EP. 41 – Two Decades of Cloud Evolution with AWS’s Jeff Barr

00:00 - Introduction & Early Career

Matt Pacheco
Hello everyone, and welcome to Cloud Currents, a podcast that navigates the ever-evolving landscape of cloud computing and its impact on modern businesses. Today, we're honored to have Jeff Barr, the Vice President and Chief Evangelist at AWS Amazon Web Services, to join us to talk about what promises to be a great conversation about past, present and future trends relating to cloud computing. Jeff has been with AWS for a remarkable 23 years, joining Amazon Way at the beginning before we even know AWS as it's known today. As an author of the AWS News blog since 2004, Jeff has written nearly 3,000 blog posts and has become one of the most trusted voices in cloud computing and the entire industry. His unique ability to translate complex technical topics into accessible insights has made him a sought-after speaker at many conferences across the globe.

You've probably seen him in today's episode. We'll explore Jeff's incredible journey and his early days at aws, dive into how cloud computing has evolved since then, discuss the revolutionary impact of AI and how it's affecting software development and pretty much everything else. And we'll get into some of your insights on emerging trends, the democratization of development through AI, and predictions for what happens next. So, Jeff, welcome to Cloud Currents. We're really excited to have you on today.

Jeff Barr
Thanks, Matt. Super happy to be here. And we've got quite a number of awesome things to talk about, so this should be really fun.

Matt Pacheco
Yeah. So let's jump in and let's start all the way from the beginning. Can you walk us through your journey in your career and tell us what stops you made along the way, Things you learned and got you excited to get you where you are now.

Jeff Barr
All right, so I, my tagline for my career is a long chain of highly improbable events. And so I, I always like to think that you, yeah, there's this interesting concept of being lucky, but I think you can also create luck by being energetic, by being interested, by kind of deciding on situations you'd like to find yourself in and then making sure if you find yourself there, that you take that perhaps like goal changing or life changing kind of action. So from a very young age I always was interested in tech and I remember I must have been about 9 or 10 years old and seeing the movie 2001 A Space Odyssey when it came out and saying, wow, a spaceship and a computer. I definitely need to have one of each of those at least.

And I still don't really have the spaceship, but I've got The lots and lots of computers. And so I always was just deeply interested in understanding. And that started kind of with math and electronics and digital logic was never super good at math, but I was always able to actually figure out programming pretty quickly. When I was in high school, I was lucky enough to get to work part time at one of the first actual personal computer stores in the country. This is back in 1976. This is before things like the IBM PC came out. This is what the original days of the Altair and the inside computers. And my official job was to actually take all the boxes of the books and magazines and literature coming in off of the, coming into the loading dock.

I was supposed to like cut those open and put them on the shelves for our customers. But I was so excited by this industry that I actually spent a ton of time reading all of this content and absorbing it and making sure that this exciting new field I was really understanding the hardware and the software and the companies and the people and how it all worked together. And what that meant was that when employees would come into the store or customers would come into the store and say, hey, I've got this computer and I want to put this board into it, that I was the one who'd actually done my homework and I understood how this actually all work together.

And despite the fact that I was like just this kind of like long haired kid in the corner, I was the one who could actually explain all these things and I'd taken the time to understand it. And since then I've done all kinds of really cool stuff. But this idea of understanding something really well to the point where you, I, and I often think of that this understanding is actually explaining it to yourself. If you can't explain it to yourself, you can't explain it to anybody else. And so I found my niche in this developer advocacy or evangelism, whatever you'd like to call it, where you effectively, you've got one foot in the tech and maybe your whole leg is in the tech because you actually have to be deep, deep into it to be taken seriously.

And the other leg, you are in marketing effectively. And so you're, you're doing exactly what I did 50 plus years ago as a teenager. You're understanding something, you're probably, you're reading the docs, you're looking at the code, you're building things with it, you're opening it up to look at, see what's inside and then you're Telling people about it and it's like kind of it. It sounds really simple and straightforward and you can come up with all these really deep strategies and, and ways to do it and. But sometimes it's. Just understand it and explain it is effectively all it takes to vastly oversimplify it. But to take it down to like what's at the heart and what makes this evangelism job so much fun.

 

05:38 - Building AWS

Matt Pacheco
That's so cool. So you were there at the very early days of a US. Could you paint a picture of what those early days were like as you were helping define what cloud computing would eventually become?

Jeff Barr
Yeah, so just I got to Amazon by a really interesting route. Before I actually worked at Amazon, I'd spent some time consulting in the very early web services space as existed around 2000, 2001. So the. In those days we had all these really complex protocols. We had SOAP and WSDLD and udi UDDI and all these very complex ways to get to web services. And I was consulting with all these customers and it was very hard to actually explain the business value of web services at that point because you got some mired down in protocols that all you could really do is you could show someone, well, here's a computer and over here on the Internet there's a server. And we actually send a message back and forth across the Internet. And as technology we're like, whoa, it's amazing. We made these two things connect.

