תעתיק 🔗
Anne: hello and welcome to episode 21 of Asynchronous and Unreliable, a new weekly podcast where we discuss the latest ideas and concepts in tech. I'm your host, Anne Curry, co-author of Building Green Software, the Cloud Native Attitude, and author of the science fiction Panopticon series. And today I have the great pleasure to welcome Sarah Wells, independent tech consultant, author of O'Reilly's Enabling Microservices Success, and former technical director at the Financial Times newspaper, which is which is how we met. In fact, I have a whole chapter in Cloud Native Attitude devoted to to to what Sarah was doing at the Financial Times and the work there about going cloud native. So basically Sarah, your background is all about getting enterprises to adopt and benefit from modern modern practices. so welcome.
Sarah: Yes. Hi. Thank you for inviting me. This is I'm looking forward to this.
Anne: Yeah, you've you've interviewed me 'cause you do the go it's go to was it go to you or was it O'Reilly? Go to
Sarah: I think it was go to, yeah. I yeah, just but they like to get two people together who're gonna chat and we've done that before.
Anne: Yeah, so we have been on a podcast before when Sarah was hosting and I was yes. So it's a reversal. Although to be honest, it doesn't really make that much difference, does it?
Sarah: No.
Anne: so so one of the reasons why I've been quite keen to talk to you on the podcast, well for many reasons, because because it's it's very it's a lot of fun to talk to you. we keep we've run into one another quite a few times in the past month or so at AI conferences in London
Sarah: Mm-hmm.
Anne: you so you've been out there, you've actually been out there as you do as a tech consultant. You go out, you talk to a lot of companies, you know because you've actually, you know, the Financial Times is probably the most successful online newspaper, is it?
Sarah: It certainly
Anne: Well,
Sarah: is very successful. I think it has a very good understanding of what it's trying to the audience it's trying to address. And and also the other thing the FT did that was very good was they did a very they did a very successful transfer from thinking about the newspaper first to thinking about about online first. And that is that's a that's something that's quite that was a challenge, I think, for a lot of newspapers to start thinking in terms of what is it going to look like when it's on the website. And then as soon as journalists got used to that, it was what what is it going to look like when it's on mobile? And it's very
Anne: Right.
Sarah: different from the layout on a on paper.
Anne: so I I I is one of my favourite chapters in Cloud Native Attitude because I I I do a lot in it about in Cloud Native Attitude where I went out and talked to people who were really doing the stuff and what
Sarah: Mm-hmm.
Anne: were their problems, what were their issues. I followed back over and over again over the course of years. And the FT and your experience there was very was a very classic experience in that it took a long time. You didn't you know,
Sarah: Yeah.
Anne: you didn't expect to be in the cloud from on prem overnight. And you weren't.
Sarah: No. No. I mean, and there's two sides to that, I think, which is the actual time it takes to get everything out of a data center. It's not an a an instant thing. But there's also the cultural change that you need to have to be able to really benefit from from the cloud, to really be cloud native. So it wasn't just we are decommissioning our data center. It is well, we're going to build microservice-based systems and we're going to use containers or lambdas or use cloud components rather than just deploying our applications in exactly the same way on AWS as we did in in the data center. So I think that cultural change is also big. And and also we were doing it at a time where many of these things people hadn't done. So it wasn't easy to go and look at what other people were doing and say, this is how you do testing in a in c in microservices and in containerized applications and and everything because it was very early days for a lot of these technologies. So you're working it out as well as you know th there's a load of things where I talk to people and I say, well, you know, it would have been great to use this particular technology. It just didn't exist when we were doing it. So Kubernetes did not exist when we were first running containers. So we built our own orchestrator. Obviously you don't do that. backstage and portals like that didn't exist when we were first starting to struggle with the idea of we need to keep track of our of our software estate. So we built our own. But it never would be my first thing. If someone else has built it, I would rather use it. But we were so early we had to do a lot of that work.
Anne: it was interesting. I mean you you told me a lot of stories about the early days of of the FT. And and when I spoke to you, I think you was you were a couple of years in. It wasn't the earliest days, you know, it was
Sarah: Mm-hmm.
Anne: you were actually successfully partially in the cloud, not completely in the cloud, but partially in the cloud. by the time the final or the last edition of CloudNative Strategy came out, you were totally in the cloud. But you didn't attempt to do it in one go. And it there were there were lots of interesting things that you learned through trial and error on the on the
Sarah: Yeah.
Anne: way. Yeah.
Sarah: Yeah, you don't get it right first time all the time. And actually that's something where as a consultant you quite often go in somewhere and you're not there for that long. But it when you know, I was at the FT for 11 years. So I was there the whole way through this this transformation. And so you've tried one thing, you've discovered it doesn't work, and you've moved to something else. You start to really hit the the challenges of understanding was this was this a sensible
Load full transcript
approach.
Anne: Indeed. Yes. I think the the the realization that you came to that I repeated to other people mo mo more often than any other was when you said that when you first went live in the cloud, the AWS bills were crazy, just like crazily expensive. And and you you learned that you had to give the information about the bills to the people who were cr who were who were running up those bills and their and responsibility and agency to them to drive those bills down. And then they did it. And that that was completely transformational.
