Screaming in the Cloud - Finding a Fix for the Cloud with Stephen Barr

Episode Date: November 7, 2024

Corey Quinn sits down with Stephen Barr, Chief Evangelist of CloudFix. With his extensive history in the cloud, the pair delve into Stephen's journey with AWS, relatable anecdotes on optimizing cloud costs, and the complex role of tech evangelists in fostering better communication between engineering and finance teams. Corey and Stephen also weigh the pitfalls of early AI adoption, how to come up with effective content creation strategies, and even postulate a hopeful vision of a tech-driven future (from a Trekkie’s point of view at least).Show Highlights(0:00) Intro(0:40) Gitpod sponsor read(1:52) How Stephen defines his role(4:26) Breaking down recent shakeups at AWS and the ever-growing promotion of AI(9:36) How will AI impact how we teach younger people about coding?(13:45) AI marketing, crypto, and other professional grifts(16:56) Stephen's history with AWS and the cloud ecosystem(20:42) Wiz sponsor read(21:30)Oversights that can easily inflate a cloud bill(25:32) Acting as a marriage counselor between engineering and finance(30:09 Stephen's creative process as a Chief Evangelist(33:54) Stephen's thoughts on the future of technology(35:28) Where you can find more from StephenAbout Stephen BarrStephen Barr, Principal Architect and Technical Evangelist at CloudFix, is known throughout the technology industry for his joyful frame of mind and deep expertise in data engineering, machine learning, LLMs, systems architecture, and all things AWS.Even as a teenager, Stephen’s digital curiosity and drive landed him at an email hosting startup working on network administration. He also worked at Microsoft while still a high school student.After graduating from the University of Washington, he continued graduate studies at the University of Rochester and Washington. Stephen has also worked as a data scientist, software developer, technical consultant and more.When he’s not researching or communicating about the power of AWS, Stephen enjoys spending time with his family at home in Seattle. His interests outside of work include science fiction, 3D printing, and the outdoors., Stephen Barr, Principal Architect and Technical Evangelist at CloudFix, is known throughout the technology industry for his joyful frame of mind and deep expertise in data engineering, machine learning, LLMs, systems architecture, and all things AWS.Even as a teenager, Stephen’s digital curiosity and drive landed him at an email hosting startup working on network administration. He also worked at Microsoft while still a high school student.After graduating from the University of Washington, he continued graduate studies at the University of Rochester and Washington. Stephen has also worked as a data scientist, software developer, technical consultant and more.When he’s not researching or communicating about the power of AWS, Stephen enjoys spending time with his family at home in Seattle. His interests outside of work include science fiction, 3D printing, and the outdoors., Stephen Barr, Principal Architect and Technical Evangelist at CloudFix, is known throughout the technology industry for his joyful frame of mind and deep expertise in data engineering, machine learning, LLMs, systems architecture, and all things AWS.Even as a teenager, Stephen’s digital curiosity and drive landed him at an email hosting startup working on network administration. He also worked at Microsoft while still a high school student.After graduating from the University of Washington, he continued graduate studies at the University of Rochester and Washington. Stephen has also worked as a data scientist, software developer, technical consultant and more.When he’s not researching or communicating about the power of AWS, Stephen enjoys spending time with his family at home in Seattle. His interests outside of work include science fiction, 3D printing, and the outdoors.Links ReferencedLinkedIn: https://www.linkedin.com/in/stephenjbarr/AWS Made Easy: https://awsmadeeasy.com/SponsorsGitpod: gitpod.ioWiz: https://www.wiz.io/scream

think that the fundamentals of cryptocurrency, it makes sense and I like it. But then on top of that little core of truth is this giant pile of scams and garbage and hype. Welcome to Screaming in the Cloud. I'm Corey Quinn. Thrilled to be able to talk with Stephen Barr today. We've talked many times previously, but this is the first time we're doing it in public on one of my properties instead of his. Stephen is the chief evangelist over at CloudFix. Stephen, how are you? Really well, Corey.