And you show this to investor, a business person, like, what's the big deal? I don't see anything all that magic there, like what's the actual business value? Because the web services of that time were like get a stock quote or get the weather or do a currency conversion. And it was really hard to actually inspire people to say, okay, well you need to take a leap from these simple things to something that could be like, actually like a value to your business. So that's the scenario around, let's say late 2001. And I'm deeply immersed in the web services space. I'm following what's going on. And at some point I find this notification that says Amazon now has xml. And I don't quite remember exactly how I learned this.

Just one of the many feeds or mailing lists or whatever I was following at the time, I, I find this and I immediately hop over to my Amazon Associates account and log in. And what I discover is that Amazon has actually taken the Amazon product catalog and put a set of basically XML and SOAP wrappers around the product catalog. So that developers can make queries to the product catalog. They can get this very rich schema of data back, and they can actually use that to create websites and businesses. So I, I see that. And through again, some improbable events that I managed to get invited to Amazon headquarters to a meeting and to a, a developer event. And at that developer event, I was totally taken by the vision that they had.

And at some point during the day, one of the people, the speakers got up and they said, well, we're so excited by this very first web service that we're going to start looking around the company and start putting APIs and other things. And I hear that. And that was like that kind of absolute fork in the road, light bulb moment where you're like, whoa, this is my whole future. And I turned to the person who invited me. Her name was Sarah. And I said, sarah, I have to be a part of this somehow. And before I knew it, I was interviewing for a developer platform position, which turned into a developer manager position. I come on board. By now, we're into 2002. I come on board and own the Amazon Associates team.

And despite what I thought was this great plan to work on web services, I'm actually working on Pearl Scripts. It was a little bit of, whoa, this is not quite what I was expecting, but I'm owning these Pearl Scripts. But I've got this awesome manager. And my manager says, well, yeah, this is your official job, but we, but why don't you spend some of your time helping out with that web service effort? And this was, it was nonspecific. It was kind of a like, just go help those guys over in that corner kind of level of thing. And that again, was one of these amazing forks in the road where I started helping out the team with answering questions on forums and writing sample code and implementing new cool features for the service.

And then before I knew it, the team came to me and basically kind of dumped a conference appearance on me. And, and they said, well, we don't like to do conferences, so you're the new guy, so this is your problem. And I'm like, oh, fine, I love conferences. I'll happily do that. I did a couple of those. And before I know it, Sarah, the same person comes to me with a job description in her hand and she says, jeff, we've got this role called web services evangelist, and we've been interviewing people for months. We haven't actually found anyone who really fits the bill, but now we see who you are and what you're doing, we actually think you are, you're kind of the right person for this job. Do you want this job?

And I thought about it for a couple seconds and I talked to my family to make sure they would be okay with a bunch of travel. And by 2003 I had that role in the business card that says Web Services Evangelist. And I, I took a very simple job description of I'm going to take what we've got and I'm going to get it out to developers all over the world. And that was the kind of the, that's how I got on board. And so we've got these initial web services, we've got the Associates web service, we've got some other search and shipment web services. And the company's pretty excited about how developers are taking to these services and not just building little demo apps, but actually creating businesses around all these things. So we see this opportunity andy Jassy comes on board.

And so we have this great setup where in this one hallway we've got Andy, then we've got Andy's executive assistant. And then my office is the next one up the hallway. So Andy starts writing this famous Amazon document called the Narrative where in six pages you need to basically identify a market opportunity and you need to lay out your plan to address it and build a business. So I am lucky enough to have a one one with Andy every week. And I get to read and put comments on his narrative as he's learning what developers are all about and what web services we should build, give them all this great feedback. And we talk about this whole idea of developers and developer relations decides I'm going to start this blog in 2004. And suddenly things are moving.

We launched the Simple Q Service, we launched S3, we launched EC2. But you, the amazing thing, when you're at the beginning of these, what are now kind of historic moments, you don't actually know that they're historic or that you should savor the moment. You don't know that you should capture it. You don't know that you should like save all the documents and take a million pictures. You're just right there doing your job day by day, step by step. And it's only afterward that you can look back and say, wow, we, that was the beginning. That was this little tiny, were rubbing the sticks together, we made this tiny little spark and we said, wow, that's an awesome little spark. And we can make this Into a fire. The fire is going great and we seem to be able to do something with this fire.

Let's keep putting more fuel on it. But you, you rarely get the perspective beforehand that says, wow, this is going to be something amazing and world changing until long, long after the fact is. The, the most interesting thing I've taken away from this is you're in the middle of all this world changing chaos and you're putting all you can into it, but you don't actually know how significant it is until much, much later.

Matt Pacheco
Yeah, you actually got to my next question. It was going to be about how has, when you were just starting the blog and you were starting to do some of the speaking engagements and doing these demos you talked about, did you envision the growth that you guys would have over that amount of time? It's, it's, oh, not extraordinary.

