to make them run in much smaller amounts of memory. All these things are going to be important because they will enable these machines to interact with us in much more natural ways and in ways that allow them to do things that were never possible before. We've not seen, you know, I mean, that said, there are attributes of, you know, artificial general intelligence that are certainly not present. You know, I talked about, you know, the attributes of artificial general intelligence, you intelligence, the ability to sense, to learn, reason, plan, adapt, and then act. Honestly, most of those things need either the models don't do them today or they have a lot of improvement.

Sensing is very strong, I think, in models because sensors are easy to build and a lot of data can come into these models. They're very good at understanding, having the perspective of being able to pick up on information. That's not to say they understand it, but it's to say they have a lot of information available to them. They're not learning the way they need to learn today. They're not learning the way they need to learn today. They are not continuously learning. They go through training sessions, and I think that that needs to change because for something to really be an AGI, the learning must be continuous. Reasoning is still weak. We are

seeing advancements potentially in that. We just heard all sorts of rumors in the last couple of weeks coming out of OpenAI about some potential improvements in reasoning and planning around mathematics. So we'll see. But these things are adapting very quickly, and changes are happening very fast. So I do believe it will happen. But time will tell. Maybe we're being overly optimistic, but I don't think so. What do you think has resulted in the speed of innovation that we're seeing right now?

I think it's sort of the natural evolution of what happens once these discoveries are made. Once the godfathers of machine learning have come up with these ideas around their own networks, and they began to be applied with more and more data. It feels like the techniques to do so continue to be refined, and the machine capabilities are now there. I mean, part of it is, for the first time, we have the capacity, these new processors, the GPUs that allow for so much simultaneous processing of data. These are totally different kinds of computers than the machines that we're using for this podcast. I mean, we're using a standard digital computer for this, where these machines are quite different in their parallelization.

So we have a whole set of new technologies happening everywhere from the silicon all the way up. And I think you'll see continued ongoing innovation in these things. Right now, the silicon in some ways is one of the gating items, but I don't think that'll be true for more than 18 months. And we'll have several, many, multiple companies competing with NVIDIA to provide chips to supply this and all kinds of vendors building things. I mean, I think it's a far better world that this technology is being developed

and created by many different people in many different organizations. And at least in some cases, being done in an open source way than if it was all done in an open source way, than if it was all done in a proprietary way by one or two companies. I'm really pleased to see that development happen. Yeah, absolutely. I mean, I think from an innovation standpoint, competition's generally good. And then also from a diversity, safety, privacy perspective, those things help as well. I think what you hit upon there is

as well in terms of like, basically people have made certain discoveries in the space that have led to this like step function and that's happening in innovation. And then people can follow on, you see similar things in the history of science as well. When a discovery happens, then there's this kind of like decade long innovation that follows up. Or if you use like a sports analogy, it's like, you know, the first time Roger Bannister broke the four minute mile barrier, suddenly everyone knew that, oh, this is possible. And then suddenly everybody started doing it, right? The other thing that's important is that, you know, one of the attributes about the way silicon computer technology works is, you works is a very small number,

handfuls of companies build the underlying technology, the fabs, the chips, all those sorts of things that enable a wide industry to be created. But once those chips are available, they're available to everyone simultaneously. And so everyone has the opportunity to do the same innovation they just have their own creativity and innovation that they need to drive and and you know once you as you say once these discoveries have been made the ability for the

snowball to start is created and i think we're very much on that right now okay so i want to go back and and talk a little bit about your time from at Microsoft and Snowflake. So you spent, you know, 23 years at Microsoft in a variety of different roles, and then five years or so at Snowflake as the CEO. Can you talk a bit about how your journey in technology across those companies, and how did those experiences kind of influence some of your thoughts around data and AI? Well, I was essentially technically raised at Microsoft in the 1990s, in the heydays of Microsoft. As I've said many times, I saw the good, the bad, and the ugly at Microsoft. I was one of the 12 witnesses that testified during the DOJ trial.

And when you participate in something like that, it leaves an indelible impression on your mind. I also flew back to DC every quarter after the trial was over to continue meeting with the DOJ and the judge to ensure compliance associated with what we were doing with the documentation. So I was very actively involved in all of the challenges that Microsoft went through. And I think I saw Microsoft mature in some really key ways. I mean, we had some incredible technology and a lot of great people. But in the early days, I think our culture and our values were a little freewheeling when compared to the way the world has expectations today.