It's great to be here on your property talking in an official capacity. This episode is brought to you by Gitpod. Do you ever feel like you spend more time fighting your dev environment than actually coding? Works on my machine issues are too familiar and the VDI setup in your organization drives you mad? Gitpod brings automated, standardized development environments to your laptop and the cloud. Describe your dev environment as code and streamline development workflows with automations. The click of a button, you get a perfectly configured environment, and you can automate tasks and services like seeding your database, provisioning infrastructure, running security or scanning tools,

or any other development workflows. You can self-host Gitpod in your cloud account for free in under three minutes, or run Gitpod Desktop locally on your computer. Gitpod's automated, standardized development environments are the fastest and most secure way to develop software. They're trusted by over one and a half million developers, including some of the largest financial institutions in the world. Visit gitpod.io and try it for free with your whole team. And also, let me know what you think about it. I've honestly been looking for this for a while,

and it's been on my list of things to try. I'll be trying it this week. Please, reach out. Let me know what you think. I have to say that you and I suffer a similar affliction in that it is kind of hard to describe what you do. How do you define yourself as far as the starting, the stopping?

So I like to see it as facilitating a two-way relationship between our customers who are interested in cost optimization and then the engineering and the internals of the company to make sure I can be a trusted source for developers to go to, for the more technical folks within the business to go to and say, hey, does this actually do what we're claiming it to

do to be able to talk shop with the developers in our, with our clients, and then also to make sure to advocate for them. So I see the evangelism as a two-way street. It's not just, it's me doing a lot of talking and that goes through content creation, through marketing, that sort of thing, but also a lot of listening and bringing that, bringing that back to the rest of the team. I've always found that cost optimization tends to be a content area that people care about at very fixed points in time, usually when there's a burning fire, and the rest of the time they could not possibly care less. So my strategy around it has always been to talk about the greater AWS ecosystem and architectural

fun things thereabouts, and just make it a point to periodically throw some of the cost content in to remind people that I exist. And I found that that's been working for eight years so far. I almost see it as like, well, like you said, it comes up when there's a fire and almost like fitness, right? You go to your doctor and he's, hey, your cholesterol is way off. And all of a sudden you're like, all right, I'm going to the gym. I'm getting serious.

I'm going to change my diet. And then we're human. We tend to regress. Three weeks later, we're on the couch eating chips. Exactly. And so it has to be this constant pressure. And I want to see myself as being, in a way, a bit of a constant, friendly encouragement, constant pressure to say, hey, here's something else you can do. And then, yeah, here's something you can try. Have you looked at your EBS volumes? Just like you'd have a personal trainer who is saying, let's push it on this exercise. Let's try and do an extra set this week. That kind of thing, to be that constant pressure.

It's interesting just seeing how difficult it is to define what we do in the context of what we do. I like to sometimes say that we're marriage counseling between engineering and finance, which feels like it is not that far from where a lot of these things tend to play. You don't have a trademark on that phrase, do you? No, no, I don't. I'm sure that will come up in conversation. You have to sit down on the couch here and what are your needs and are you being heard here? That's definitely the case. Exactly. We can talk about insecurities as a safe space Trusting. It'll be great. It'll be great. So lately, it seems like there have been some shakeups over at AWS. There are a lot of things going on over at AWS as of late. I know that they've been chasing a lot of the ML dragons lately. They've had some significant and notable departures that I'm not seeing backfilled. That doesn't mean it's not happening, but it's not happening in the public eye.

My Rolodex, people I know at AWS seems to get slimmer every year. A Rolodex is a list of contact cards for those of the audience who are, you know, younger than we are. It's funny that those analogies are starting to break a little bit. The save button is still a floppy disk.

Yeah, I mean, so just at point in time, we've learned about Matt Wood's departure and there's no official word on where he's going. As of this recording. This is one of those things where just invariably, as soon as we say this and the track is laid down, 10 minutes later, there'll be an announcement and it'll be all over the place and we'll sound like fools. Check the show notes for an asterisk somewhere.

Exactly. Yeah. But I guess with anyone involved in the generative AI space with that larger of a persona, you probably have a lot of opportunities coming your way. So that's my speculation. I don't have any inside information as to where he's going. It's always interesting to me watching how senior executives manage transitions between companies, the way that they talk about it, how they structure it. Like when you talk about like the Uber executive types, the executive VPs at giant companies,

in many cases, they'll do a departure. They'll have a long handoff period. There will be a farewell tour with a lot of the clients. They're generally on gardening leave for three to six months. And then it comes out where they're going to go is generally a step up. The inverse of this is stepping down immediately to spend time with family. It's like, ooh.