Jeff Barr
The one thing I do remember is that in the early days before we had to be much more careful with numbers and with financial information. There was a set of internal mailing lists. It was like S3 daily metrics and a bunch of us run that list. Then there was EC2 daily metrics. And I remember probably getting 10 and then 100 and then seeing the first thousand dollar day of EC2 and thinking, wow, that's actually pretty amazing that we're selling that much. And this was at the point when EC2 was, I think it was 10 cents per hour per instance. And you start to think, wow, those are real customers and real people actually paying money for this.

And what I also remember from those early days and traveling and talking to people was, you know, I'd go somewhere and I'd talk to some, to a great audience somewhere and I'd, you know, get to my hotel and have a good meal and think I'm out there selling storage at $0.15 a gigabyte and compute at $0.10 an hour. And I'm thinking, how does that actually multiply out and add up to my dinner? Like I'm kind of trying to relate those two. I, I still remember that aspect of this weird Amazon frugality of like, okay, well something doesn't quite add up here. I'm, I'm having this really decent dinner after a long day on the road. But I'm talking about all these like nickels and dimes and pennies to actually pay for it.

And so you kind of get that weird kind of cognitive dissonance of like, how does this all fit together? But it actually did. And you got the sense pretty quickly. Like the first. I, I still have the email somewhere. Probably our first $10 day. We're like, whoa, look at that. People are actually using this thing. And how cool is that?

14:45 - Cloud Evangelism

Matt Pacheco
That is pretty cool. So you've been doing this for a while. You've been the chief evangelist for a bit. How has it evolved over the years? And also as you've grown your offerings across multiple regions and probably 30 plus regions. How, how's that changed?

Jeff Barr
Well, I, I think that's what's happened is we certainly have a lot more services than ever before. And people think that when you have the title of chief evangelist, that you somehow know everything about every service and every last nuance in every detail. And it's very flattering that they would ask those kind of questions. It's like, you know what, I'm just like you. I'm doing my absolute best to try to keep up with all of these amazing things and to try to understand and explain them, figure out what's going on. I think were always pretty careful with what we would think of as like, the messaging discipline to make sure that were accurately describing what these services were all about. And I still remember, especially in the early blog posts where I would insist on every last detail being correct.

And I continued that through the whole 20 years I did the blog was every last detail had to be correct. And it was partly out of just wanting to do the best possible product. But I had this weird fear that someone might misread something I wrote or I had the wrong facts and I could actually mislead someone and their startup would crash and burn. And oddly enough, when I first started being out in public and people would come up and say, are you that Jeff guy from aws? My biggest fear, my weirdest, totally irrational fear was like they. And I'd say, yep, that's me. Kind of carefully. I always had this weird fear that they would say, well, I followed the directions in your blog post and my startup now exploded and it's all your fault.

And I, that, that totally unfounded fear somehow kept me making, kept me on my toes for 20 years to make sure that I, I would go to extreme lengths to make sure that I wasn't just passing along things people told me that anything that I would write, anything I would say that I would verify by talking to developers or using the service myself, or pulling up, going to the lead developers and saying let's go through the code and show me how this actually works so I can be really confident that I get it.

Matt Pacheco
Yeah. I was going to also ask you, how do you balance staying technically current with all of these new services and balancing that with communication and your evangelism?

Jeff Barr
Because I know it's a real challenge as a developer advocate, because as a developer advocate you often have a background as a developer and you've spent your entire career building things and you still get to build things, but there's a little bit of, you're putting on a little bit of an act because you can't be a full time builder anymore because you've officially, you kind of turned from the light side to the dark side and now you're effectively part of marketing. And, and developers are just natural skeptics. And when you say, yep, I report up into the Chief Marketing Officer, like, oh, you're just like the marketing person. You're, you're not real.

And so you, you have to work extra hard to make sure that you are, that you're accurate, that you're honest, that you don't accidentally adopt a whole bunch of like very abstract marketing vocabulary that I, I, I try to stay very much kind of like hands on the keyboard, feet on the ground. And I, for sure I keep all the messaging straight and correct, but I don't like elevated into these very abstract kind of conceptual benefits. I like to be very direct about how things work and make sure that it's, that when developers hear what I have to say, they'll say, well, maybe that guy's not quite a developer anymore, but he knows our language. And we can look at his background and say, hey, he's not BSing us, he's giving us the truth.

18:35 - Adapting to Change

Matt Pacheco
And I'm sure as we talk about new tools and new technologies and you speak about them like AI at your conferences, there's that still that understanding that you have that knowledge and you can talk to some of those products. And I know there's probably some difficulty. There's, there's constantly new product. I, I just appreciate what you do in explaining some of those more complex tools and concepts to your audiences. I do want to talk about AI. That's why I'm shifting to AI, because I know that's what everybody's here for. That's what everybody wants to talk about these days. And I love it.