Sarah: I think visibility in pretty much in in lots of aspects. So certainly in terms of your costs for things like cloud. If you show people what they're using, they'll turn around and say, Well, we're not actually using that. We could we could get rid of that. visibility in terms of what services have you got and who owns them and how secure are they and anything that that gives people something where they can look at it and say, we haven't done this thing. It is really useful. So one example we had was we we created this thing called the system operability score. We were trying to improve our runbooks. And we had lots of services. I mean, we we probably had a couple of thousand services across everything at the FT. And some of them had very limited good documentation. So we had this runbook format and we created the score where we basically said, here are the most important fields, here are the next most important fields. And we we kind of aggregated it all together and said, Here's the score for this. service, here's the score across this team, here's the score across the group. And that visibility for a lot of people really encouraged them to think, okay, I know what I need to do to improve the run books. We had teams that went from nothing to a hundred percent run book coverage very quickly.
Anne: kind of lead each other kind of thing, the you sh sharing best practice, sharing what good looks like.
Sarah: I think, yeah, show sharing what good looks like. But also the very first thing like this at the FT was way before I was I was a director of technology and our operations team would send her out an email once a week with the the five noisiest alerts. And I've got to say the moment where I became where I actually concentrated on this was when my team had was in that top five noisiest alerts. I was like, that's embarrassing, let's fix that. So I think that you can Like anything that that shows people, hey, this is actually really loud, can you fix it? It is good. Yeah.
Anne: Yeah. That's very good. So one of the reasons why I'm quite interested that I'm very interested to talk to you at the moment is you were right in at the big at the beginning of Cloud Native, or in
Sarah: Mm-hmm.
Anne: a in a proper major enterprise having to do it and you successfully did it and it took a long time and it was really difficult and you learned loads of things, loads of try and error. Now you're out and about at what is
almost the equivalent moment in the next giant change in the industry.
Sarah: Yeah.
Anne: AI? How how is
Sarah: Yes.
Anne: how is it the same and how is it different?
Sarah: I think there's more worry about where this is going with AI. It f it feels like it's a bigger transformation than than moving to the cloud. I don't know where I stand on that. I I feel like at the moment my f my view is that it's an extremely useful tool. It is going to change the way that we approach coding, but it looks more to me like like it's a next generation of programming. you're still
Anne: Yeah.
Sarah: thinking about software engineering, you're just using a different tool to do it. I think it affects everyone. So quite qu there there's always been an ability for lots of software engineers to just get on with what they're doing and not worry about the fact we're moving out of the data center or or we're not worrying about the change in the way we do architecture. But this is this is going to impact pretty much everybody. It
Anne: Yeah.
Sarah: it's really right there and being used by by engineering teams and it's changing the way those teams work quite significantly. so and I think it's like there's a lot of s there's far more people I think that are sitting there thinking, what's it does it mean for me and my my career as
Anne: Yeah.
Sarah: well.
Anne: Yeah. I mean speed wise, how do you think it compares to the beginning of cloud native, say at the
Sarah: In terms of adoption, it feels like it's going very quick. It
Anne: Really tell us what
Sarah: really does. Yeah. And it's interesting because I think it's outpacing organizations' ability to understand what they're doing. So
Anne: Yeah.
Sarah: there's a big gap between the number of organizations that are that have people using AI and and organizations that have really thought about about how they want to govern that. So I think there was some recent research that said you said basically almost no one's got like AI specific governance. Like 25% of of companies don't have an AI policy, 38% have a comprehensive policy. And yet 90% of organizations have people using AI. So there's so there's a there's a gap. The other gap is in lots of those organizations, the AI use is being done through personal accounts, which means that you really don't know what's happening. But But the but again, interestingly, the research says, well, the the number one thing that gets post pasted into these personal those personal chats that developers are using is source code. So your source code is being loaded into chat GPT by your employees. and you're not really on top of that. So I think there's it's there's a real pressure to say that you're using AI to get on with it, but there's also a a gap. And I think it of exacerbates that thing where where it's always been the case that that people think of compliance and security as something that slows you down. And people have really invested in the last five, ten years to make it less like that. But now you you really have to work super hard to keep up with what's happening and to be n and to be helping people be more secure and more compliant without being seen as a as a real gatekeeper in front of all of this.
Anne: the security risks at the moment are crazy. I saw one of your other fellow people who were featured heavily in Cloud Native Attitude, which was a a chap called Greg Hawkins of Starling Bank. And he was
Sarah: Mm-hmm.
Anne: I was talking to him about this the last week and he pointed out to me that I think it was the week before last, Microsoft's Patch Tuesday, where they release a whole load of patches to all of their kind of enterprise software, was the biggest ever, 600 patches that needed to be applied that week. and that was that was second only to the previous week, which was the second greatest, which had been the greatest ever patch Tuesday, which had been 200 patches. And and these have to be applied by enterprises. Although you know, to to everyone's desktop, to everybody's systems, it's it's astonishing, isn't it?
