S2E1 - Scaling Your Data Infrastructure with AWS's Jon Hammant
Welcome to episode one of series two. I'm excited to announce that we're joined by Jon Hammant, leader at AWS and hugely knowledgeable on compute, technology, cloud, and, of course, one of the main topics we're talking about today is AI and data. So let's dive in. Well, welcome, Jon. Thank you for coming on the show.
Aaron Phethean:I guess the first thing to do is really have a few intros and tell us a little bit about you. You know, you work for AWS, and you have a a customer facing role, but but tell us, what's what's in your day to day?
Jon Hammant:Yeah. Definitely. So, so Jon Hammant. So I've been with AWS about four and a half years or so. Initially, was running the compute business for The UK, then running the compute business for Europe, and now I lead the specialist business to The UK.
Jon Hammant:So that's pretty broad in terms of scope. It's both business and commercial, which is relatively unique for for a lot of these roles. And, basically, I look after all of our core computing services, so that's everything from EC two to networking to high performance computing to security, our enterprise applications and modernization. So that's if you're thinking of moving more, I wouldn't say legacy, but more, heritage workloads. It would be nice to have been a heritage workload.
Aaron Phethean:Very, politically correct.
Jon Hammant:Yeah. It's important. So the heritage workloads that you wanna move over, whether that's either Windows or VMware or SAP or, anything that contains in service, so I leave that. And then also, there's the really fun bit that I look after. So all of our data, AI, storage, and analytics businesses.
Jon Hammant:So we're pretty Well,
Aaron Phethean:that's that's certainly part of the appeal of of having you on. I tend to so I've been doing these, you know, a whole season now. We're into season two. And every episode, I've brought up AI, and I've had a bit of chat. You know, it's very, you know, relevant to data, of course, but it doesn't usually come up in the first five minutes or even in the intro introduction.
Aaron Phethean:So if he takes us back to the kinda compute and the, you know, the the kind of work that that Amazon does and and, you you know, the kind of infrastructure you help customers with. It feels like AI is in a very similar trend to early computing, cloud computing, where adoption is kind of ahead of controls. Yeah. I sort of wonder if you're seeing organizations start to panic as they adopted more and have not quite figured out, you know, the the constraints. I know that they're struggling on the delivery side to have enough compute for AI, but I wonder how it surfaces for the customers and and what you see.
Jon Hammant:Super. It's a super interesting topic. So I think there's a numb so I so I do think history kind of has a habit of repeating itself, and I think some of the things that we're seeing AI go through at the moment kind of remind me of the early days of the iPhone, if you remember kind of back when that released into into a business context. So, you know, people talk about their jobs being removed by AI. I like to think my job has previously been removed by progress.
Jon Hammant:So I used to be a, a Blackberry administrator, a Bez administrator. Oh, wow. And there's lovely little job that is many, many moons ago. And, you know, at the time, I thought black I I still think Blackberries are really cool. Like, they were just they were just fantastic email advisers.
Jon Hammant:And it was interesting because the iPhone kinda turned up. And all of a sudden, I think the reaction I remember at the time because I was working in, Lambrooks currently. The reaction was, oh my god. Do not let people plug those in. Like, that is a terrible thing.
Jon Hammant:There's data leaks. There's all these issues that can happen with that. Yeah.
Aaron Phethean:And I
Jon Hammant:think what very and at first and that was a pushback for, you know, a year or so you saw happen. And then suddenly, you suddenly saw the the kind of the enterprise readiness of them coming in very, very quickly. The people move, they they used to do this, and the, the admin tools came in, they got a bit better. And all of a sudden then, yeah, you flick the you there's very few people who are using a BlackBerry for work now. They're all using iPhones.
Jon Hammant:That makes sense. Or they're all Androids. And I kinda think the same thing is happening with AI. So, you know, it's come in that we've hit the zeitgeist of both whether that's either Claude or ChatGPT or, you know, DeepSeek now. You have so many people that are using this in their daily lives, whether that's either from working out cooking recipes to, you know, helping plan a holiday to asking it to to help them learn a language, whatever they're trying to do in a normal.
Jon Hammant:And I think
Aaron Phethean:I heard on the radio the other day, there was a announcement, I think it was public sector, you know, some more controls. I think they might have even banned its usage because they're worried about the data leaks.
Jon Hammant:Yeah.
Aaron Phethean:And I kind of feel that that's probably one of the best barometers of its market penetration. You know, basically, even the public sector, even the other one of the industries that's sort of renowned for being behind the curve is using this tech, and people are trying to innovate
Jon Hammant:with it.