And to be fair, we learned that in the process. This whole idea of what was valid and what wasn't valid was very much that law was established as a part of the antitrust case that was brought against Microsoft. So the understanding of how you do things really grew out of that. And I think I gained an appreciation through my time at Microsoft and then subsequently on the role that values play in the way products are created. And I very much see values at the center of products. And when we went to build Snowflake, it was a situation where we had incredible technology with a lot of potential. But it was also important to build a company that people would want to work with, and that's where the values came in to play.

And I'm very proud of the fact that people like working with Snowflake, and they respect the company for what it does, and I feel that's very much part of the values that were created early on. Yeah, you talk a little bit about that in the book, creating these values early on at Snowflake to help keep the early-stage culture. Companies go through all these different growing pains. It's much easier to have a certain type of culture of 25 people, but when you're 2,500 or 2,500 people, things get a lot more challenging. It's one thing to create these values, but how do you actually maintain them? What goes into going from some values that you put on a whiteboard to actually making sure that people are following those things? When you're hiring, they fit into that value system. Well, it's always interesting to see whether companies

are truly values-based or not. And I think, you know, ultimately, you know if a company is values-based if the employees reference the values as a part of their thought process in making decisions in their day-to-day job. If no one knows what the values are and they pay no attention to it, regardless of whether it's on a whiteboard or on the wall, whatever, it's meaningless. But if the people inside the company really think about it, then it is a values-based company. I think that begins at the top and it requires constant reinforcement from senior management. Every time I did a team meeting, I did an all-hands team meeting on average about twice a month while I was CEO. I always talked about the values of Snowflake as a part of that.

I look at other companies that are values-based that I respect, one of which is Amazon. I'm not sure I agree with every one of Amazon's values, but that is a values-based company. The leadership principles that Jeff Bezos set there are something that very much those people at Amazon live with. That company is very, very customer-focused. There's a lot to learn from values-based, customer-focused companies. I mentioned Amazon. Nordstrom is another really good example of a company that's truly a values-based, customer-focused companies. I mentioned Amazon. Nordstrom is another really good example of a company that's truly a values-based company. Retailers, I think, have an interesting tendency to do that because they're directly competing for customers. So values turns out to be very important there.

Yeah, one of the things I've always really liked or respected about Amazon is they're pretty clear and upfront when you're interviewing there. You're thinking about it, about who they are as a company. And, you know, it's not going to be a fit for everyone, but that's okay. It's better for you to… Not an easy place to work. It's not an easy place to work. Yeah, absolutely.

Yeah, a lot of people don't make it past six months, but they're very clear about this is, you know, our expectation. This is who we are. And it's kind of nice in some ways to know a little bit about what you're stepping into. Maybe you underestimate what's going to actually happen when you do that. But they're very clear about who they are. And they've done a lot of great things for the world. In terms of both the advancement of cloud technology, just the advancement of retail in general. And again, when I look to companies to hire people from, Amazon's one of my favorite companies to hire managers from because people learn to manage

at Amazon and they learn to lead. And there's a lot to be said for that. Yeah. I've heard people refer to Amazon as like the land of the 5,000 CEOs or something like that, where essentially each little product. Well, yeah, the land of the 5,000 CEOs or something like that, where essentially each little product. Well, yeah, the structure. I mean, it's interesting, the structure that Jeff put in place with all those product leaders, each of which essentially having their own P&L structure. There's an enormous number of P&Ls in Amazon, and that works for them. I don't think it would work for every company, but it works for them and the structure they put in place.

Yeah, so during your time at Microsoft, you worked alongside people like Bill Gates and Steve Ballmer, hallmarks of the industry. Very famous people that have changed the world of technology, changed the world in a lot of ways. How did that experience influence or shape the way that you think about leadership and even some of these things that you're thinking about in terms of these value systems? I think in every way, to be honest, because that's what I was raised on. The expectations of those leaders were something that was ever-present at Microsoft. One of the interesting things about Microsoft is that leaders at in the technology group in the product groups are very technical at microsoft and they know their business and they know their products better in some ways than than the people underneath them so microsoft requires you to be able to

understand the at least as to be a senior leader at Microsoft, Bill and Steve required you to have an idea of the full picture and how all the pieces fit together. But they also expected you to have a command of the details. And there's many, many infamous things that Microsoft would do. One of the most interesting of which was the mid-year review process that would go through the sales organization. They no longer do this, but you would spend five to six weeks on a tour around the world to look at every country and the results of every country. And it was a deep dive into the financials of every single country. And it was a grueling experience.