We can read between the lines there. Yeah, that sends a message that isn't terrific. And the fun thing is when companies don't realize that that's the message it sends, and it's a legitimate, oh, that's actually what happened, but no one believes it because of the messaging. They need to get new PR for them. Honestly, lately, it feels like so much of the AWS corporate comms apparatus has been aimed at talking about machine learning. And when you start digging really deep into it,

the teams that build and run the services that we know and depend on are still there. They're still doing things that are neat and that are going to be trotted out and reinvent, which is freaking terrific. The problem is, is that that's not at all what gets talked about. I mean, yeah, certainly I'm a machine learning AI optimist, but it has to, it can't take the place, right? I think the pie should get bigger, right? It shouldn't redraw the pie. I really hope reInvent's keynote is not a rehashing what we saw at Google Cloud Next earlier this year, wherein the entire keynote was about

AI and it never pivoted away from that. The problem I'm seeing when these companies flood the zone like this is that when there actually is a fascinating and useful breakthrough that hits from wherever it happens to come from, we're never going to see it because it'll just sound like, no, no, this one is actually transformative and revelatory. You've said that for the last five disappointing flops. So what makes this one different? transformative and revelatory. You've said that for the last five disappointing flops. So what makes this one different? And I think that there's a line, right? And I think that Steve Jobs used to represent this

in this reality distortion field, right? Where you want to be far enough ahead that it's optimistic, but not so far ahead that you have lost sight. Hey, we still have to have EC2 instances and databases and managed services and that sort of thing. And I talk to customers. I fix AWS bills. I see the spend. There is an awful lot of money in cloud and a surprisingly small percentage of that is even tangentially related to AI. Companies have not suddenly given up doing the block and tackle the things that generate the revenue and keep

people employed. You're a 50,000 person company. You're not suddenly going to pivot everyone to AI. You're just not. I mean, I guess I could speak for myself personally and the company where I'm at. I think right now people are doing the similar things, but they're doing it differently. Like right now, whenever I'm trying to do something, I'd spend some time with, how can I do it with AI or using AI or trying to make the scale? But it's also the classic programmer trap, right? Of how do I spend 40 hours to automate this 10 second job? That's part of it. I found that it's useful for certain things and not useful for others. One area where it excels, obviously, is coding assistance. That is effectively a superpower.

Every developer I work with these days is using some form of Copilot or something similar. That is magic. Doing it, something similar for writing blog posts or whatnot, I find it helpful, but only because when it spits something out, it is so bad that I feel the need to dive in and fix almost everything. So there's like 10% of what it originally had that was there by the time I'm done. Usefully for me, that's structure.

So great. I've just filled in the content with things that aren't inane. Yeah, I really like using it to modify and to play with text and help me rephrase this another way and help me turn these notes into a more useful general content. I think starting from zero and saying, write me a blog post about cost optimization, you're going to get out the effort that you put into it. Oh yeah. But even that is helpful for those of us who struggle when staring at an empty screen. Yeah, absolutely. Well, I guess, so here's a question, especially with the coding

assistants for young people who are just starting with programming. Like, how do you, I still wonder about that because I have kids who are of the age where they can start learning about programming, but like, is it now worth it to talk about here's how you write a program or should we just focus just on here's how you use an AI assistant? I mean, cause I don't know much about assembly language. I know the principles, but nothing beyond that. And is programming, is Python going to be assembly for kids? I'm taking my guidance from educators, by and large. Math is a good example of this. Very often we'll see that, okay, we have calculators everywhere now. They're in our pocket. They're ambient in the room around

us. So being able to say, okay, you're not going to have a calculator with you is ludicrous. But I have a second grader who's still learning math without aid of a calculator because you need to understand what is reasonable for a calculator to return. And I feel like there's a similar progression on this, but it comes down to what is your goal? Are you trying to be a computer scientist? Are you trying to get a job as a front-end developer as your first gig out of school or bootcamp? Are you just looking to something that keep the lights on 40 hours a week? What is your goal here? I've always found that a deeper understanding of fundamentals is a terrific thing to fall back on when automated things break, which they tend to do. We've all seen ridiculous suggestions from AI, some of them funnier than others. It's not nearly as funny when it makes

it into production, but there's a strong sense of you need to have a grown-up looking at these things. But senior engineers can see this, sure, but where do senior engineers come from? We didn't just emerge from nowhere. I love how you mentioned math. Have you read the essay Lockhart's Lament? Not yet. Okay, so it's about this math teacher. He is terrified that he's talking about, imagine if music was taught the way math is, and you have to learn all of these scales and all of these theory before you could ever touch an instrument. Isn't that how classical piano training works? The idea is you can just sit down at the piano and just start playing, try and make some noise, and you can learn a lot just by playing.