Aaron Phethean:And the the the weird
Jon Hammant:thing is, I think so I think people are everyone's using it. I think the reality is that a lot of companies are in this situation of going, well, how do we take advantage to it? Because the reality is if you I mean, I can tell you categorically, if places ban it and they say, thou shall not use generative AI to do things, 100% people are sitting there and they're using it on their phone or they're just using it. We we talked to a a CIO of a large organization recently, and, he was saying that they they see on their their web proxy effectively the amount of usage of of unauthorized tooling that's going on. It's it's it's bonkers.
Jon Hammant:So what do I think where does we kinda that's for a while. I do think a lot of companies have realized that, actually, them spinning up their own internal agents is a really, really useful thing. So we've we've got it internally. Something called Cedric that, basically, we use for pretty much everything. It's fantastic.
Jon Hammant:It's got the Amazon docs within it. It's got loads of context. It's all secure. It's all internal. So I think in terms of that maturity life scale at the moment, most companies that aren't running some kind of internal invocation of an LLM that are of any size realistically will either be looking at team packages on some of the the good external ones, or they're gonna be working on how they kind of roll and integrate their own.
Jon Hammant:I think it's I think the value is really when you put your business data into it as well.
Aaron Phethean:And I I definitely I mean, it's probably our own personal perspective and talking to clients and and thinking about how to get, you know, real value from it. What what I noticed is that everyone tends to be naming them. So they were they all kind of, like, creating them like their pets, although they're kind of like and I think what what's interesting about that to me is that once you start doing that and you start treating them personally and you start thinking, well, this is how Amazon does it. Well, of course, you want to tune it even further. Of course, you want to, like, make it, you know, even sound like someone at Amazon Talks or, you know, someone in your company that uses their lingo.
Aaron Phethean:You know, that that seems like a a real natural extension for at least, you know, obvious from when they start naming them.
Jon Hammant:No. No. %. I think you can see, so we we're working we work with a very large advertising company at the moment, and they often talk about brand voice. So I Mhmm.
Jon Hammant:You know, making sure that the thing speaks with the voice of your brand. And it's definitely something that we're seeing, especially things that are public facing. You know, if you go onto a, let's say, for example, you go onto a user a chatbot that comes with an alcohol company, compared to a chatbot that might come from a travel agent. Like, you want the thing to feel and interact differently. I do think in the longer term and where I'm really excited about this, honestly, is I think getting to a point where I can have an assist I'd really like an assistant.
Jon Hammant:I think an assistant would be super good, especially one that knows knows more than me and knows things more.
Aaron Phethean:It knows everything. Right? That's that's sort of the appeal is that they went every single scrap of information is available. I mean, exactly. Assistant wasn't at that level.
Jon Hammant:And I think it's also I think it's so for years, we existed in a, a geocentric universe where people viewed that effectively everything orbited around the Earth. I think there was this, you know, thousand years till Veronica turned up. And there was a sudden realization that we're actually when a heliocentric thing were orbiting around the sun. It's all and I kinda think the evolution of intelligence is gonna go through the same thing at the moment. Even you hear people talking about AI, they talk about, you know, it's artificial human level intelligence.
Jon Hammant:So much of it's compared to humans. I think the really interesting opportunity for me is the kind of the extension of what intelligent means going into different directions. I think we're we're kind of getting that. I think some of the larger, ML models at the moment, they they just have a different kind of intelligence when you play around with it. It's fantastic.
Aaron Phethean:I think that's an interesting measure of the progress. Obviously, there's tons of excitement about the models, you know, and then DeepSeek's announcement, you know, disrupting the market and, you know, a lot of really panic about where the value lies. Yeah. I suppose as a data company, you know, delivering technology to move information around, my kind of heliocentric view was always that the data is the intelligence. There's there's a fine tuning of how you deliver it, speed you deliver it, you know, fidelity, the the there's there's, you know, how you you get to it.
Aaron Phethean:But, really, it's pointless unless you have the knowledge, unless you have your ad answer.
Jon Hammant:I think that, and I think you're gonna see so I do think there's gonna be revolutions in the, the ISV SaaS side of things. I think when you think about what a lot of what a lot of software does effectively, it's you know, three tier architectures are still the architectures that mainly mainly exist. I mean, they're they're Microsoft's order. But, you know, effectively, you have data that's being stored somewhere. You do something with that data, and you present it back to a user.
Jon Hammant:I mean, that that is a kind of, you know, a a Yeah. Majority of things you do. I think it'll be really interesting when you go forward and you think all of a sudden, well, what is the interaction layer on all of this? Well, actually, do you even need the interaction layer? Does it just become, okay, you have raw data that's sitting there.
Jon Hammant:You have an agent that knows how to work with it, knows how to work with different pieces. I think it, you know, I think it's it's gonna be interesting some of these very large SaaS businesses that have made a made a living off effectively that that thin slicing that's the the almost the intermediation that's that's been occurring for quite some time.