Jeff Barr
The entire DA field is, you have to be really current, but there's Always far more things than you can invest your time into. So you have to. You have to pick things really carefully. But you al. You always feel like you are. Like when you watch a surfer and they're kind of like balanced on that, Like, I don't know the exact word, but they're like balanced right there on that wave and everything's perfect. Like all the forces and the waves pushing them and they're just like teeter tottering right there on the wave, but everything's perfect. Sometimes the job feels like that where you, you step away for one second or you don't pay attention and you. You fall off and you can't catch yourself because the entire industry just kind of passed you by. So you. We're. We're always facing that.

That existential stress of keeping up and staying relevant and making sure that we're not somehow leaving ourselves anchored in the past.

Matt Pacheco
Exactly. Exactly. And there's. There seems to be a lot of those big waves coming more often recently.

Jeff Barr
And. And I love that. And one to decide. Very early on in my career, I remember being at a. At a party when I was about 20 at this company I worked at the time, and having this, what. This guy who seemed like impossibly old to me, and I'm looking back, he was probably at least 50, and I'm well past that now, but he seemed impossibly old. And were. He was telling me how his mainframes and his job control language and all that was like the ultimate in computing. And I was trying to convince him that personal computers were the way of the future and he was just not having it. And I realized he had just gotten very set in his ways. And at that point being.

I'm pretty sure I was 20 at the time, I remember just thinking, I. I just can't grow up to be like that. And nothing against him. He was. He was an incredibly nice, competent guy. But he had. He had decided to kind of anchor his. Anchor himself in the past versus really like respecting and honoring the past, but always paying a lot of attention to the future as well. And I think that's a really interesting balance that we as technologists have to really strive for, make sure that we understand a lot about what's come before and. And make sure that we say, yep, those were all great things for the time, but. But there's a cool new thing coming, and we. We can't let the.

The polish and the perfection of the previous generation kind of overwhelm the fact that the new Generation is, it's, the next thing is never perfect. When you start, it's always a little rough around the edges. It's still being iterated on to become perfect. If the developer is keyed into the customers and listening and learning, they're making it better really quickly. So you need to kind of grasp that. And of all the things that we get taught in school, the thing that you don't get is that they don't really tell you, hey, we're dropping you into this very dynamic industry. Everything's going to change all the time. And part of your duty of being in the, in this field is you need to invest your own time in making sure you keep up.

And that it's the, the change is just part of what happens here. It's not, it should never take you by surprise that something changed. Right? You're, it's just going to happen. And you know, maybe we need to actually teach people better mechanisms for making sure that they're, I don't want to say coping because it's, you don't want to cope it with it, you want to enjoy it and appreciate it and take advantage of it. But we teach people a little bit better about how to make sure that there's ways that you can introduce change into your, your working life.

22:35 - AI Revolution and Global AI Adoption

Matt Pacheco
So, so speaking of the generations appreciating what came before, what's coming, how has the generational shift in developers kind of affected how you approach your education and evangelism?

Jeff Barr
It's been really interesting the last couple years talking to developers and just hearing that on the one hand they're super excited about all this new power that they get from all the gen AI powered coding tools. And on the other hand they're saying, this is really interesting because there's a tiny bit of a threat because they say, well, is this going to replace me? Is this taking away part of my job? Is it taking away the fun? And I've spent a lot of time this year traveling. I've been to a lot of really amazing cities all over the world and I've been trying to actually paint this really neat historical picture that is one of the fun things you get to do in this industry. If you've been around enough, you get to see a lot of change.

And when I was in college, I'd already spent a lot of time with the earliest of the microcomputers, but also I get to college and we're taking a step back and we're doing mainframe programming on punch cards and I was simultaneously moving forward and moving backward, it felt like, but I absorbed so much from all those great experiences and being able to go in front of an audience and say, look, you know, at the beginning of my career, I was putting in zeros and ones on the front panel of a microcomputer and I did punch cards and I did paper tape and I did microcomputers and mini computers and all these things in between. And binary to assembly language, to compiled languages, to AI powered specifications and saying these are all effectively, they're all improvements.

And then what I always like to point out is say the, we're always starting with a great idea. We then have this intent and we want to solve the problem somehow. But what's really happening here is that the kind of, the abstraction level of that intent has just been stepping up over time. From directly putting in the zeros and ones to writing an assembly language where there's a little bit of symbolism in there with the instructions and the addresses. To the higher level language where we say, okay, well I don't really care about the, the details of which instructions to use. We trust that to the compiler, but we're saying like, declare some variables, do an addition, print out the result, let the compiler figure out all the details.

We're just making one more step forward with all of our AI tools where that intent is not expressed in a programming language, it's expressed in a human language. And I very specifically didn't say it's expressed in English because I think when we talk about accessibility and all these AI tools, you can use the language you are most comfortable communicating in, which might not be English. And I think that's actually awesome. We think about how do we get more people to build cool apps and solve problems? Well, we make it more accessible to more people around the world. And I'm seeing that these language models are, have that ability to, to admit more people into the field, which I think is just awesome.