And I feel like with AI, I can play with programming more freely. But you're right, you still have to go back and learn the basics. You still have to go and learn algorithmic thinking and data structures to some extent, but being able to play more easily. And I think a lot of poor math education is basically like learn all of this stuff, but you don't actually just sit there and play with a number, right? Draw a right triangle and try and understand the relationships between the length of the sides or try and figure out why numbers, when you square them, are always positive or something like that, where you can just sit there and play and you don't have to, it doesn't have to be formal. That's how I learned a lot of math. So much of it was just listening to teachers droning on about

stuff that I already got the first time they explained it. And I'm just sitting there playing with numbers in a notebook because this was before the time you could just waste time on your phone. We could be bored back then. And the exploration that you just do for the fun and the joy of it is where you learn a lot. Yeah. It's being second or third grade and we're doing those, what are those trigonometry proofs where you have like angle, side, angle, and it shows that, okay, if you have an angle and a side and an angle, then these two triangles are, what are they called, congruent. I use some argument that we hadn't officially learned yet. It's like, well, you can't use that because we haven't been taught yet. But it's true, because

if you have this and this, then that has to be true. Yes, we haven't learned it yet. I think some of this formalism can kind of beat the joy out of things. Oh, yeah. Things I love, when I find them being taught formally, become things I hate. I'm going to get grief for this, I'm sure, but so often in English

classes growing up, they pick some of the worst books, and you can tell those books are terrible because they won the Caldecott medal. And it's they have the big seal on the front of it. And it is so heavy with meaning and symbolism that they forgot to tell an actual story. And it's just like, I love reading, but this is crap. If I didn't already love reading, this certainly wouldn't help. I'm remembering an online English class that I took. We were all supposed to write a comment on the book. And I said, the thing I liked about this book, it led to one of the best naps I've had this year. It was so boring and incomprehensible, right? There was no fun in it.

There was no story. It was just lots of metaphor and simile and stuff to analyze. But there's no, there's nothing, there's no substance. Which brings us back to AI marketing, which is what worries me, because in many cases, there is substance behind these things, but it's attenuated, and it's difficult to get to a point where you can see it being useful. And there's also a lot of defensiveness that factors into this. People are, in many cases, worried about, how is this thing going to take my job away from me? As a counterpoint, I don't see too many jobs except at the margins where that is a realistic possibility in the near term. So there is time.

If the crypto world is hard to sort out, AI is going to be so much worse, right? And I think that the fundamentals of cryptocurrency, it makes sense and I like it. But then on top of that little core of truth is this giant pile of scams and garbage and hype. And so I think with AI, it's going to be even harder to sort out what's hype, what is someone just with a cool website and going for venture capital and taking us off a ride versus, hey, this is a new innovation, this is a new model. I think that's going to be harder to sort out. And that's where, well, hopefully people, I'm trying to keep myself in the know as much as

possible and try to be able to sort that back to the reference point for that yeah it's i think there's going to become a future where this is just how computers work like it just it's unthinkable to sell a computer these days without an ssd attached it's similarly it's going to go oh of course it's going to have an ai compatible gpu equivalent whether it can actually display graphics or not is almost beside the point. Yeah, maybe we've kind of topped out on what we need in terms of graphics, but AI, yeah, AI focused processors

probably have lots of room to grow. For text-based LLMs, for the conversational stuff, I'm getting great results just on my existing Mac hardware, which is not terrific when it comes to GPU, because that's never been its goal, but the performance is just fine. I'm getting, in some cases,

what, 200 tokens a second? Yeah, are you using, like, Ollama, or what are you using? I'm using Ollama with a bunch of, obviously, submodels under that, where people have, where I've done a lot of, I've done a lot of playing with some of the custom models, people have tweaked and done some enhancements on top of Ollama itself,