Sarah: I think it yeah, it's terrifying. It it so it's just AI makes it a lot of things easy to do because it with less like there's less investment to be able to try and do things. So there are lots so within an engineering team, there are lots of things where you go, I'll just build this one off tool because I can do it really quickly using my AI agent. In a way that I would never have been worth it when I was having to do it myself. But this is the same that's happening with security. It's much easier to go, is there any vulnerability here? It's going to improve the security of everything, is just in the meantime, it's going to be extremely painful. and that's that's that's one aspect of it. The the other aspect is that so I was reading, so I'm writing a workshop at the moment about governance for it for AI-enabled organizations. Particularly looking at at some of the security side of it. And there's a really interesting but there's basically some research that says you look at the code that's being committed and look at you test the agents, they're getting really good at writing code that's that's acceptable quality, but they're not getting any better at catching security issues. So if you are a bit overwhelmed in an organization because there are so many PRs because everyone's using agents and you're starting to get and you're not doing a full PR and frankly, code reviews were always a bit problematic for catching stuff like that. But if you don't have anything that is looking for security vulnerabilities in the code that's being produced, you can't you can't trust that the that the AI agents are going to be good at this. The interesting thing is that they are particularly bad at Java. And that looks as though it is to do with the stuff that that it's being trained on. So there's a lot of insecure Java code that these
Anne: Yeah.
Sarah: that that will have been used in to effectively
train all of the of these agents. So I think it's quite scary is that the security is being left behind.
Anne: Yeah, it is it's interesting you say that about Java because Java's come up several times in the in the i in previous podcast episodes.
Sarah: Yeah.
Anne: I talked to Holly Cummins, who I don't know if you know from Quarkus, you know, the about and it's been a move for many years that Java hasn't been very good at you being able to turn it off to apply patches and things like that. And and so so
Sarah: Mm-hmm.
Anne: they've been they invented this new version, Quarkus that. could be turned off and on and patches could be applied. But that's still not the majority of people who are running that.
Sarah: No.
Anne: Especially if they're not saying, I mean, because Quarkus was like, well, you're really going to be in a mess in the cloud if you can't turn things off and on again. But a lot of a lot of Java folk are not in the cloud. In fact, most people still, enterprises are not in the cloud, let's say, or not properly in, not fully in the cloud. So they can't necessarily patch these Java systems all that easily. And Justin Cormac, who I don't know if you've met before, who was one of the founders of Unikernels, was on a
Sarah: Mm-hmm.
Anne: couple of weeks ago to talk about about how good AI was at writing in particular languages. And one it one thing he pointed out was that it was very good at writing Rust, mostly because Rust is quite modern, and most of the code that's written in Rust is quite recent and follows really quite good modern best practices. But languages I mean he used C and C plus plus as an example of a language where the trouble is if you've got a huge a huge kind of hist historical mass of code and of
Sarah: Yes.
Anne: very variable quality, the AI gets very confused.
Sarah: Yeah. I and I imagine that there is a large amount of code for Java. And if you're not saying only look at code that's been written recently that will be using more modern approaches, you're going to end up basically predicting pretty average code. And I think this has been a general question. Yeah. Been
Anne: Mm. At best.
Sarah: a general question, isn't it? Is you're going to get the mo the average of all the code and the average is probably pretty bad. I think that's gonna be a problem. what's interesting is it's just that it's not improving either.
Anne: Mm.
Sarah: So a lot of the improvements that are happening for AI agents, it's around the LLM. That there's a lot of work being put into to make sure that that that everything that people hit and they you know, it it there was that that basic big jump earlier this year where suddenly coding agents were really, really usable. The much less frustrating. Many of the things that were kind that were irritating went, and I I assume it's a lot of the work in the in the sort of harness around the LLM that that the coding agents are providing, but it's not happening for security. And that that worries me. and the the research suggests that there are certain types of security vulnerability that are easier for agents to deal with. And the ones that are they're really bad at are the ones that require you to have that broader context. So I was reading something about cross-site scripting vulnerabilities are quite hard for agents to catch, but SQL SQL injection is quite easy because it's very localized. You can actually look at that particular query. And I think that this is something I need to do more reading on because I'm just really fascinated by that.
Anne: Yeah.
Sarah: but I've been doing a bit of work with security engineering teams recently, and it's just tough. It's tough to think about how you're gonna kind of make sure that everyone is still secure and stays
Anne: Yeah.
Sarah: secure.
Anne: Yeah.
Sarah: and it and it has to do with knowing that there are some vulnerabilities to patch, getting everybody to patch them quickly, knowing that code is now getting produced quite in large volumes and people are not necessarily able to stay on top of all of the of all of the code reviews. Maybe they're they're making decisions about what things they review. How do you put things in place to to provide you with automated governance? Because I think you're gonna need you're gonna need that. You can't you can't expect to be able to automate so that you can perfectly grab everything, but you ought to have as many tools as possible i i in your CI C D pipeline to tell you whether things are going are looking wrong.
Anne: so those are the right things to do and those are the things that you're telling people to do. What kind of uptake is there on good practice?