I mean, it would start at 8 in the morning and end, I swear to God, at 2 o'clock at night and then start up the next day at 8 in the morning. And it was a constant, you know, you needed to be on the whole time because Steve, in particular, when he was doing that, was on. And Steve has this machine-like ability to deal with numbers and understand numbers and how they relate together. So he was able to ask questions at a very rapid rate, and it would cause many challenges. Sometimes it was rough. I mean, Microsoft culture could be really rough. In those reviews, people were sometimes torn to shreds.

It was not uncommon that general managers were fired after that review. That was not an uncommon outcome. So it was a tough environment, but it really required a level of discipline. Again, there's some things I wouldn't – I'm not sure I agree with some of those aspects. I certainly don't agree with having a really negative review in front of people if you can avoid those sorts of things. But in general, I do think that forcing people to operate at a high level of thought and to be prepared is a lot to be said for that. So I learned all of that at Microsoft, and it's something I carried with me. Yeah, so if you go back to some of the early days at Microsoft and of course other companies like Leaders,

Bill Gates, Steve Ballmer, Jeff Bezos, the list goes on and on. They're somewhat notoriously known for being somewhat difficult to deal with, at least back then. They're all difficult. Yeah. So I guess, has things changed, I guess, is that because of, has things changed I guess is the bottom line, or do you need to be that way in order to create these types of companies? It's a great question, right?

First of all, the ones that are successful continue to have multiple sides to them from what I can see. Most of these people are very, continue to, you know, they continue to have multiple sides to them. And, you know, from what I can see, most of these people are very, very driven, and they have, you know, their idiosyncrasies as well. Certainly, contemporaries like Elon Musk are no less difficult than those guys were. So, but I don't know. I mean, I think the characteristic is very driven people that have objectives. And in some senses, they're going to focus on achieving those objectives. And that sometimes requires people to make difficult decisions and sometimes be difficult. Sometimes it helps because I really do think it helps people to push them to achieve more.

I mean, I wouldn't say it's the way I would do things. I mean, I say all this, but that's not how I ran things, to be straight. I don't think people would say I did that, although I could be difficult in reviews, too, if people are not prepared. Generally speaking, I don't think I was quite as difficult as some of those folks are. But on the other hand, I'm not as successful as those folks were either.

So I spent a lot of time thinking about this. as difficult as some of those folks are. But on the other hand, I'm not as successful as those folks were either. So I spent a lot of time thinking about this. How big of an asshole do you have to be in order to be super, super successful? It's a great question. Probably a little bit more than I am. Yeah. Well, you probably need, I mean, I think a lot of it comes down to you need, at the very least, to have a very high bar in terms of the expectations that you're putting on people so that people are forced to meet that bar. And sometimes if they're not meeting that bar, you need to give them candid feedback.

And a lot of times you don't have time to necessarily do that in the nicest way. So I think there's probably a balance there. I think it would be unfair to the success of these people to just say the correlation is be a giant asshole and then you'll turn into Bill Gates, right? It's a little oversimplified. For sure, it takes more than that. However, the attribute does seem to come with those folks in my experience. tell you is that really brilliant technical people, while they don't tend to have the same assholishness as Steve and Bill and some of these others did, they are difficult too, in my experience. It's very rare working with these brilliant technical people that they're just the

same as everybody else, let's put it that way. People who have these brilliant capabilities, you know, have their attributes that that comes through in their personality. And I've learned, you know, over time to really work with a wide variety of different people. And one of the most important things I'll just say is just trying to maintain relationships with people through all of that. That's, to me, something that's very important, that no matter what the issue is at hand, the relationship with the person is actually almost certainly more important. Yeah, absolutely. So you mentioned how prepared you needed to be in this deep dive that you would do where you're doing this world tour. Was that also one of your motivations for setting up your

own cloud data center in your house that you ran so that you could essentially understand at a deep level what your customer was going through and also how all the products worked? Yeah. When I was running Windows Server, this was back in 2003, 2004. I was working on a new home here. I decided that one of the things I would do is put a server room in it and really try and run myself as a medium-sized business. Not a small business, a medium-sized business. We'd built this product called Small Business Server that I was pretty proud of. It made it very, very easy for a small business with one server to put everything together and run their entire business on that server. We sold many hundreds of thousands of them. They're very, very popular.