of course, people playing with Gemini as well, or Gemma, whichever one, the one that you can run locally is like, Google's bad at naming things. Who knew? Yeah. My most capable local machine is a Mac M2 Max. And yeah, I use a Llama as well. And it is impressive what you can do locally. Although I do wish I had maxed out the internal storage. I didn't realize we're going to be downloading these 70 gigabyte blobs. Yeah. When I wound up building out my Mac Studio M1 Ultra back when the pandemic was just getting started, it was the first generation of Mac Studio. I went for the middle of the road processor and I wish I'd gotten higher just for the GPU story on that. It's hindsight. Is it worth paying another $7,000 to correct the mistake? Not so much. I was in an Apple store just the other

day and it's like, I can't think of a reason to upgrade my stuff at the moment. I mean, this is an M1 Max and then I have an M2 Max Studio. Yeah, I'm on a laptop myself, M1 Max. I got it as a quick laptop in a hurry for re-invent in 2021, and I haven't felt the need to upgrade it since. Of course, I also used to travel exclusively with an iPad Pro, so I'm probably not the best example of this. If I need some actual heavy compute, well, that's what the cloud is for. So I'm curious as far as what is your history with AWS? You've been around the ecosystem an awful lot. It's fun talking to folks who've been in this back. We can remember when there were enough services that most people could hold them all in

their head at the same time. Okay, this is an interesting one. So I was doing a research project with some professors at the UW. I was an undergraduate at the time and they were finance professors. And we were looking at this lending marketplace called Prosper. And Prosper, it was a peer-to-peer lending platform where someone who was applying for a loan, so they wanted, I don't know, $5,000 to deal with a plumbing emergency, and they'd write who they were, and they'd have their standard loan information, but they'd also provide a writing sample and a photograph. So you get all the standard loan information, right? What's their credit score? What's their debt to income ratio, that stuff. But you'd also have their writing in

a photograph or several photographs. And at the time we had a relatively new service. This was 2007, I think we had Mechanical Turk. And so I wrote a bunch of Perl scripts for these professors to put all these images onto Mechanical Turk. And for those of you who don't know Mechanical Turk, you can pay one cent or two cents or three cents and you can get people to do these little jobs called human interface tasks. A lot of those jobs are now replaced with image recognition, that sort of thing. But at the time, you'd use it for looking for something in an image. It was a good use of that. And so we'd ask, is there a dog in the photo? Is there a puppy?

Is there children? And try to figure out, did these photos that people were putting in to their loan applications, did they add extra value in terms of people estimating probability of repayment? Maybe the people who put puppies in their loan photos, are they better loan candidates and why? These were behavioral finance professors and so i spent a lot of my undergraduate my last year writing a mountain of pearl scripts i learned about mechanical turk i actually i asked my dad so my dad jeff barr chief evangelist aws i asked

him we set up a little meeting with these professors and it was all it was just such a little ecosystem at the time it was so so fun just how small it was. Yeah, we wound up building out a lot of stuff in Mechanical Turk in the early days at Expensify for receipt optical character recognition on that and understanding what a receipt is because there's a lot of variety in receipts. These days, systems are pretty decent at doing that. But, you know, 2012 when I was there,

that wasn't the case. Mechanical Turk was great for it. You'd spend fractions of a penny per read on this, or a couple pennies, and it would work out. Yeah, absolutely. So that was my first exposure to AWS. Later, I was in grad school, and I was helping another finance professor. She had this really expensive GPU rig, and she had written this simulation that had run for like three weeks on her desktop, running the simulation over and over again for different parameters. And she had submitted this paper to get what is called reviewed by her peers. And she got really nervous that she wanted to make sure that the simulation was reproducible, but she only had three or four days to do that.

And her simulation took three weeks to run on her local machine. And so at that time, there were six regions. And so I maxed out what my account could do, which was 100 instances per region. And that was the first time I had written a program that could spend a substantial amount of money. And it was really simple architecture. It was pull from an SQS queue

and committing private keys to GitHub. Yeah, I accidentally helped someone start a Bitcoin farm once. I don't think that there are too many people from those days who didn't. I'm so glad that GitHub will detect that now. That has saved many a person from financial ruin. This episode is brought to us by our friends at Wiz.