Sarah: I think you have to you've to do a couple of things. You need to make it really clear what the expectations are for people. Because my my experience is that that software engineers generally want to do the right thing. They don't always know what the right thing is because you're not always clear about it. So if you can say, here's what we want you to do to try and make sure that your code is secure, that that they'll do it. And and they'll and if you can put things in that are that automate doing some kind of validation and then say, hey, you need to fix this. That that's very appreciated. So at the FT we created an engineering checklist that that had a selection of things that you needed to do to get to build code and get it to production. So not writing code or testing. We assume teams know how to do that, but it explained what was expected if you needed to, if you needed to procure something, what you what you were supposed to do for security, what you should do for accessibility, all what documentation you needed to have in place. And I I had a Prince Vengineer who came up to me and said, Well, you know, I really, I really like this because I know if I can tick all of the boxes that I've that I've done my job. And I think that's so I think being really clear, having that policy in place
that says, you know, only use these AI tools because they're the ones that we've approved. I know you want to use everything, but actually, if you need something new, you have to have a security review. It has to be reviewed by legal and compliance. That works provided you provide. A bunch of tools to start with. So you're not saying, hey, you can't do anything, but you're saying, here's a good subset and tell us when there's something missing that you that you would like to get. That's that's a good thing. Explaining what's expected from people, and and there's that whole sense of now there are now people writing code who are not software engineers. How do you help them to build something that is that is secure and isn't going to Is it isn't gonna be a shock to everyone when they discover that it exists. Because you hear stories about people who go to someone went to their GP and discovered that the GP had vibe coded something, vibe coded something that effectively meant patients' data was just available, but they don't know this. How would they know that? So I think if you're in an organization, you need to you need to provide that kind of guidance to people. Here's what you need to worry about, here's how you check that you're okay. one organization I work with have a They they've always had a thing about a security review before something goes into production. And it's pretty it it's intended not to slow everyone down, but have a chat with the security team, you can you can do some threat modeling. and the there's a customer success manager there who knew that this was a requirement and had written a a Chrome extension and and cook came and arranged to have a security review. And I thought, well, that is fantastic, because you've
Anne: Yeah.
Sarah: managed to communicate that this is something that matters.
Anne: Mm.
Sarah: And you've made it into not an in too intimidating or or horrible a process that that they're engaging in it.
Anne: You mentioned last time I spoke to you a a failure mode that you had observed. that and you could totally see how this happens, that actually people ask the AI, what's the
Sarah: Mm-hmm.
Anne: what's the right thing to do? Is this the right thing to do? And actually that's not That is not sufficient because AIs tend to go, yes, that's fine. And and that can revers result in kind of escalating issues and complexity and risk taking and because the AI is so encouraging.
Sarah: I think that I think that's true. Although I don't I don't tend to struggle too much with the with I find with coding I I use Claude, I don't find Claude too kind of, yes, that's brilliant. what I think's interesting is that that you have to be asking the right questions and providing the right guidance. So if you don't know how to frame the conversation, it's very hard to just say, make this secure. You want
Anne: Yeah, yeah, yeah.
Sarah: to say run a security review, look for these things, look for the OAR top 10, something like that. Something that that actually you have to have that context. And I know we were talking about for junior engineers, you don't necessarily have that context. And I think it is it's the danger is that a junior engineer goes off with their c with their coding buddy and does a load of work and is going the wrong direction. And if they were working And working with a senior engineer, they'd they'd get a course correction early and they'd be learning as they went. And the danger, I think, is that you sit there with your coding agent and you come back with this thing and someone else now looks at it and goes, Well, that is just not the right approach. It's it's far too complicated. You've got this massively long method in the middle of it. All the stuff that, you know, you would look at it and go, These things worry me. But But someone's already invested days of their time and it's very hard to then fix that. So
Anne: You know.
Sarah: I'm worried about how what it looks like to to work where if people just go and work with a an AI agent and then they come back with their ta-da, here is my code, where's the collaboration and the learning and how do junior engineers become people who know the right questions to ask?
Anne: Yeah. Yeah. Well I guess they probably learn the hard way when they take it all. But the trouble is it's too much of a big lesson at that point. There's no point going,
Sarah: Yeah, I
Anne: throw all this out, you don't really learn anything. You have to learn the mistakes one at a time, don't you?
Sarah: You you have to you have to learn it in the time and also it's that thing about it's much easier to fix someone's opinion about something before they've finished it. You're very invested in anything you in anything that you've written, like finished. So you need you need really to have that course correction while you're doing it. I totally recognize this because I feel like when I'm when I'm coding with an agent, I've got 25 years experience of the things that could catch you out. So things like, is this a is this a CSV? Well, I'll bet you there are some unencoded commas somewhere in there, because that's what always happens. This has happened. and I j I think there's a load, and and this happened to me fairly recently. It's like, would you so I just said to the agent, you know, check for this. so I I do feel that we have to find ways to Still let people benefit from working with an agent, if if that's the right thing, but also learn from their other colleagues.
Anne: Yeah, it's it's interesting because John, my husband, who you know, who's who's often on this podcast he was saying, it's interesting because that's not a new failure mode. You have to well if people are off doing something, you have to, as a manager, touch base with them regularly
Sarah: Yes.
Anne: and say, What are you doing? Is everything okay? Describe it to me. You know, you do have to keep a a a an eye on them. so you can't let them go off too long. But I think the
issue at the moment is that they can Between two touch points now, they can do a lot, can't they?
Sarah: Yeah. Yeah. I think there's a a thing where you you want to go and work out how you're going to do something without revealing that you don't understand it. And that's always been the case. that's why pair programming was good because you had to you if provided you had the trust in that in that pairing and in or ensemble programming, you could learn together and that was really good. And I think the pro problem for me is that agents make it really easy to go off and just learn go I don't know how to do this tell me tell me more about this but it stops you from from really admitting that there's something that you're not not familiar with.
Anne: Yeah.