They're still running some businesses today. It's pretty old stuff at this point, but it really modernized a lot of small business. On the other hand, if you were a bit larger of a company, you had to deal with all the complexity of all of the different Microsoft products. I mean, small business server put all those things together. It swizzled it in a way that made it work. But if you, a little more than that, you had to deal with it all separately. And it was really, really hard for people to work with. And so to understand that better, I at one point had 11 servers running in my house, running a whole set of different tools, Windows Server, SQL Server, Exchange, all these things. And management, the system center management tools, which were underneath me at the time. And what I found is, you know, I would be asking all these questions of engineers and many of which, you know, I was very good at finding problems that they had not found before. And so I found all kinds of bugs and things and certainly found how complicated it was

to set this up. And so I put myself in a situation where I had, you know, in some cases when I would go into a product review, I had more knowledge of the actual use of the product than some of the people that were building it because I'd spent the weekend trying to make the darn thing work and usually failed in the process, usually had problems. Now, you know, in retrospect, it's a white elephant. Nobody needs a data center in their house anymore. Everything's up in the cloud now. So I've still got it sitting there. It's a white elephant. Nobody needs a data center in their house anymore. Everything's up in the cloud now. So I've still got it sitting there. It's a white elephant in this house. Someday it'll get turned into a bathroom, but for now it's still sitting there. Well, I think one of the challenges

that a lot of organizations face is that a lot of times the engineers that are working on the product, they're not necessarily the consumers of that product unless they're working on a dev tool or something like that. Especially server products. Yeah, exactly. So they're not necessarily interacting with it the same way. There's many, many people that work on the Android operating system that have never built an Android app. So they're not going to necessarily interface with it the same way that a consumer is. So it's easy for them to miss certain things or assume something's easy when it's not actually easy to do. Yeah, and I think what that really just shows is just having the end user perspective is always important, whatever that means.

In this case, the end user was an administrator inside an organization. But my target was not, you know, my target was, and my concern was all those companies that did not have the skill and expertise to hire the people who understood all of this, but the companies that still needed to use the products and make them work. And it really speaks to the fact that we need to make these things simple. Now, what really solved the problem, or at least brought the world forward in a huge way with services. You know, in the days when I was talking about servers or small business servers, you know, you had to install the whole thing in your environment. Now, pretty much all you need to do is have network connectivity.

All you need to do is be connected to the internet, and boom, boom, boom, with a few SaaS applications, you can set up a small business very effectively today. Things have gotten so much easier because of services, because all of that setup and much of the administration has been taken off the burden of the end user or the administrator. So the world's changed in a very positive way with services. And I think that has been one of the things, you know, you talk about how technology accelerates. Well, services means that the technology is accessible to many more organizations at what is effectively a much lower cost. And so the technology allows more people to get the benefit, and that's what continues to speed up everything around us. Yeah, and I think that's why you also see a significant growth in technology companies too, because

essentially the go-to-market speed is exponential compared to what it was 20 years ago. You don't need to spend the first two years buying hardware and setting up an on-prem data center with a bunch of boxes in a closet. Snowflake never had a data center. Everything was always in the cloud. It would have been so much more complicated to have to do that ourselves. So much more complicated. You mentioned Snowflake. You joined Snowflake as the

CEO after a stint at Juniper Networks. What kind of sold you on the vision of Snowflake? I'm sure you had lots of different opportunities you could have explored. What was it that compelled you to join Snowflake as CEO? Well, I left Juniper in late 2013, and at that time, what I saw was a major transitioning happening towards the cloud. The one thing I knew for sure is the cloud was going to be successful. And the question was what technologies were going to propel the cloud forward. And I'd come to the conclusion that I wanted to build something, something earlier stage.