Matt Pacheco
So, so with the English language barrier, kind of talking about how has AI kind of removed that and address that and well, I wish I.

Jeff Barr
Could understand how it truly works, but this is the. Now, now this is a really interesting dividing line for a lot of us that have grown up with all this tech is that in years and decades past, we could go down through all the levels of abstraction. We can say, okay, well I'm using a compiled language, but I can actually look at the source code of the compiler and I can turn on a bunch of flags and I can actually see the assembly language generated by the compiler. Okay. I can see how that turns into instructions, I can see how those turn into bits. If I really wanted to, I could say, well, what is the digital logic that actually make, made the processor?

And if I want to go deeper than that, it's like, well, what is the, how are the transistors work and what is like the actual like physics and electronics that they're at the level of like individual, like atoms. How does all this stuff work? And you could literally fit all that stuff in your brain up to the compiled program and you could kind of go up or down all of these abstraction levels if you wanted to. The LLMs are different and that they're trained and the output of all this training is a bunch of data that you can't really look in the data and say what do these individual values mean anymore. It's this very indirect representation of.

Effectively we've taken all the amazing stuff that we've learned as an industry over years and decades and decades, all the code, all the best practices, all the documentation, all the samples, all the commentary and we've effectively taken all that, put it in this gigantic bucket and said, and handed that off and said train on all this. And we're kind of packaging up all the knowledge but along with that, because all of the training gets turned into these vectors that are these like multi dimensional representations of information. The same training process. And I don't, this is where my understanding really breaks down. Those, the, those individual concepts that they're encoding, the vectors, they're somehow independent of a particular human language. And I, I don't quite grasp fully why I'm hand waving to myself at this point and I'm telling.

There's this big void that says Jeff, you have no clue what you're talking about here. But I, but I, at least I'm aware that I don't understand what I'm talking about. But what I do know, and this has been the most fun thing ever, as I've traveled from country to country this year, I, I have made this point of finding a local language and always. And translating a prompt from that local language and putting that into an LLM and getting a perfectly good program out of it. When I was in Peru earlier this year. So in Peru they speak Spanish, but there's this native language called Quechua that's this very obscure language that is, you know, from hundreds and hundreds of years ago. It, it did not overlap with today's computers and, you know, by centuries, it's a really old language.

But I, just for a fun experiment, and this was like my first step of my experiment, I, I used an independent translator to translate my prompt from English into Catchwa. And then, and I, I simply said, I want to add two numbers together and print the results. And then I took that, I put it into one, I think I put it into the Amazon Nova model and I said, and I put that Catchwa prompt in there and no one's ever advertised that this model knows how to speak this particular language. And I, I, but I, I don't fully grasp why it does or doesn't, but I got a perfectly good program out of it and I said, oh, this is kind of neat. And then as I've gone from country to country this year, I always would pick a local language.

And I was in Armenia and I used the Armenian language, which again, how does the language model know how to map Armenian into Python? It's amazing. I don't, I don't quite know how that worked, but it worked just fine. And the message I gave to developers is the models meet you where you're at. It is no longer a prerequisite to becoming a developer that you have to learn to read and write English, so your native language is likely going to work for you as a developer. What I do also tell people though, is that becoming a really good natural language communicator, a great reader, a great understander, a great writer, these are now skills that weren't really good at developing in ourselves and within our industry before that are super, super important.

You need to be able to sit down with your customer, deeply understand the problem that they have, and then cogently and lucidly express that problem to your programming tool and say, this is the issue and this is how I'd like to go about solving it. So we're now demanding that the big step, I think with AI is we're really saying you're, you're effectively human to human communication skills mean a lot more than ever before.

Matt Pacheco
That's really cool. And that's a really cool, interesting use case of AI that we don't often talk about that much. So you're talking about globally right now. You've, you've recently gone to, let's see, Japan, Peru, Armenia, Italy, Thailand, India, probably A lot more. Do you see any regional differences between kind of the AI adoption and cloud trends that we're seeing here in the us? Are there differences? Are there similarities? What did you notice?

Jeff Barr
I'm seeing a lot more similarities than differences. I think developers all over the world are really picking up on AI and gen and saying, wow, this is super powerful and this lets me just express myself in new ways. And I can actually, they're not thinking this replaces my need to understand the tech. They're saying, because I understand the tech, I can make use of it in more powerful ways. But people are, the developers I talk to are still making sure they understand the fundamentals, but they're saying this now empowers us in really deep new ways. And as far as I can tell, that's global. I haven't had an audience yet where they said, no, I'm sorry, but we just don't buy into gen AI.

Everybody I've talked to, I can't say I've been everywhere but the places I've been this year and last for sure universally excited about it.