Sarah: so I d I think we have to work out how we how we change collaboration to include the fact that many people are are using agents.
Anne: Yeah.
Sarah: meet meter did some research into developer productivity. In 2025. And they were that it was interesting research because they basically found that people's perception of how productive they were with AI agents versus not without AI agents wasn't very accurate. So they tried to run the same research in 2026, but they actually found that they had lots of developers who, even when paid, would not do tasks without the AI agent. So they had to change the style the actual form art of their ex of their research because people just could not be removed from using their AI agent. Which I think is I don't know, I think it's slightly scary.
Anne: Mm. It's
Sarah: Like that's that's a crutch pretty quickly. That's become a major thing that people are like, Yeah, this is how I do this is how I do my job.
Anne: Which is interesting. so we're both O'Reilly authors, and w yeah many years ago, kind of like 10, 20 years ago, 30 years ago in our jobs, if you didn't have a book, you couldn't do your job. You know, it was like
Sarah: Yes.
Anne: you had to look up everything because you
Sarah: Yeah.
Anne: there was no there was no way of finding it out. Almost always you'd sit there coding with a book and you couldn't do it with that book. And you couldn't even even basic things without the book, because you didn't
Sarah: Mm-hmm.
Anne: you didn't commit all of that stuff to memory. So in some ways it's interesting. my guess is an awful lot of programmers in the intervening period couldn't do coke programming without Stack Overflow, which is which now seems like Yeah, indeed. Yeah.
Sarah: Or without their IDE. Yeah. So like auto-complete, all of these things where I was like, I don't have to remember the syntax because there's something to help me with this. Yep, Stack Overflow, I'm I'm I can benefit from other people having hit the same problem before me. And and agents do even more of that. There's a there's a there's a really great sense of I've got this error, tell me how to fix it. And nine times out of ten, the agent immediately is able to Tell me what the problem is, including in including in languages I don't know very well, which you know, especially in languages I don't know very well.
Anne: Yeah.
Sarah: and it's been sufficiently good that I that I tend to think, yeah, this makes sense to me generally. I've still got that
Anne: Mm-hmm.
Sarah: the still got those software engineer instincts of I don't think I like your design for this. But yeah, I d I do think it's you're right, it's just the next iteration. It's the next iteration of I now have something that helps me with the with the the basics. Just
Anne: Yeah.
Sarah: the basics is bigger now.
Anne: Yeah yeah. Yeah, I mean it's it's it's a long time since we wrote code, very much code, without any reference at all, isn't it?
Sarah: Yeah.
Anne: it's it is a continuum. also things are changing very quickly, it's very hard to know what direction things are gonna go in. As you say, the issue is who's going to teach the the the senior programmers, but then also the issue is are are we are those are we just going to age out senior programmers with the senior programmers?
Sarah: I feel that a year ago there were a lot of people saying, hey, I can vibe code an application in an afternoon. I don't have to be a software engineer, I can do that. I think there are still people that are building, are building things, but they tend to be very much personal stuff. That bit of turning something into a product that you can That you can sell to other people and that can be used by large numbers of people still requires software engineering and something that is maintainable. I know that there are some some thoughts that maybe maybe in five years' time you you you'll just have the agent rewrite the code from the specs every time that you need something something new. Maybe maybe that is the way it's gonna go. But my my sense at the moment is that there's still value in in having that software engineering mindset, the design, the the like the the and I don't know if we're gonna get to the point where we where we say we don't need those people, you know. I think we will still need people and I suspect we'll still need them once a lot of people start retiring out of out of the profession.
Anne: I mean, we certainly we're in in the transition at the moment, and the transition could go on another 20 or 30 years, you know, who knows? These it always
Sarah: Yeah.
Anne: takes a long time. But the senior engineers now could be around in the industry for another 20 or 30 years. I mean, it's at COBOL,
Sarah: Yeah.
Anne: there are still plenty of people writing COBOL and the the people who support the the the architecture of COBOL have not retired, not fully. Not least because they're getting paid very good money.
Sarah: Ha ha ha.
Anne: So, you know, it might be that people like us who you know, will just continue to be tech consultants for the next twenty or thirty years and then hand over the baton to whatever is the fable of them of the time.
Sarah: I do think it we
are still very early days in it it even though it feels like everything's changing very rapidly. I think when we look back on it in five years' time, it'll be well, of course we didn't understand what that was going to be like because
Anne: Yeah.
Sarah: we're not we're not there, we're not there yet. and I guess I guess it reminds me and I was probably only just starting to work in in IT at the time that the internet really was was taking off. I I my first job was during the dot com boom. And there's a there's a lot of people making their comparison of we thought, you know, all of these companies were going to be massively successful and some of them went bust because we totally we we just didn't didn't didn't make the right predictions. But then eventually the the internet did happen. It's just I'm I was thinking of a company that I loved when I was in my first job in IT, which was called Urban Fetch. And
Anne: Okay.
Sarah: it would so it would basically you could order food or CDs or I'm gonna date myself here or anything and have it delivered to your work location for the price of the thing. and it was extremely popular in the software company that I was was working in. They might as well have had someone sat just outside our office that were constantly bringing stuff, but they went bust because they didn't have any any real ability to pred how they were gonna get and make money. They're basically charging us the
Anne: Thank you.