Wiz is on a mission to help every organization many a person from financial ruin. In other words, nobody beats the WIZ. WIZ connects in minutes using an agentless approach that scans both platform configurations and inside of every workload. They provide a deep assessment that goes well beyond what standalone CSPM, CWPP, and other unknowable acronym-based tools offer and find the toxic combination of flaws that represent real risk.

To learn more, visit wiz.io. That's W-I-Z dot I-O. What I find interesting is that the large scale of company that I tend to spend most of my time talking to, if there's a breach and someone gets keys and starts mining Bitcoin, it's a lot harder to detect because, yeah, my personal bill is what, 50, 70 bucks a month, something like that, depending on how aggressive I've been with serverless that month. And yeah, OK, now it's $5,000. I'm going to notice that or 50,000 or 500. Whereas when you're spending tens of millions of dollars a month, that starts to really disappear into the background noise. You got to be paying a lot of attention to even notice it happens. Oh, gosh. I was working for a company that shall remain nameless, but they gave every consultant and

every person who needed it the same root credentials to their AWS account. Everybody. There's no, no, I... In my early days, someone emailed me theirs, which scared the crap out of me. And so they had this app that could have run on a Raspberry Pi, and the spend was like $30,000 a month and growing. And they didn't know what to shut off and what not to. And it was a well-funded organization, so they could kind of live with this. And I think there was some accounting maneuver to subdivide this.

But yeah, it was really terrifying what could happen through these bad practices. Oh, yeah. I care a lot less, to be very transparent, about the big companies taking the unexpected financial blows. They can absorb those. It's the independent learners who are just figuring out how this stuff works. Maybe they trust the wrong person on Discord or whatnot, and suddenly they're staring at what they believe to be an insurmountable amount of money. $700,000, that's more than I'll ever make in the course of my life, which is not generally true, but it feels like it's the world is ending then. And AWS has never been great at making people feel reassured while they're looking into that. I haven't heard too many stories where they haven't made it right, at least the first time, but you're twisting in the wind for a week or two first,

not knowing. Yeah. There's that moment where you're like, oh my gosh, what am I going to do? And does it even cross your mind? Okay. Do I, if I get on chat, are they going to forgive me? And if I call up support and yeah, like you said, they all, they usually let you off the hook unless you were phenomenally egregious and have done this multiple times. But if they determined that it was a genuine mistake, they'll, they'll look after you. Yeah. It can be frightening to get a bill that's multiple times your monthly income. Oh, yeah, it's horrifying. And I'm just glad that it seems like I haven't had seen nearly as many of those stories lately. I think a change that is absolutely for the better

that is in the process of rolling out is mandating MFA when you spin up an AWS account. It's so many other things mandate this, and it's just such a no brainers. Well, it'll act as a break on growth. If someone's not willing to give a phone number for a text message or use Authy or something like that, then I don't know that you want them in your cloud environment in the first place, do you? No, I agree with that. Use MFA. That is a good idea. And even the inconveniences that it can sometimes cause, like if you're not good with your backups and you lose your phone, well, yeah, that could hurt. You're going to have to go through some manual steps to deal with that. But the blast radius protection that it gives you is well worth that. And that's what we always say, limit your blast radius, right? If you have

an AWS account, you can turn off people being able to start GPU instances or the SAP HANA instances, the X ones. You don't want to accidentally turn one of those on. I haven't, I confess, I haven't put a lot of those in place for our internal corporate org, just because everyone who has access to touch that stuff has multiple years of experience within the AWS universe. So there's, at some point, we're absolutely going to have to get some religion around stuff like that. And we do have segregation of who can look at data, but we, at this point, it's assumed that, oh, if you're spinning up a GPU instance, you probably know what you're doing. Maybe it's time to revisit that assumption. Yeah, I think we have it in our larger org,

just to make sure, okay, if, and it's not to stop people, but it's saying, okay, there's a hurdle because this hurts. So just make a hurdle in place so that if you're doing this, you know that you're doing it. Don't, it's principle of least surprise on some level, I think is really where it winds up hitting. Speaking of surprising people. So when you talk to folks who have been at big companies and are being confronted with areas for improvement when it comes to their