Aaron Phethean:I I actually think it's, you know, quite a, you know, interesting insight across all areas of software. So I think what you've described in my terminology is, you know, a system of record where you look after things and then the way you make decisions, you know, you have analytics, so you have some kind of, you know, warehousing of that information downstream. The disruption, I agree, seems to be that there is no real need to input it in the future. It's just sensing. It's collecting.
Aaron Phethean:This is what's going on. And then you just have the output. Well, this is this is what was actually happening. You know, that that's kinda tended to be the weakness of many CRM systems is that if nobody input it, it was it was a kind of pile of useless information. But you don't need someone to be, you know, painstakingly managing these things in the future.
Jon Hammant:I think that, and I think also, it's I really like x k c d, if you've watched that, the with the the comics with the the stick people.
Aaron Phethean:Well, well, Sun's an actual hero. He's a loves loves him.
Jon Hammant:It's just brilliant. It's just brilliant. But and there was one there was one of them where they said about, making a new, charging standard. They kind of had, right, here's 13 different here's 12 different charges. We're gonna make one new charging standard that's gonna rule them all.
Jon Hammant:And then what you end up with, you know, 13 charging standards. Of course, you did. And I think what we've kind of where it's interesting to see how AI is evolving. I think a lot of this that you're gonna build like, how do we use AI in a business? What you're gonna do, a new tool that you're gonna go to.
Jon Hammant:So you all you do is you're gonna, how are we gonna stream those other tools? We're gonna build a new one over here. You're gonna go to it and use that one instead. I think very quickly that paradigm is going to shift. And, actually, what you're gonna find probably within you know, we I'm seeing it more and more in places that it's just gonna become part of the workflow of what you do.
Jon Hammant:Is that is the reality of it. And I think
Aaron Phethean:Certainly, in every product, there's you know, we can see every vendor putting it in the product, and some better than others for sure. I I wonder if you see and talk to data teams. And so there's there's kind of two angles I'd like to think about it from. Yeah. There's what the organization does.
Aaron Phethean:And, you know, the data team, I think, is really crucial part of supplying that that capability to the organization. And then there's the data team themselves, like, how they benefit.
Jon Hammant:And I
Aaron Phethean:and I wonder if you're speaking to data teams and and and working with data teams and and
Jon Hammant:Yeah. We we really are. And I think there's so so on the different lenses around that, I think, if I look at where do I see some of the near term implications of AI, I do see it in productivity on technical development or technical work. I think there there really are some amazing, amazing opportunities. You know, it's fantastic.
Jon Hammant:I I code with I code alongside with Claude now. It's just brilliant. It's Mhmm. It's like I've got a pair programmer who's far more capable at me than me and knows things better, and I can just I can't necessarily always trust them, But most of the time, they're pretty right. And as long as you kind of work out where they're gonna go, you're you're kind of so I think this whole idea of more pair work with an AI, this I think is gonna come, and I think that will massively benefit some of the data teams.
Jon Hammant:I think also it's interesting thinking what the the schema format of the future is going to be.
Aaron Phethean:I think it's a
Jon Hammant:super interesting thing. So we, so we were working with a very large meteorological, association. We work quite closely with them. We're we're talking through through kind of different options, things we can do. And one of the bits of advice that we're giving to them is use more free text.
Jon Hammant:So when they're storing files, when they're storing blobs of data, include just a human written free text explanation of what that thing is. And it's kind of weird thinking that free human free text would have been the thing that I would have, you know, made my toes curl a bit different.
Aaron Phethean:That's the most useless way of storing it, whereas it turns out to be the best.
Jon Hammant:It is. It is. It's it's really and I think that more and more so we've we've been looking at agents and how, effectively, how agents will communicate via APIs. I do think the API of the future will be language, natural language. Yeah.
Jon Hammant:And and I think that's a really interesting change. It just hasn't hasn't ever I
Aaron Phethean:think, you
Jon Hammant:know, and also that yeah. And next language has millions of years of evolution. We've designed to get good at saying things Yeah. And coming Must
Aaron Phethean:be able to convey information accurately and at scale, you know, it it it works. Yeah. It's been proven to work. One of one of our, clients was dealing with information supplied to them from their customers. So they're actually in a private equity scene.
Aaron Phethean:And, obviously, you get delivered decks. You get given information. Sometimes you get the ability to tweak the forecasts and there's macros and there's spreadsheets. And and we were sort of, you know, wringing our, you know, hands and thinking, how do we deal with these, you know, macros and spreadsheets? Actually, I think fast forward a couple of years, and the supply will figure out that they better give it an AI ready way.