When I was at Juniper, Juniper was a company that had a lot of good things, but also had made a lot of mistakes along the way. I was trying to fix a bunch of the mistakes that had been made. I came to the conclusion that while that's a very worthwhile exercise, it's not what I wanted to do. I wanted to build something new. I I thought about, you know, I wanted to join a small company and I was looking for something where I saw, where I would see a material ability to impact the world. And when I, when I heard and met with Benoit and Terry in my first meeting with the team, I, when I listened to what they were building and I understood it, I realized that if it worked, and at that point it wasn't really clear it was going to work because it hadn't happened yet, but if it worked, it was going to be industry impacting because no one had ever built a SQL database that could scale like Snowflake

could scale. The technology behind this allowed it to scale. I'm a reasonably technical guy, and when I listened to those guys talk about what they were doing and how they were doing it, it made total sense to me. It seemed like it could work. And they had done the experiments to actually validate that you could move the data fast enough and all of the technology could work. The pieces could fit together. And so that's what really sold me on Snowflake was the fact that it was going to be game-changing, and we just needed to make sure it worked and then execute effectively.

And part of all of these things is the product is at the center of it, but there's also the company and the execution around it that's also very important. So that's really where I focused. Yeah, so we were just talking about how Snowflake never had its own data centers. And it made this choice of, or I guess their vision was to be this cloud-first data warehouse, eventually much bigger than just a warehouse to the data cloud. But was there ever tension at any point early on where I'm sure people were like, hey, we'd love to sign a contract,

but we need an on-prem system. So I could see a sense of feeling like maybe you're leaving some revenue on the table while holding steadfast to this vision of the bigger thing. Was that ever a consideration of deviating from that vision and offering some solution? No, because we didn't know how to build this. I mean, literally the only way we could build it. Snowflake was only made possible because of AWS at the time. The only

cloud that could possibly support Snowflake in 2014 was AWS. Google Cloud was basically non-existent at that point. It was barely existing. And Azure was very early and had enough challenges. There's no way we could have hosted on Azure. And really, the thing that allowed Snowflake to work as well as it did was the fact that S3 is such an amazing product. And the fact that you had the ability to store an effectively infinite amount of data very durably. I mean, S3 never forgets anything. It never loses anything from what I can tell.

And, I mean, it really is. They say it's 11 nines. It's more than that. It's more than that. And you could get the data relatively quickly. And the fact that it worked the way it did was what made Snowflake possible. There were three things, three technologies that came through in the cloud and sort of at the same time that made it work.

One was blob storage, infinite blob storage. And the only reliable blob storage in 2014 was S3. The second thing was virtual compute on demand, the ability to spin up a virtual machine and then let it go away. And again, that was pretty early days for that. And then the third thing, which is not as exciting, but turned out to be very critical, was 10 gigabit Ethernet. Because until you had very fast Ethernet speeds at relatively low latency, the overall performance of something like Snowflake wouldn't work. And it's the confluence of those technologies coming together that allowed us to really make it work. So the cloud was the only option. Now, we had customers telling us we need

to go on-premises all the time, but we always just said no because we didn't know how to do it. We didn't know how to do it. Yeah, I guess that's a good reason. Even today, I mean, it would be difficult to replicate snowflake on premises i mean it's undoubtedly doable you could put the pieces together today but it would be very difficult yeah it's probably difficult for the client as well because oh my god sacrificing the services right hey you don't want snowflake is always describe Snowflake as an aircraft carrier battle group. I mean, it it is the infrastructure to bring Snowflake into a region was very significant. It's not a small thing. So, yes, you don't want to run it yourself.

I mean, there's no way anyone ever could. Why do you think, you know, the incumbents at the time, like AWS, Microsoft, Google, like was Redshift, there was BigQuery. Why did they miss the Snowflake opportunity? Oh, they didn't know how to do it. I mean, hey, you've got to separate these things into different things. So Amazon had built the cloud, and then they acquired Paracel technology, which was the fundamental technology behind Redshift. You know, Redshift was, I've said many times, Redshift was Snowflake's best friend.

Because Amazon and Redshift proved that the cloud was viable for data warehousing. And Redshift is actually quite a good product. It just didn't scale any better than any on-premises product did. It was essentially an on-premises product just ported to the cloud. So it had the same characteristics that any other on-premises database had, and it was pretty good. It wasn't by far the best. At the time, there were better data warehouses. Teradata was a better data warehouse, in my opinion. There were others that were also good.