cloud spend. You and I were talking about this earlier. It's why I call myself basically marriage counseling between engineering and finance. There's a lot more people and behavioral aspects to this than there are raw math. The reason that cloud costs are complicated is only partially due to the fact that arithmetic is hard. There's a lot more to it than that. What's your take on it? I completely agree that many of the issues are people issues, right? People who might be scared of change or want to protect territory and don't want any... There's a lot of those types of things. And also, people don't want to look bad. Their careers are on the line. So if you say,

look, we can save you this amount of money. It's easy for someone to react and say, wait, why didn't you do that? You were my AWS person. Why do we have to hire these external people who are then pointing out what you should have been doing in the first place? But that's what you don't want is this finger pointing to start happening. So ideally when people see the savings they can do, they have to say, okay, I know you were doing the best you can. We brought in some experts to help you. And look, there's so much more on the table. We're really excited. Now we can do this project and that project. And you're right, it's a marriage something successful while trying to do it for the absolute least amount of money physically possible saying no to something else. Where do you draw the line? What is the priority? Because when everything's the priority, then nothing is. Where do you want people's energies focused? Because there's a certain mindset, and I admit I'm one of them, where I will spend, belieffully, a month just knocking a few hundred bucks off the AWS bill. I cost more than that.

So what is the actual value that I'm contributing? At some point, one of my roles is to tell people, okay, stop saving money here and go back to doing the thing your business does. That's a really, really good point, right? There's only, there's diminishing returns to this process. That's why having to automate what you can, I think we're big believers in that. So for example, GP2 to GP3, right? That should be automatable.

I view them in buckets on some level. Like the first pass is generally, okay, great. Click three buttons and you'll save 10%. The next 10% is going to require some re-engineering work. And the 10% after that is borderline taking hostages in Seattle.

So there's a, it gets harder and harder once the low-hanging fruit gets found and sorted out. But there's also a question once the low-hanging fruit gets found and sorted out. But there's also a question, increasing the enterprise, of how you handle governance. How do you get 1,300 teams all rowing in the same direction? And that's where it has to be, is that relational thing, getting the top-down buy-in, but then also having the credibility with the engineering teams. They look, we're doing things. This is, we know our stuff.

We're trying to do things that are generally applicable. And that's the other thing we find. I'm sure people in my position, your position, you get to see a lot of different AWS accounts. And so you can think very generally about them. Whereas often when you talk to someone at their company and they've been there 10 years and they've nurtured it from the very first EC2 instance up to where it is today, it's very easy to convince yourself that no one else uses AWS like we do. It's one of those ideas where you're right. At enterprise scale, no one is going

to use it exactly like you do, but there are patterns, history rhymes. There are, okay, great. Maybe your naming structure is different, but you're still going to run into the same foundational problem that other companies are. I i mean it's like the marriage counseling example again right you're doing this every problem is unique because you can't come in as a consultant under the tagline you're not special it does not go well because who who wants to be like everyone else yeah exactly but at the end of the day they said there's patterns there's things that you can do that are generally applicable and that's what one of um like talking to private equity firms who have a portfolio of

companies and they want a policy across their companies that often resonates a bit better we found because they're in the mindset of thinking at a process level about companies and a process level about operation whether than like some idiosyncratic, here's the one AWS account for this one, I don't know, company. There's a lot there, but I do want to get to one more topic before we call it an episode

because you are a chief evangelist. Huzzah. Talk to me about content creation. What's your process? Quick self-promotion. We have a show, AWS Made Easy. Myself, my co-host, Raul Subramaniam, try and stream every week.

We interview AWS guests. We try and cover AWS news. We rate and review articles, that sort of thing. I got some really great advice early on, which was get to 20 episodes. And I heard somewhere that on the Spotify distribution, most podcasts are like three or four episodes and then they just vanish, right?

It's a couple of people. They meet up for a few drinks, do it in their garage, launch a podcast. They lose momentum. So my first thing was get to 20 episodes. And then we did that. So number one is prioritize shipping it over quality. And I'm sure you've seen that with video editing. There is no end to how much you can kind of polish an episode.