Matt Pacheco
Yeah, that's really interesting. So another question, let's talk about data a little bit, because with all these AI applications and all these cloud apps, you have to consider data. It's quite important. And in the past, when we spoke, you proposed that data might be more important than apps in the future. Can you explain that concept for sure.

Jeff Barr
A little more so. It seemed to me that in times past, which we're talking about inception, up to a couple years ago, we put a lot of energy into figuring out what is our problem to solve and what does the app look like that we want to build. And we wrote these massive documents to document our understanding and to specify what we'd like to build. And by the time we actually had a running app, we made this incredible effort in time and money and people. And so that final app, we've invested so much in it, we're like, okay, it's really precious now because of all this energy we put into doing it. But now we're going to say, well, it doesn't take so long to do any of those things.

We can, we can figure out what this is and we can build it pretty quickly. And we, we're not so bought into the app itself. We're kind of, we can think of the, the prompt as really the investment is going into the understanding and into writing the good quality prompts. Actually turning the prompt into the app was maybe not quite trivial, but it was a Relatively short time period with not a whole lot of people doing it. We're not as bought into that realization of the app. So I'm thinking that this future and this is totally separate from the whole Egyptic concepts that are coming out, but just the apps themselves that we build, maybe they're useful for, I don't know, a week or a month or a year. And the next time we say, well, do we incrementally maintain the app?

Maybe we say, forget it, let's just keep those prompts and our understanding, let's build a fresh one with the newest tools, the newest technology. But I actually do think that because we're putting less energy into these apps, ultimately the data is probably going to have a longer life because we're saying it's easier to collect the data, it's easier to get it back, easier to build cool things that use it. I, I do think that we're going to say, well, let's invest in data for the long term, that we're, we're saying it's now going to be easier than ever to get value from this data. So let's put up more money into collecting it and curating it and making sure we understand what it is and putting it to use.

One awesome use case I've heard from a couple customers lately is when data warehouses first started to come into prominence a couple years ago, the kind of overarching message was, well, you don't know quite which of your data is the most valuable to keep. So just keep everything, just throw everything in the data warehouse and later you'll figure out what to use it for. Companies did that, but they're saying, you know, that's a lot of stuff. And is it. Why should we actually keep putting more kinds of things in there? What can we do with what we already have?

And so this incredibly cool use case, I heard from one of our customers, they said, well, we took our data warehouse, we found all of the tables and all the schemas that we have, and we took it and we just threw it into an LLM and said, what can we cook with all these ingredients that we already have? And it turned out that actually they got some really cool ideas for apps. And rather than saying, well, we can't do anything until we collect more, even more data, it was almost like opening the refrigerator and saying, okay, I've got some of this and some of this. And I got this in the closet in the pantry.

What can I make with this for Dinner tonight and that same approach of remixing, recombining, getting the best value of what you already have, I think the LLM is actually going to really help with that. So I see a very bright future for more kinds and probably longer lived data as well.

36:02 - Future Development

Matt Pacheco
That's really interesting. So a lot of companies are still trying to figure this out. They're not quite at the point where they're ready for AI. What is some advice you can give to people listening potentially on what they can do now to prepare their data for an AI driven future?

Jeff Barr
Well, even before that, I'd say it's totally reasonable to feel like you are behind and that you're a little bit confused and that you're not sure where anything is going because effectively everybody talk to will put on a pretty brave face and say yep, we've got our strategy figured out and this is where we're going. But then they'll also say there's so much uncertainty here and we can't make plans that are going to be 12 months, 18 months, 24 month plans anymore. We have to do things in shorter increments.

So, so there's, so they're kind of having to rethink their application development kind of not life cycle, but kind of the pace and saying let's do a lot of small to medium sized things very quickly so that we don't actually pick a direction and then end up far away from where we thought we wanted to go.

Matt Pacheco
Cool. Let's talk about the future. I love this part of the episode when we talk about trends and all the things you're excited about. So looking ahead maybe what, five to 10 years, what fundamental changes do you predict in how software is built and deployed?

Jeff Barr
Wow, 10 years is a long time right now. And it seems to me that I've probably seen more change happen in the last two years than in the 20 years before that. So try to go, multiply that by five at this point and say where is it going to go? I, I would say that we'll, we probably will continue to see so many new models emerge. I, I don't think that as an industry we can have a world in which there's a brand new awesome model every month and that we suddenly we try everything, we say, oh, this one was great, this was our January model, but we tried our prompts and here's a totally different one for February. Let's, let's try use that one instead.

I think we're going to see this Crazy evolution for a while and then probably some level of, I don't want to say static, but stability, where probably the models will probably iterate within a version versus whole new models emerging from nowhere. Like right now I can leave on a trip and one model is considered the awesome one. And then two days into the trip a different one is like, oh no, that one's better. By the time I land, it's like, yep. And now that one is, is now the thing everybody prefers. And so but it's hard for people to figure what, what will we, what can we invest in if we think everything is changing all the time? And that is certainly giving people a little bit of cost for concern of saying we can't make any long term plans.