Aaron Phethean:And then there's sort of no point in coming up with a, you know, technology to, you know, work out how to tweak your macro and and, you know, build a formula out. Just people will supply it differently. They'll they'll understand how to supply it differently.
Jon Hammant:1100%. And I think, I do think the way of working is going to change in the near future. I really do. I think and I think you're right on your your earlier point about, I think, the deep seek style moments. I think it'd be really impactful.
Jon Hammant:I think finance has really started to wake up in terms of what's doing. I think one of the things that was, for me, the most interesting out of it happening was the move to using more reinforcement learning. I think that from a from a technical perspective, it gets super interesting that, you know, for ages, there's so any large scale technology change is just a number of overlaid s curves. So you just stack s to that. That's where, you know, where that's everything from, like, lithography to, you know, algorithmic development.
Jon Hammant:It's just lots of s curves that are rolling over the top. I think what you've seen with AIML in the the near term has been the the s curves that have been Moore's law, so you're just taking advantage of that. And that's going to pre training. That's going to some post training with reinforcement learning with human feedback. That's going to know, increasingly test time compute some piece around that.
Jon Hammant:I think now reinforcement learning and self effectively self play within these models
Aaron Phethean:is
Jon Hammant:gonna be really, really impactful, especially for areas that that have defined answers. And that that's the really I think the trillion dollar question for me is, you know, are the LLMs getting us or are the AI systems, sorry, getting as good as they'll ever be in things like, psych on a psychology or in philosophy, for example. Things that are less less constructed and less constrained in terms of their outcome. But for things where there is an outcome, so for sciences, for maths, for, you know, the, that kind of area, Yeah. You can show and for finance, being a perfect example, you can show an outcome that you desire from it.
Jon Hammant:So is there a limit to how much you can effectively iterate internally to get to that point? And the answer is no one really knows where it's gonna go, but this and you kind of argue you you argue whether it's effectively, it's synthetic data and at what level does synthetic data can be useful to continue to to to move this course.
Aaron Phethean:I think one one of the things I see, organizations grappling with, so, obviously, the the technology's new, the infrastructure, the constraints, how do we use it, you know, sort of volume, which which leads to some cost pressures perhaps. And, you know, one one of our core value propositions is is kind of lowering that that cost of managing and then looking after your data. But what I see the other thing they're struggling with is is the pace. So, yeah, you're talking about some very specific opportunities and features around the technology. And I see people investing, you know, customers investing a lot of, their their time currently just in trying to keep up and understand it and find the opportunities before you even get to leveraging it or or, you know, getting an outcome.
Aaron Phethean:Yeah. And it must be quite worrying or at least frustrating for a a kind of senior leadership team to think, okay. Disruption might be coming from this. How do we spend our time and our our, you know, our money? And, you know, that the outcomes aren't obvious.
Aaron Phethean:You know, it's sort of to some extent, it's it was easy in the kind of mobile era. We need a mobile or we need a, you know, either surface or the customer. This is not so obvious yet.
Jon Hammant:I think that, and I think there's a strange dichotomy that exists at the moment that I can, you know, I can almost, I can I was just, like, shooting stars? I I always feel like they they had a dove from above in this show. Like, there was view AI has been this, like, dove from above. You're like, cool. You cue this thing down.
Jon Hammant:And I think it's amazing that I can effectively summon up PhD level intelligence to go and do some really, really, really ultra clever things.
Aaron Phethean:Yeah.
Jon Hammant:Yet still, for me to put together my I've got I've got loads of expenses that I haven't done, and that is gonna be a three hour slog of comparing receipts. Yeah. And and I think it's it's really interesting. I think there's not there's there's so many bits where I think we are kind of getting towards that turning point that I think we start to see large scale impact. I I I think we're, you know, on the edge.
Jon Hammant:I think you I don't think we've had that kind of true moment of implementation yet.
Aaron Phethean:No. I think in certain use cases, people are like, okay. Well, that that's a perfect one to use it. You know, that that should be available. I think within teams that I deal with, it's still not that obvious.
Aaron Phethean:So, you know, give you a couple of examples. So I think the early notion was, let's write SQL better. Let's write write queries. Maybe the developer tool, the kind of assistant to development. And, actually, I think then there's sort of a a growing realization that, well, one of the the most demanding things on the team's time, which is fundamentally where the cost comes from Yeah.
Aaron Phethean:Majority of the cost and and sort of where the opportunities lie, both, you know, servicing their customers better than their internal customers, is that they answer a lot of questions. They dig into a lot of things. And there was always this kind of self-service, you know, analytics type dream or, you know, people self servicing. Actually, like, the tool used by someone or the tool used by, you know, the person who has the need, there's gonna be tons of time saved there and actually real organizational context asking things about the organization. I think less so in the truly nuanced kind of Yeah.