But the thing is that Redshift did, it was cheap, cheap, cheap. I mean, the price of Redshift was a fraction of the price of NetEase or Teradata or any of these other guys. And so they validated the fact that the cloud could solve a set of problems. Again, they didn't solve the fundamental scaling problem. What happened is because people on the cloud would have a tendency to want to work with a lot of data. When they would move to Redshift, they would have an initially successful experience. But then as they grew, they would hit a wall. And that wall was, you know, difficult to penetrate. I mean, it was just as difficult. Now you have to make multiple

systems, the same techniques you've been using for, you know, in the past. And when that and so what was what was able to happen for us is we would go in behind for those customers that were having difficulty scaling Redshift and just solve the problem for them. You know, Snowflake was the scaling solution to Redshift. And so it was a great benefit to us. Amazon wasn't very happy at the time. But in the end, I think it worked out great for both companies. And, you know, Amazon and Snowflake are good partners right now. Google was a different story. I mean, Google had Spanner. I mean, Google always had, first of all, Google and standard databases are not, they don't go together. Google always built their own data infrastructure

in their own way. So they built Dremel, I'm sorry, I said Spanner, Dremel. They built Dremel as their huge query engine and they turned that into BigQuery. But even up until 2017, BigQuery barely supported SQL. It had its own proprietary API, proprietary language, and SQL was a secondary language. It wasn't until 2017 that they flipped over and made and made sequel their primary interface and it was

only at that point that they really started improving the product to compete effectively with with snowflake and i've often said if bigquery was a amazon product snowflake would not have been as successful as it was because bigquery you know bigquery wasn't sequel in the early days but the darn thing scaled and, just like other stuff Google does. As Google turned it into SQL, it's become quite a competitive product. Yeah, I think Google obviously is very good at scale, but they're typically building a lot of these things to satisfy internal requirements and not necessarily thinking about how does a third-party engineer consume this product in a way that makes sense. Google is different, and you do things the Google way. Absolutely.

That's one of the advantages Snowflake still has, because BigQuery still tends to require you to do things the Google way. And if you adopt the Google way, you can be successful, but it's not necessarily the way that most companies operate. So there's sort of the convergence of all these different technologies that I think is leading to this speed of innovation that we're talking about, the step function, GPUs, cloud, all the data that products like Snowflake, Databricks, and so forth now allow anybody to store essentially infinite data.

Data is the love language for AI that's leading to all this innovation in this space. So in terms of, I guess, looking out on potential negative consequences of this innovation, what are your thoughts around the impact to jobs? I think Mark Andreessen wrote an article earlier this year about how AI will save the world. One of the big things he talks about is how technology always creates more jobs in the long term. But it's hard to ignore that there could be some short-term impact around certain types of jobs. What are your thoughts around what sectors might be impacted in the short term, and how will this kind of shake out maybe a little further down the road? When it comes to what Mark wrote, I agree with what he wrote in the long run, but he really, I felt like he gave short shift to the displacement that occurs, the human displacement that occurs in any technological revolution. Every technological revolution that's ever happened

has displaced people, has been put people out of jobs. The thing that's different this time is the speed at which it's going to happen and the number of different places that it's going to impact. I mean, I think we will start to see jobs that are light, what I would call light administrative jobs get replaced by these artificial intelligence systems in the medium, short to medium term, say the next two to five years. I don't really think that we're going to lose

any, you know, in the short run, I'm not sure any jobs. I think there's a lot of noise and this is going to happen and a lot of, you know, there'll be individual impact on people, but I don't know that in the short run we're going to see any jobs be impacted. But the jobs that do light administrative work that are basically connecting X to Y and just doing so with a few emails, those sorts of things like administrative assistance,

I think we are going to see some displacement associated with that in the short term. I would say in the medium to long term, the thing that's going to be very interesting is going to be how robotics will begin to implement, begin to affect everything. Because I really do believe that the next five years will be about intelligence and computers gaining intelligence. But then the following 10 years is going to be all about robotics and how robots are going to impact our lives in every single possible way. And they're going to be replacing

more and more work that people don't want to do with manual labor, with robotic labor. And so I think we're going to see a lot of that. I do worry, I mean, when the cars are good at driving themselves, which hasn't happened yet, but honestly will happen within 10 years, that's going to have a major impact. On the other hand, you know, it's hard to find long distance truck drivers right now. And that's probably the first type of driving that will be replaced by robotics. And in a lot of cases, there's just shortages of people. You know, I was at a hotel recently and I waited on hold for 20 minutes just to find the right person to talk to. That job is going to get eliminated.