We, we can't make any long term commitments. We, we want to learn, we want to try to stay current. So a lot of small to medium sized experiments seems to be a great way to get toward the future. I do think that all the training that people have had to date is still going to be really valuable. But I also think that we're going to see different skill sets emerge. It'll be incredibly important and the people that have a deep understanding of business and this incredibly good ability to describe what they'd want the outcome to be and the problem they want to solve will be highly valued.

And maybe the paradox, the challenge there is that some of us originally got into computers because were very comfortable with the zeros and ones and the fact that we didn't have to argue with anybody, that the computers had to listen to us and that there was no uncertainty and you either got a result or you got an error message. And the LLM world is different than that where everything is fuzzy and you can argue a little bit with these models and you can actually like back them in the corner sometimes. And I used to say, well, you can't actually, you can't force the computer against its will to like compile your code and run it.

But with some of these models you can actually put a little bit of extra emphasis and like almost insist, hey, this is the way I would have done and almost browbeat the model into doing your bidding for you. So that kind of zeros and ones, black and white on the human, you're the computer. I think that's on its way out, if not already. And like I said, there's some of us who were better at talking to computers than to humans for a while. And we're like, oh, that nice shady kind of part of the world where we're just lock ourselves in the dark room, we don't get to do that anymore.

41:15 - AWS Innovations

Matt Pacheco
Very interesting. I want to ask you what, in a, in the context of aws, what is the product or service or offering that when you go to your speaking engagements that you get the most excited to speak about? It could be anything. But I'm curious, so.

Jeff Barr
So this year I've really been talking about two things. I've been talking about coding assistance. And so we started out with the Q developer and the Q developer command line. Just last month we launched Curo, which is our newest AI powered coding assistant. And I only presented to two audiences so far, but they're super excited about Kiro and the fact that it's what's called spec driven development, where with the earlier coding tools you kind of just said, you know, make me an app and this is what it should do. CURE imposes some structure and you go from specifications to requirements and the requirements actually point back to individual parts of the spec. And then you say, okay, let's actually implement the spec and that then actually maps to a certain set of tasks that are carefully coordinated.

So starting to put a little bit of organization and structure around the use of the programming tools. People kind of get that and they say, you know, you can. The challenge that I think people saw with some of the other coding tools is you can be going really well for a while and everything's awesome. And then you put one bad prompt in and you just kind of like you fell off the knife edge. You were bounced on this knife edge, everything was perfect. You're like, wow, this is amazing. This is the best thing ever. And then one bad prompt and suddenly like, okay, this will never work. And so having a bit of structure, the ability to back up, the ability to actually put these things into source code control as you go is something that I think developers are kicking to.

The other aspect is not a product per se. There's this whole group of effectively this idea of proving programs correct, which the bigger model is called formal methods. And one of those formal methods is called automated reasoning. Automated reasoning is a technique that we've long used inside of AWS to build these services like S3 and DynamoDB. They have to run at tremendous scale and to be as reliable and error free as possible. This is the idea of proving programs correct to make sure that you don't have very deep, hard to find errors and race conditions and situations that emerge once in A billion times. Formally verifying the correctness of your program is something that is going to become increasingly important. It's a little.

It started out very academic and it's slowly making its way into the mainstream and I think we'll see that become more and more prominent over the next couple of years.

Matt Pacheco
So I want to ask another cloud question. So the cloud landscape has changed over the last few years. There have been changes, well, evolution of products from you guys, from Azure, some changes on the VMware front as it relates to private cloud for a lot of businesses. How do you see the landscape changing in the future? More. Do you see more of a public cloud only? Do you see more of a hybrid? What are your thoughts on that?

Jeff Barr
So I mostly focus on what we do in aws, but what I've been psyched about with AWS for quite some time is that we have a lot of different ways that you can actually deploy compute, power and storage. So we've got what we can think of as the traditional regions which then kind of branched off into the gov cloud. But then we've got this ability to do things like using the outposts, which are effectively kind of a server blade or an entire rack full of aws. We've got things like the wavelength zones that are within the domain of a 5G provider. So we've got multiple different places where your same knowledge and experience and training and tools that you build can be used to deploy and build, deploy and monitor your AWS applications.

So what I see is there's a lot of different ways you could put those same skills to use. I was at some events last night and people are talking about a future where data centers are built in orbit, for example. And the idea that a data center in orbit would have the same potentially APIs and services as something on Earth. That's kind of cool to hear about people talking about that as a nearer term future than I ever would have thought. Like people are actually quoting years and months for when that might happen. It's not science fiction. It's like these are things that are possible, not quite here now, but they're on somebody's calendar.

Matt Pacheco
At least cooling won't be as big of an issue, right?