Sarah: cost of whatever they're delivering us. But I think now we have all these things that instant delivery. It's just it took a long time to to catch up with this is possible. It was
Anne: Yeah.
Sarah: too early. And I think we probably have something similar happening with AI. The things the people that look like they're going to be really successful, maybe they're not the people that in the end are going to gonna be the ones that that are still around in 10 years' time, but and and what we're gonna be doing with it might change as well.
Anne: Yeah, yeah, no yeah.
Sarah: So that's a complete sidetrack. I miss Urban Fetch.
Anne: Yeah, it's yes, it's it it is very interesting, isn't it? That we just don't know. I mean, it's like you and I met ten years ago, cloud native, and we thought at the time cloud native was such a you know, more resilient systems, more scalable systems, faster time to new features. It felt like everybody who wasn't cloud native was gonna be out of business because it was such an advantage. Ten years later, that's a really There are still loads of people who most people are not.
Sarah: I think you start to realise how many organizations faster is not like the the bottleneck isn't there.
Anne: Mm.
Sarah: Like it there is no point. And this is going to be an interesting thing, I think, with the fact we can we can write a lot more code, we can respond much more quickly to requirements. I don't think we're very good at working out what we should build. And when I think about when I was when I was
running a software development team. I used to look at our backlog and after a while I stopped worrying about any user stories that were below about the third slot on that backlog and the ones we were working on, but the ones that were coming up afterwards, because by the time we got to number three, we would have reprioritized, something new would have come along, we would be doing something else. And probably those stories that were in slots four, five, six were not things that we we should be doing. And I'm worried that like one of the things that will happen in the AI is we're going to code everything and it's all going to go live and and suddenly you've just put tons of new features in and no one actually wants them and no one's really worked out whether it unless you're an organization that's actually very good at predicting at being able to tell whether what you've built is something people want, then how are you not just gonna end up building loads of features that that don't improve your product?
Anne: Yeah, there's a Simpsons episode where Homer Simpson designs a car and it's ridiculously over engineered with every single thing you could possibly have in it. And it it's it's a total flop because it's ridiculously overengineered. People don't want everything in the world. They want they want someone to to curate what goes in their car for the moment.
Sarah: Yeah, maybe we'll have the flip side of everyone just we'll start to have products that are like, just do this one thing. Like the like the original Google search. Do you want to just
Anne: Yeah.
Sarah: search for something and have the results be useful? Yes. That would be great.
Anne: Yeah, that is true. Although I'm learning loads of new things as a result of talking to the AIs and them doing the research for me that's that I wouldn't have to do. And then I'll follow it up and double check it all. But I'm not having to unearth all the documents. I can be pointed at.
Sarah: Yeah, I do think that that a quick br a broad research of bring me some bring me the a list of things in this area, it it can be very easy. I and I think I've found some things that I've really struggled to find through search. I have found
Anne: Yeah.
Sarah: by by looking with an AI that is that is useful. And a completely non tech thing, but I was I've just been Trying to to remind myself of of I I learned Italian quite a long time ago, but not to a very high standard. And I'm like, yeah, let me just get let me just brush up on this. I'm going on holiday to Italy in a bit. And actually an AI is really good at adapting the conversation to be at the right level. So I'm chatting away badly in Italian to to Claude and it's helping me to remember all of this stuff.
Anne: Yeah, that that is astonishing 'cause I have to say, I mean, I know everybody's obsessed with Joolingo, but I never really I didn't really like Joolingo because it kept trying to get me to practice things that I knew I wasn't gonna say and not to practice things that I knew I was want to ha won't have to use almost immediately.
Sarah: Yeah, that was that was what I was trying to do, was just okay, let's practice the things that I'm actually gonna do if I go on holiday. Like let's let's
Anne: Yeah.
Sarah: practice going into a restaurant. Yeah.
Anne: Absolutely, yeah. The things that say when you did your at school, when you learned French at school, it was quite useful stuff like in the restaurant or at the time, at the post office, not quite so much, or at the petrol station, which obviously was of no use to you whatsoever when you were twelve, but it's
Sarah: I remember finding that really bizarre and then years later going, that's why we learned that when we were doing French.
Anne: Fate Le Plan, fill it up. Does that even mean? I don't I haven't got a car, I can't drive, I'm not gonna be able to pass my test
Sarah: Ha ha ha.
Anne: for another five years. But yes, yeah. but yes, Duolingo, I'm sure it's lovely, but yeah, but you've been getting more value out of just chatting to Claude.
Sarah: getting some value definitely and and I do yeah I'm I do acknowledge th the many issues with AI
Anne: Yeah.
Sarah: but it's also a technology that is just slightly mind blowing how good it is given
Anne: Yeah.
Sarah: that it's just predicting the next token it's that's quite that's quite amazing.
Anne: Yeah. it's utterly bl mind blowing what it's doing. and yeah, we have really no idea what what the end result is going to be. But so you you told me a story which possibly turned out to be a as you said yourself. may have been just a myth internally at the FT at the beginning of the of like loads of physical newspapers, real news newspapers were going out of business and everybody knew everything was going to change. and but you didn't necessarily know what it was going to change to. So the priority was to have the ability to make change. I mean you you you you you told me the story of what was it the a Kodak being the
Sarah: Yes.