Aaron Phethean:Predictive side that that the opportunity seemed initially.
Jon Hammant:I think that I think also there has to be a an ask as to what are you what metric are you trying to optimize for. Mhmm. So, you know, I do think there's there's almost like the the the whole paperclip maximizer view of AI going wrong. So eventually, things decide to it's told to make some paperclips it does, and it takes over decides to take over all of humanity to continue to make paperclips. I think there's a a more micro worry that can happen where you start things.
Jon Hammant:So so AI is very good at certain it's very good at writing lots of documents.
Aaron Phethean:And so,
Jon Hammant:you know, we could sit down and we could make a huge huge corner copy of documents for people to go and read. But, realistically, that has no business outcome. That's not what any of our customers want. I think there's and it's the same. We sit in development a bit.
Jon Hammant:So, yeah, we've got a number of cut few customers that have, effectively, bought in AI driven development. That's definitely something that's coming. That's something I'm really passionate about. I think really makes different. A few of them, their initial impact and the initial assessment hasn't necessarily always been as good because one of the worries is if you give a lot of, like, more sometimes more in the genius side of development in the end.
Jon Hammant:I think it's just just you end up with a lot of maybe something more junior developers who write lots and lots and lots and lots of code that isn't very good and you kind of isn't really useful to an extent. Like, the best line of code is one that you don't write. And I think it's it's interesting when you you come up with a system that's very, very good at generating large quantities of things that humans have to deal with. I think on the flip side in terms of companies that are that are effectively going on to this, I think you'll see some companies waste a lot of time by making a lot of by getting AI to write a lot of things for them that a lot of people time spend people take to read without any benefit for its existence.
Aaron Phethean:I totally agree.
Jon Hammant:I mean,
Aaron Phethean:that about where our mind goes almost instantly from that is, you know, there's a sort of deluge of dashboards that's sort of constantly lamented by, you know, our customers and our kind of industry. But, obviously, you could create dashboards a lot faster. Doesn't make any more useful.
Jon Hammant:I mean render. Yeah. I It's I mean, it's interesting. I used to be I I used to work with the data warehousing team quite, like, quite early in my career. And even then, I think there were, you know, everything if you're gonna make a dashboard, if you're gonna look at analytics, what's the business value?
Jon Hammant:What's the outcome that drives? If you're just gonna show stuff that doesn't really and also it needs to you know, when you think about going through all these from a data perspective, stuff needs to be actionable. If you're just looking at something and going, that's kinda that's kinda nice to see, who cares? Like, it's just no there's no like, it's it's and so I think you're you're kind of exactly spot on. I think the ability to create vast amounts of dashboards is gonna be an interesting tool.
Jon Hammant:I think also there's there's definitely I've seen in the world of business sometimes people blame the lack of dashboards and view on things as being the you know, the reason we can't do this is because we don't have a dashboard that does it. And, actually Yeah. I I feel sometimes that maybe isn't the isn't the underlying why that that is sitting there.
Aaron Phethean:Yeah. Exactly. It never was. It's a, you know, it's a I guess, as an emerging space, there is a race to figure out how to use it, and that that's probably hard to get away from. And, you know, one of the things that I, you know, challenge a few guests with is is why the, you know, the curses, why the kind of developer tools might have ended up being the the first place.
Aaron Phethean:And to me, it's because it's explainable that you're developing something. And And we just sort of touched on the point of view. It's not necessarily a high quality thing, but it's it's, you know, how to develop something explainable. I wonder if, like, thinking about all the different customers and things you see customers doing, Like, do you think the trend will be more that way? So more creating something else that's understandable or more autonomous and more intelligent of itself?
Aaron Phethean:Yeah. I I
Jon Hammant:think it's really so so what in terms of the trends we see, I think this year, you will see far far more to that. It will just become a default in the same way that CICD became a kind of default. And, I mean, I it's always it's interesting looking back up when I was first working at Accenture doing DevOps, and you can't talk to people about DevOps and people like, oh, I don't know what that is. That's a I continuously integrate. Why would we we totally wanna, like, test it and and then now we're in a situation where most developers will will basically have a workflow.
Jon Hammant:They'll push it to get it. It'll go off, and it will do stuff, and it'll it'll roll out. I didn't. So so I think we're kind of in the same situation now, but with AI. So I think in probably a year from now, just custom there will be a default that if you're coding, you're using it for some kind of smart also complete using some kind of research around it.
Jon Hammant:I think it's very, very much seeing seeing people using it for testing. I think that's one of the areas that we would recommend going for first. If you're gonna write things with AI, actually writing the test harnesses is is probably more valuable because it's it's very good at generating tests. As to why it's happening, I mean, it's kind of if you imagine what's happening behind it, it's it's it's really interesting to think of how babies learn. So, like, it's really it's really cool when you see so I've got, like, three kids.