I mean, that job needs to get eliminated because very clearly there isn't anybody to do that job right now. Yeah, and businesses, it's so expensive to field those calls. In the short run, I don't feel like it's going to eliminate any jobs because we have a shortage of labor and a lot of this labor that can be replaced in the short run is the stuff that people don't want to do anyway. But I worry in the medium to long run fairly significantly that there'll be huge disruption to people. And I don't know what to say about that, honestly. Honestly, I don't know how to avoid it. I know it's going to happen. I hope that it happens in a way where more opportunity opens up for these people. But again, the challenge tends to be not so much the young

people who have their whole lives ahead of them and have an opportunity to learn new skills. It's more people that are closer to my age, close to retirement, that have been doing something their whole life and that thing is no longer meaningful. What are your thoughts on some of the fear around the existential risk of AI? Do you feel like any of that's legitimate, or are you sort of err on the side more of being hopeful and optimistic? I think it's all legitimate. I mean, I think it's all legitimate to talk about that. Do I think it's going to happen? No, I don't. I mean, I don't think that's, I don't think the Terminator scenario, you know, that are one of the variations in science fiction. I don't see

that being the outcome of things. I do believe that we as people are evolving. I mean, we are evolving at an amazingly fast rate. And we don't realize this, but, you know, we're already beginning to merge with these computers. I mean, how much time do you spend, you know, looking at an electronic device every day versus talking directly, you know, to a person? It used to be, you know, when I was young, you know, I watched TV, but most of the time you spend talking to other people and friends and people over, now things are shifting so that the communications are happening electronically and more and more we're becoming consumed by these devices. So I think we will evolve in ways that are very, very significant. Do I think that the chips will get implanted in the brain? Yeah, I do.

I mean, I don't think that's crazy. I mean, I don't think I'm going to have one, but I suspect my grandchild will someday, or certainly their grandchild will. And I think that more and more will have connections to these external systems that become to dominate more and more of our lives. So as we start to wrap up, is there anything else you'd like to share, Bob? Well, I guess the one thing I would say is that I am optimistic about things. I mean, when you look at it,

you hear about all these existential problems and all these other issues, they're all very real. But technology does make the world better. And it's the single thing that has driven the productivity of people and, in general, the living standard of people to improve. And I do foresee that these things are going to improve the living standard for people across the world. I don't think that there's any question about that. Those disruptions I talked about are also a very big concern, but if you look globally, I think that the right things are going to happen.

We will see AI used for every possible purpose. And, you know, it's just interesting. I just was reading today that I think 20 countries have just put together some sort of treaty about how to work with AI and safe AI. But, you know, which is great. i mean the internet's not a safe place you know hey there's parts of seattle that are not a safe place to walk to walk through you know you don't need ai people don't need ai to create a lack of safety for others you know ai could make it worse but it will also and it will make it worse in some way i mean it's a great spam device you need to you know you want to create creative spam? I mean, for sure, AI is going to help people. If that's what you want to do, AI is a great tool for it. It'll also be a good tool to help find that

and prevent us from having to worry about it. So I see it both ways, and I think, again, it comes down to people, and it comes down to the values we create. That's what I always keep coming down and saying is it's about the values. In the end, I'm glad that we're going to see thousands of different systems with a wide variety of values. Yes, some of them will have values we don't agree with or I don't agree with or someone else doesn't agree with. But overall, I think that's the way the world is, right? And I think that this is just another tool. And certainly

not the, you know, when I consider tools like nuclear weapons, which have no positive use for humanity. I mean, AI has millions of positive uses, and it has some really negative things. So overall, I am very positive. And I guess that just says that even though, you know, people have many negative attributes, again, I remain an optimist, you know, in the quality of people in the long run. Yeah, well, I think that's a great place to leave it. I'm also feel I'm an optimist as a father of a one and three year old. I feel like it's important for me to remain optimistic for their future. But Bob, I want to thank you so much for being here.

I think you've had an incredible career. It's a great book. We'll link it in the show notes. I highly recommend that people listening to this read it. And you're welcome back anytime. Great. Thanks a lot. I appreciate you. Thanks.