Anne: the the the example of that Kodak didn't fail because they didn't know what was coming. They knew what was coming. It failed because they couldn't move they couldn't turn the ship around.
Sarah: Yeah, I mean they built the first portable digital camera. There's a photo. It's basically it's basically the size of my head and it looks like
Anne: Yeah.
Sarah: it's made out of bits of Lego or something. It's but it's amazing. The problem is it it's the investment in what you you do. So Kodak built a digital camera. They had photo sharing software, but they saw all of this as being the support to their existing film business. They couldn't change their mind. about what their business was. Their business is selling film. And then they hadn't really understood the rest of it. And I I think I said I got told by my CTO that the board at the FT had said we want to avoid that kind of constrained thinking. We want to make sure that we have the ability to be flexible and to be
able to see where things are going and change quickly. And that's one of the things that I think that the cloud Help with microservices, help with DevOps as well. All of the changes that we went through allowed us to build new things really quite quickly and measure whether they were having the impact that we thought they were having. And I mean,
Anne: Yeah.
Sarah: there's loads of stories about companies that that just are too invested in what they do to be able to understand that what they do is no longer relevant. You know, I think that Blockbuster is the other example. It's like, You have to understand that actually rental of actual physical physical things is is gonna stop.
Anne: Yeah, yes, yeah. It it's interesting because we the reason why we saw one last week is we're both on a panel at a conference in London in September called or October called Fast Flow,
Sarah: Mm-hmm.
Anne: which is about kind of team topologies and be being able to iterate quickly and and put fast time to value and all that kind of stuff.
Sarah: Mm.
Anne: and and I think we talk about fast. as the ability to go fast and put new features out and all that kind of stuff. But really the the the resilience part of fast is the ability to change fast, isn't it? The ability to change, the ability to change at all, really.
Sarah: Yeah, I think it's the it's definitely that and fast feedback loop. So the benefit
Anne: Yeah.
Sarah: of being f of fast is actually the feedback you get pretty quickly. and it's something that I really I think you see when you start to be able to release code more frequently is that you actually sometimes make a change and decide it didn't work and roll it back. Now that never happened when I was working in software organizations where it took us three months to build something. We could build something that was completely the wrong thing, but once it's gone live, it's never being taken out because we've invested too much in that. and I remember the example at the at the FT was that the web team put star ratings onto the page that listed all of the films. And the the hypothesis was we'll get more people reading our film reviews by Teasing the actual star rating and it went live and they discovered that people didn't actually go and read the review. In fact, fewer people went to read the review. They just went, two star, not gonna go see that. Five star, yes, I'll go and see that. So they reversed it. And I just went, okay, that's actually shouldn't be a surprise, but it's surprising that you had a hypothesis, you knew how you were going to measure it, you measured it, it didn't succeed, and you went, right, let's get rid of that code then. and I think that happens because it didn't take forever for that code to go live. It didn't c it cost an enormous amount of of effort. No one had to really put their job on the line to say, here's a feature we should build.
Anne: Yeah. Yeah, which of course you ha you used to have to. I mean, that's in the old days of waterfall development, you you were never gonna go, the last two years, let's you know, like Dallas style. Well let's just wreck on that.
Sarah: yes, forget it, didn't happen.
Anne: Yeah. Which oddly enough, it's a bit that kind of takes you back full circle to the thing you were talking about earlier, where where one of the issues were if A young junior developer is working with an AI can move very, very, very fast indeed. And therefore they do a lot of stuff. And you can't leave them alone too long without kind of checking in because they'll
Sarah: Mm-hmm.
Anne: if they build a whole feature, they become very attached to it, and then it's hard to roll any of it back in their minds emotionally. You can't do it. The emotional rollback is too much.
Sarah: Yes, I think that's absolutely I I I think it's I know I see it in myself. Once I've once I've invested a lot of time in something, much harder to to take the the feedback there. And you know, you've you've written books, so I when I wrote my book, I had an amazing group of technical reviewers. So they're reviewing it for for the content and but by this point you've been writing stuff for for a year, so you really have to be You have to really know that these people are excellent and you really trust them and then they they they ask you a question, you go, no, I now to I need to rewrite this whole chapter and they're right. But but I think you have to try it's hard to keep your mind completely open that actually it'll be much improved and and you need to throw some stuff away and you need to to change your approach to something. The earlier you can get that feedback, the better, because you're not sort of set on it. And
Anne: Yeah.
Sarah: this always makes me think about I reading something about The way people what happens when people find a new house? Which is that you go and look at three different houses and they all have pros and cons. And once you make the decision that this is the house you're gonna put an offer in for, you rationalise all of the things that weren't quite as good. So you might have gone, well, you know what, the bathroom was quite a little bit small, but then once you put an offer and it's like, well, the bathroom is perfectly Perfectly compact and you know, and I just think that that that's what you do right with everything. It's like you make decisions and then you totally rationalise why it was the great s the best decision. So you
Anne: Yeah.
Sarah: need to catch yourself before you have invested your time in in constructing that view.
Anne: Yeah. So so indi so it's interesting that that was an advantage actually in having co-authors on on building
Sarah: Mm.
Anne: green software, that because we had to get every chapter reviewed by all of us as we went,
Sarah: Mm-hmm.