Jon Hammant:It's it was just like a little science project team, the things, like, work out what's going on. And you think, like, a baby will put its foot in its mouth, for example, when it's when it's really small. And I think what it's what's it doing, well, it's kind of it's using reinforcement learning. So it's calibrating effectively using the most sensitive area that it has in its body, so its mouth. And it's using that to calibrate the other sensations within its body.
Jon Hammant:And so it kind of it learns through you.
Aaron Phethean:It it
Jon Hammant:won't it keeps going through. The reason why I think there's probably gonna be an increasing acceleration in the use of AI within things like development and technical fields is because it's really, really easy to self play. Like, that's the that's the kind of interesting thing. And Yeah. That's true.
Jon Hammant:Fundamentally, if you think how does intelligence intelligence comes from two two ways. So intelligence either comes from, learning and learning from something else, so that kind of, like, knowledge, effectively knowledge transfer, or it comes from self play. And, actually, when we look in in terms of history, so, you know, if you get the the big things that happened, so if you go back to AlphaGo and AlphaZero, it was really interesting seeing the work that went into that. And the real moments actually started to come from alpha zero where it was all reinforcement. All set up like the the face covering.
Aaron Phethean:Completely new ways. Actually, though, I don't know. You you I'm fairly into chess. Right? Like like like playing online chess.
Aaron Phethean:You know? Some of these strategies, you know, a lot of excitement in those early sort of reinforcement learning type approaches is that, well, this is a strategy no one had yet discovered through all of the humanity and quite and chess is a game that's quite good at documenting outcomes and patterns and making recommendations, and yet it's discovering new things. And I think probably there's still an unsolved challenge of how you get the learning back in. You know, testing is one thing. You can't necessarily test in on an organization buying companies, for example, or, you know, investing in Yeah.
Jon Hammant:One of the one of the other things that I lead that's probably most exciting is our high performance computing side of things. I think that is that for me is, like, just super, super interesting in terms of everything from drug discovery to computational fluid dynamics to, like it's just it's just a really interesting it's interesting helping science. It's it's kind of why I like doing it. I think it's interesting when you look back at how science works. It's so much luck driven.
Jon Hammant:Like, it really is. When you look at, like, large like, you take, stainless steel, for example. So it's pure luck pure luck that we discovered stainless steel. What we weren't looking for it. We we didn't specifically go after it.
Jon Hammant:We didn't think anything with the qualities that have it even existed. It was just literally through, a British scientist trying basically looking for new shell casings, trying loads in a row, and then finding one that he thought had failed. It didn't even have the qualities that he wanted it to have, and then suddenly working out that this is we've created this this fantastic new material that, you know, means effect that you can eat with a fork, but I don't taste. There's no reason that it happened when it did. It could have happened much earlier.
Jon Hammant:And equally, it probably could have happened a fair bit later. I think what really incites me about computing at the moment is we're kind of on the we're on the spiky exponential side of Moore's Law at the moment that, like, exponential changes. Everyone kinda realized in COVID that exponential one once it starts going, it's pretty pretty going, if that makes
Aaron Phethean:sense.
Jon Hammant:And I think we're now going through this really interesting situation through exponential change in tech. And, actually, that just so for me, it changes the kind of luck surface almost. So the ability for us to just try things which have huge, huge combinatorial, unlikeliness of happening Yeah. It means because we can chuck so much compute to the whole thing, we can do some really amazing things. And that that that really does excite me, actually.
Aaron Phethean:Yeah. I mean, I I definitely agree that, yeah, that kind of experimentation and I feel like that's a human ish trait. Yeah. Obviously, you could build, you know, a a set of agents, a set of AI that's still willing to experiment. Yeah.
Aaron Phethean:But it tends it it seems to me at the moment that tends to be a weakness. Like, it's it gets quite good at perfecting too early on almost. And maybe that is also the kind of reason why stainless steel's type discovery happened so late is that, you know, there's not really a willingness when you're perfecting something that's already quite valuable. Don't throw away.
Jon Hammant:How how do you avoid being stuck in the local maxima? That's the that's the kind of that's the reality of it. And that and I think, you know, like, it's interesting, but I think in my career so so some of the most fun I had in my career were things when the feedback when that feedback loop was at its shortest. So for example, you know, I used to, used to work on a gambling, like e gaming. It's really good because you can get stuff live quickly.
Jon Hammant:It's all really, you know, really quick paced. And I think one of the things that we need to work out, it's gonna be interesting as AI starts to move into large scale organizational direction. Because the thing that we need to make sure is, one, that we're putting that feedback loop in, and, two, that the thing that it's optimizing for is the thing you you kind of really wanna optimize for. Like, how do you to your to your point, how do you avoid AI driving a local maxima that effectively maximizes short term short term profits for a lot of people, but destroys the three, five year, ten year horizon on a bunch of these companies? I think that's Yeah.