Jeff Barr
It's actually very different. It turns out, it turns out that some of our cooling is dependent on having air and having gravity, so that, that the heat actually has a direction to go in. And so there's this interesting thing that getting power is actually easier that there. There's these ideas you can deploy these gigantic solar panels in space and you can get power pretty easily. And, and I, I don't fully understand the economics, but at some point the power you can get from being in like low Earth orbit is, the cost is low enough that it's going to offset the cost to actually get the data center into space, but. Which to me seems like an incredible trade off. But people who've done the math say it's actually real reasonable.

Matt Pacheco
I can't wait to see once they start taking those steps to get to that future. It's really interesting.

Jeff Barr
Yeah, the thing is you get the mind blown kind of continuously in the, in this field of like things you just thought were total science fiction turn into reality and like, you know, the work that we're doing with quantum computers and how that works going to say, yeah, it's like, and you look at these things and the quantum computers all look like some alien artifact out of a movie, you know, because of the way they have to be chilled. And there's all this, you know, copper plumbing and there's wires and all this stuff going everywhere. It looks like this thing that we just kind of found that was built by some advanced civilization we don't even understand. But nope, we actually do and we built it ourselves. And it's always amazing to look at.

47:16 – Outlook and Advice

Matt Pacheco
I mean, people used to joke that the government computers back years ago were big and took up an entire room and only ran on a few kilobytes of ram. So it's the, it's funny how it's evolved and now it's on what your little phone can do more than what those computers could do years ago. Okay, another question on advice. So people just starting in the field, where should they focus on building their skills? What would you say?

Jeff Barr
I would say that you can't go wrong with learning to be a really great communicator. And I'd also say that there's a real big. One of the really big differences with the AI tools versus everything else is that for a lot of my career, spending time offline and reading documentation and reading the manuals and reading the code was the best way to understand something. With the AI tools you could read about them a little bit, but the actual interaction and the feel and the responses of saying something and watching the responses come back and iterating that is not documented anywhere. You have to actually experience it for yourself. So going hands on with these models is super, super important. And another big Difference. And this is, I think, where some of the discomfort that a lot of us have is.

In the old days you could read the whole manual and the manual said, okay, like 1, 2, 3, 4, 5 to 100, here's all the things that this program can do. And here's all the language statements, here's all the commands, here's all the functions, here's all the data types. And once you know all those, you're like, okay, I have the whole vocabulary of this language or this operating system, but the models don't have that list. It doesn't really exist. And you, there's no list that says this is the exact set of 10 million things that this model knows how to do. And it's a lot more squishy and analog than that. The model can do a lot of different things, but it can do some of them really well, some of them really poorly.

But then it can things that maybe the first response wasn't as great. You can help it by giving it more information and it's more of a symbiotic kind of co creation relationship. Then it either does or doesn't do something. So getting very comfortable with that and saying that I'm going to. When it doesn't do something right, it's not always, oh, the computer's broken. It's like maybe I didn't communicate my needs properly.

Matt Pacheco
Excellent. Last question. If you had to restart your cloud and AWS journey from the start, what would you do differently?

Jeff Barr
Oh, wow. So the only thing I would have done a lot differently would be to make sure I capture so much of the history. You look back and say, wow, I was there for so many amazing different things to happen. And we had these initial meetings and we discussed things and we made great decisions and the result of those decisions were implemented and resulted in these awesome services and these successes. But you're so busy and so immersed in doing those things that you don't realize that you should capture it and make sure that you have this really good recollection of who did what and why you did it. And just sort of in many years when our grandchildren look back and say, where did all this cloud stuff come from? Where'd the AI come from?

That it's well documented and we've got that sense of captured history that is just so important.

 

Matt Pacheco
Need some cloud historians?

Jeff Barr
Well, the thing is you never know that you're creating that history until it's too late. That's the thing and when we can criticize anything in the history of, oh, that wasn't the way it happened. I totally get that. Because the significance of these events isn't clear in the minute. No one says we're about to change the world by launching Compute Power. Right? It's just a simple thing you put out there. And yes, there were a bunch of decisions to get to that point, but okay, here's our first EC2 instance. That's great. Then you do a second and a third. Then you launch another region and some auto scaling and some ebs and a thousand more things over the next couple of years. And then you're like, wow, look at this amazing mountain that we've climbed. But we didn't.

We were solving customer problems all the time. We weren't actually setting out to climb that mountain. It just kind of. We just did. And it's only in retrospect you realize what you've done.

Matt Pacheco
So cool. Well, Jeff, I wanted to thank you for being on Cloud Currents today. It was a great conversation. Learned about a lot about AI and development and all some interesting going on at AWS and at your engagement, speaking with developers and people across the world. So I appreciate you coming in and telling us a little bit about that today.

Jeff Barr
It's been fun.

Matt Pacheco
Thank you for our listeners. Thanks for listening in. Stay tuned for more episodes. You can find our podcast anywhere you get your podcasts and we'll see you soon. But thank you very much.