Anne: it kind of meant you never you never got that far ahead. and my husband John he r he read all the chapters as well. I had another another Chap that I often work with, and Charles Humble,
who's often on this podcast, he read quite a few of the chapters as well as we went along because he
Sarah: No.
Anne: was he's used to being my editor, he's been my editor on various things in the past. So, as well as the editor at O'Reilly. But of course, the O'Reilly editor is not technical, so they don't provide you the the the kind of like if someone says, well, you know, you should change your grammar, that there's no emotional problem with that. So you could just say, I'll change my grammar, that's no problem. but if you're technical and someone comes along and go, I think you've misunderstood that, I it's terrible. that's that really is a punch in the face, isn't it?
Sarah: Yeah. I think it worked for me because I I mean so I I just so I had a a great set of tech reviewers like Sam Newman, Daniel
Anne: Yeah.
Sarah: Bryant, and just a bunch of other people. So Hel Patel, Anna Shipman, Nikki Wrights and and all of these people were just giving me fantastic feedback and I really like trust them all. So I know if they're saying something that it's right and it was such a better book as a result of it. But I was not expecting how much rewriting I would end up doing.
Anne: Yeah.
Sarah: I saw I saw a tweet while I was writing the book, because this is before Twitter went rubbish. But basically someone said there are no finished books, there are just exhausted authors. And and I was just like, Yeah, you you you know,
Anne: Uh-huh.
Sarah: do you just hit a point where it's just I can't look at this chapter one more time. Anyway, I realize we've gone completely sideways on we're
Anne: I know. Well you do get to a point
Sarah: not talking about AI anymore, but yeah. I think bringing it back to it. It it's basically you have to find ways to get input from other people before you've really got set in the direction that you're going. And I think AI, it's much more likely that you'll make a lot of progress and then have to be reined back.
Anne: Yeah. you need a whole load of trusted people to burst your bubble at
Sarah: Yeah.
Anne: every point, yeah. And of course AI yeah, because a lot of these things are quite painful. I mean, writing the book was is unbelievably painful. And if AI seemed to do it for you, it'd be very hard. I mean, I d when we wrote both our books, it was the AI wasn't wasn't really an option. And so, know, we weren't tempted. I was talking quite a lot to AI at that point, but it was in no way would I have considered it to be ready. Now I kind of I'm annoyed when I read people's stuff that's clearly been written by AI because it has all the
Sarah: Yeah.
Anne: you know, the uncomfortable truth, all those things, it annoys you when people give you something to read and
Sarah: Yeah.
Anne: it's and they clearly haven't actually written it themselves or, you know, they've maybe just done some tweaks to to edit it, but it but it's okay. it's plausible. when we were writing our O'Reilly books, there was no temptation because it was nowhere near good enough to do that. so yeah. People writing now do have a temptation that they have to get over that we didn't really have to get over.
Sarah: I think comes down to what is the core thing that you are doing?
Anne: Mm.
Sarah: So when I'm using AI for coding, I kind of feel like my core responsibility is to have an outcome in mind. Several like the feature that the the actual functionality that I want, and also the other aspects like the security, the quality, what do I care about about this particular software? And I feel like I'm still doing software engineering, I'm not writing the line the actual syntax. And and that's good because I that's fine for me because my focus is solving a problem using code, not I'm going to be I need to be writing the actual code. And I think with with writing, I could see how like AI helps with with research and it can help with with like reviewing, but but if you have your AI write something, then it's not you.
Anne: No, no, I i it totally isn't. It's it writing is such a helpful thinking tool. You cannot,
Sarah: Yeah.
Anne: you cannot pass it over to somebody else to do. And writing software is not ri is not a thinking tool. It's just a tool. but writing language is a thinking tool.
Sarah: And designing software is a thinking tool because it's how you like when you're designing and architecting it, you're working out what it is that you're trying to do.
Anne: Hmm. Yeah. So yeah, so designers, architects, and ops folk, I think, are going to to be around a l a lot longer as well, you know, thinking about how to scale, whether what scale you need. What do you need from your system? It's so context dependent, isn't it? It's like you could there are all kinds of different things you could do, but what do you need it to do? What does your system need it to do to support your customers, which will be different from everybody else's customers?
Sarah: Yeah, absolutely.
Anne: Yeah. So we are now coming up to an hour and it's been a really good chat. w we should probably wrap it up on the grounds that everybody's brains will be full.
Sarah: Yeah. And also we've really it's been a broad, wide ranging chat.
Anne: Gone all over the place. But I've really enjoyed it. So thank you very much, John. And
Sarah: Me too. Thank you.
Anne: and thank you very much to all our listeners and watchers on YouTube. And I will catch you again on a future episode of Asynchronous and Unreliable. Thank you very much. So thank you for listening to the last episode of Asynchronous Unreliable. But now I have some work for you. We're very keen to answer questions that are submitted by listeners, take suggestions for who we should contact to come on the show next. we're you we're always very happy to do that. And you don't have to record a snippet yourself, you can just send in a message. so you can contact me in if you're on YouTube, you can add things to the comments. If you if you're not On YouTube, if you're listening to this on a podcast, you can reach out to me on LinkedIn. I'm Ann Curry. Very happy to connect to people. Just just reach out and connect, and I'll and I'll connect to you and you can send me your your questions, your ideas, your thoughts. please do let us know. So thank you very much indeed and we are very, very keen to hear from you. Thank you.