Jon Hammant:That's gonna be really, really interesting to see effectively play out, which will end up with some massive losers and some massive winners.
Aaron Phethean:There's a there's a whole scene there because, you know, your your kind of DevOps history, you know, that that's also been my history, actually. So I was with a core banking vendor for quite a long time. And building platforms, building technology change promotion, you know, it's all it's all part of why those kinds of projects take a very long time. And then, you know, I think the kind of relevance to now and this this conversation, I mean, it's it's a, the compute side of things. Yeah.
Aaron Phethean:A little bit, b, the kind of process. You know, the there's a there's a whole ton of things. Actually, I'd love to ask you about you know, I I love the Phoenix project, and I'm writing an article now about about the book. I I love it because it's it's described how organizations change and how they Yeah. Build up things.
Aaron Phethean:And and also it's a little bit cringey because you're like, oh, I know that character. I don't know that situation. I don't know. It's sort of that hopeless feeling. Yeah.
Aaron Phethean:So what Yeah. What's the link for you between, well, cloud and DevOps and and AI? Are are they driving one another, or are they are they completely different disciplines?
Jon Hammant:So I think, so I can definitely talk first. So so, you know, where do I think? So I think, fundamentally, it's been really interesting to see how cloud is changing the business model of just so many different things, actually. I think that's one thing that that I really do like. So if I look at so I look at the eight so e c two has huge number of instances now.
Jon Hammant:Even more, I'd I'd I'd quite a stat, but I'd be wrong quite honestly because there'll probably be a new one.
Aaron Phethean:What ballpark? I mean, it's it's one of those things that's like that, you know, it's hard to imagine how much compute there is. I gave the talk to a bunch of, you know, five, six year olds at one point. It's it's mind blowing for them that there's computers everywhere.
Jon Hammant:No. It's it's unbelievable. I mean, I wanna I wanna say we're over nine nine hundred thousand instances live. I think we're I think we're around that. We're we're a lot of different unique instances, which is it's kind of there's one instance in all of it that that's, you know, now it's been super silly, but it was kind of one of my favorites.
Jon Hammant:There was a specific video transcoding instance that we had, and all it did, it was the genuinely, the best and the cheapest for transcoding h two six, h two six five streams effectively. Multiple users. It was it was fantastic. Yeah. And why did I think that was really interesting?
Jon Hammant:Well, it was something that couldn't exist in other contexts because each one of them was moderately expensive, was was very expensive for us to to build and put together. It had a low user take up effectively. Like, people wouldn't need that many or they'd need some or need it for a bit, but not a huge amount
Aaron Phethean:of time. Mhmm. And so
Jon Hammant:if you think in it to how you build that in a traditional business, almost that friction of going out and selling them and getting them to people made it just impossible too. You just couldn't have built a business without that. Yeah. Yeah. All of a sudden, as you can change the business model around, so people come to the come come to the infrastructure while the infrastructure's coming to them.
Jon Hammant:I think it means you can have far more specialization in really kind of interesting esoteric, sides of things. And I think that's been real and I think that's gonna happen both in terms of it's happened in terms of infrastructure and services, but it is that yeah. More and more, it's that very specific service lens that's gonna Yeah. That almost, compartmentalization of things is is super interesting.
Aaron Phethean:I have a, I have a bit of an obsession at the moment, and, I'm sure you've seen some of my LinkedIn posts.
Jon Hammant:Yeah.
Aaron Phethean:I I essentially I think
Jon Hammant:a lot
Aaron Phethean:of what we just discussed around computing and availability and scale, you know, and actually, like, the friction to a, you know, process of obtaining it is, you know, if you if you remove that stuff, demand kind of massively increases. And I I one of the things I talk about when I talk about pricing, what I'm kind of obsessed around pricing is as a company, a software vendor, I I think it's quite abnormal to try and lower pricing. You know, actually, so transfer wise, this is a customer who's trying to lower pricing, lower lower cost of of moving money. That's sort of one of their fundamentals. And you think that just it just doesn't sound right.
Aaron Phethean:Like, you know, from one perspective, you think a company like that wants to maximize profit. They want to charge more. But they're in a utility kind of business. They're in a business where people need more of it, should, you know, have access to more of it, and then will demand more of it. So, you know, I think the, you know, the challenge AI on the horizon, the type of service demand, I see being a a a data led utility Yeah.
Aaron Phethean:Compute led. Yeah. You know, these these other sort of there's gonna be surrounding things that are driven by this this this central demand that everyone sees.