I've never touched it. I've always just handed it off to professionals. It's the most common piece of advice I give to people who ask me when they're starting out. I'm like, okay, what microphone, camera, et cetera, should I buy? Doesn't matter. Doesn't matter. You can record something just fine on your smartphone to get you started. If the content's good, people will forgive an awful lot in production value.

And if your content isn't good and isn't compelling, it doesn't matter, because no one's going to stick around to see it. It's trash, but it's so well produced trash doesn't matter. No one wants to watch trash. There's an attention economy. I agree. But also at some level, you do need to pay attention to that stuff. And maybe it's just, you know, when your head gets into it, you hear a great piece of content, but you're like, it sounds like it was recorded in a bathroom. maybe that's part of uh having been in that content creation space for a while right maybe i would have never noticed it before and now i notice it's almost how like people audiophiles say that it kind of ruins music for them because they say oh it doesn't sound right it was recorded on this instead

of that so in general try and prioritize shipping try and do the best we can quality wise in a bounded box of time and money and effort. And yeah, most importantly, like you said, make it usable and try and distill down our experience of dealing with a lot of AWS accounts to make it useful for people, people to be able to submit questions. And we try to make shorts and clips.

That's been really nice to be able to give people a small taste of it because our episodes are an hour-ish and being able to have a 60-second clip that shows, it just plants a seed and then a link to more. That's been really nice. It's nice to see people grow, but I think that people also try to early optimize that. And the problem is there's no end to it, right? There's literally no end. And then as a tech person, right? It's fun to nerd out on the gear and the stuff and the switchers and the distribution. And then you have the AI layer.

We say, can I automate everything? And then it takes 30 minutes to get your audio working to join a Zoom call. I've been there. Yeah, exactly. I mean, if I open up my audio devices, it's like nine different things, right? Because you keep changing your setup. Okay, sorry. Is it this one?

Is it that one? Is that one? Oh, wait, I'm on my UMC 404 HD today and not my Blackmagic. But I think it's with any approach, right? Try to do the best you can within a bounded level. And ideally, it's about the listener and adding value for them. And we always try to ask our listeners for feedback, answer questions when they ask, Not making it about ourselves, but about remembering who the audience is and trying to add value for them. I find you have to write with someone in mind. Otherwise, you're just speaking in generalities and that doesn't help. Yeah.

Think about who is our persona. Our persona is someone who they have AWS. Managing their AWS account is one of their many responsibilities. And if they listen to us, they can pick up something useful that can help them without having to hit refresh on the AWS What's New page like you and I do. Exactly. I have one last question for you, because I'm going to invite you to embarrass yourself as, you know, it is always the way it works when you make predictions. Where do you see the future of technology going? You've been around in this space a long time. I have as well. Where do you think the future is? Is it darkness? Is it awesome

stuff? Is it something in between? I'd like to be an optimist, and I hope that we're heading more towards the Star Trek universe than a dystopian future. I think about some of the building blocks that we're working with right now with LLMs all the generative things there's a great scene in star trek first contact where the crew of the enterprise is transporting to earth and they've traveled back to the late 21st century and they say okay beam us down switch us to late 21st century civilian clothing and thinking about okay what had to happen for that to work right there has to be some ai about well what what clothing were the civilians of this time wearing and what were they doing and there's a lot of stuff that had to happen in order to make that work and i think we're playing with those building blocks right now right that are

our descendants are going to be playing with you know bedrock 3.0 which might power a transporter um so maybe maybe that's the case or i'm hopeful that that's that's the case the flip side of that is the amount of monitoring that we can do on ourselves and trying to squeeze the humanity out of some of the processes in the name of i don't know efficiency so i worry about that i'm hoping that uh in general i'm an optimist so i want that i'm hoping that we tend towards the Star Trek outcome. I do too. I really hope you're right. I really want to thank you for taking the time to speak with me today. If people want to learn more, where's the best place for

them to find you? Find me on LinkedIn, Stephen J. Barr. Type that in, look for a guy with my face, and that's probably me. So thank you. Stephen Barr, Chief Evangelist at CloudFix. I'm cloud economist Corey Quinn, and this is Screaming in the Cloud. If you've enjoyed this podcast, please leave a five-star review on your podcast platform of choice. Whereas if you hated this podcast, please leave a five-star review on your podcast platform of choice, along with an angry, insulting comment that your descendants using Bedrock 3.0 will one day discover.