Jon Hammant:Yeah. And I think so it's so one of the things I I genuinely like about working at Amazon AWS is the the fund some of the fundamentals of where our business model sits and how the economic drivers work. And it's always nice because, yeah, you can always not believe people of their reason. People can turn around, and you can think people do stuff for whatever reason, and that's fine. But there's definitely a different thing when you can see the economic incentive that is driving that company to do things.
Jon Hammant:I think it's far more, like, honest in a weird way. Effectively, for us, I mean, we've lowered prices so many times. And kind of why have we done that? Well, the the whole business model is that we wanna try and make sure for customers that the individual cost of a widget, whatever that widget is, in terms of if that's data, if that's a service, if that's product, we wanna reduce the cost of of those individual widgets, and we wanna do that to effectively enable you to grow and to sell more of them or to reduce more of them. And that is that is the business model.
Jon Hammant:Like, no matter what else you do. And I think that's really good for me because even the fact that we're a a consumption based business, you know, we we can we can go I can go off today, and I can convince someone to sign a bit of paper that says they're gonna use x amount of AWS or or money if I can't. I can I can ask them nicely to sign a bit of paper?
Aaron Phethean:You can hope.
Jon Hammant:Yeah. You can hope. Exactly. But the reality is unless they actually use it, unless they actually consume that, that revenue is not being generated. And the only reason they're ever gonna consume that is if it has benefit to them.
Jon Hammant:And the only reason they're ever gonna keep consuming that is if it has benefit to them in the future. So it's this really nice situation that we get into where we start to align economic incentives for everyone, and that's from the end consumer, the customer from us, AWS itself. And that's that's kind of where I think Evinced is really a job to. And I think that that is a different business model from maybe some of the software, vent. You know, I used to buy a load of software off all kinds of different companies of which I won't mention their names, and, you know, they would definitely
Aaron Phethean:You referred to very early on, you know, half joking here, the the kind of legacy. You know, you actually used it as heritage.
Jon Hammant:The heritage.
Aaron Phethean:I I sort of feel one of the reasons that software and technology becomes legacy is because they don't they don't they have not adapted, you know, have not adapted the way they work and the way they charge. You know, I do I do feel very frustrated with enterprise type software sales processes. Yeah. Then the outcomes and and are often far short of what this sort of promise was, you know, that there's these huge beasts, you know, whereas the kind of, you know, much smaller incremental, you know, the the value exchange is is much more real time in in, you know, a kind of a utility or
Jon Hammant:a small service. This is what I love. I mean, I loved being in the web of DevOps. I really did. I thought for and I and I know it's now a thing that just it's just accepted, so I totally get that.
Jon Hammant:But it was I loved it at the time because there was some it always reminded me of, kind of like a a a drunk person trying to find a kebab shop after an evening out where, like, the reality is if you if you just keep walking one direction and you only look at your map and try and work out how you're finding your food, like, once every twenty minutes or once every hour, you just end up really lost, and you don't you don't end up there. But if you kind of instead, if you constantly kind of check which way you're going, you you kind of meander into the right path. And I think it's, you know, the only thing that tells you really if your customers like your software is your customers using your software. That is that is it. There is nothing there is nothing else.
Jon Hammant:Everything else is just a just a proxy metric. And I think the fact that software delivery has changed so much in terms of that you can, yeah, you can code it. It can be but when we launched, we I've I've not seen our reasons. It's it's millions of times a day AWS is deploying to prod. And I think that's been such an interesting thing.
Jon Hammant:And, again, I think that is something that the fact that you can blue green deploy, the fact that, you know, if I if I turn around and might think back into my data center days, like, I wanted to deploy a whole another production stack. Well, that's gonna be a really challenging thing because Yeah. I've gotta have a whole other production stack for the thing. I even if I need that thing for
Aaron Phethean:Well, there's now business as normal. Right? Yes.
Jon Hammant:That's that's the
Aaron Phethean:way you do things. And then that's perhaps a a good point to try and wrap up on. I I I love that analogy. And my my kinda takeaway from hearing that is that think carefully about your destination. You do wanna go to the kebab shop.
Aaron Phethean:Yeah. But then check really often that you're on the right path to get there. That that just feels like a really good bit of advice for anyone listening. You know, when they've got a big migration project, where they've got something tiny to deliver, they've got AI demands coming down the pipe, think carefully, and then just check really often.
Jon Hammant:Yeah. No. Exactly. I think I think there's no there's no substitute for that feedback loop that you can get through. The more that the more you get stuff into production, the more that you accept that it's gonna fail sometimes, and the more you keep it going, the the better.
Jon Hammant:But fantastic job, actually.
Aaron Phethean:Cool. Well, thank you, Doug. Thank you very much for coming on the show.