Livestream AMA: The Impact of AI on Software Engineering

All right. Hey everybody. Uh, thanks for joining. I um I'm excited about uh what we want to talk about today. Uh we know that AI is at the forefront of everybody's mind. And so something that I talked with uh both Nick and John about was, you know, gathering a group of people where we kind of get multiple different perspectives about how that's influencing the industry and kind of where things are headed. John and I had a lot of, you know, a lot to say about that on a recent podcast on his coding chats uh podcast and uh I wanted to open the floor up also like I said to to newer folks and so hopefully Oh, there's Ebene. He is just joined. Okay, you've been added to the stage. So, we have Ebene and uh William and Shy. They are all newer engineers. you know I I wanted it to be people who are actively full-time employed you know so folks who are uh in their early years so all these folks are I think less than five years in engineering and so I wanted to give all of them a chance to sort of share first what their experience is like and we'll start with them and then we'll go around to uh Nick and John and myself and we'll share a little bit about what our experience and and thoughts are but we wanted to I always Like it's a that's an Amazon thing, right? Jeff Bezos says that the uh the newest person goes first so that they're not feeling like they're following on from the more experienced folks. So meanwhile, uh this is very much meant to be an AMA and so AI is of course the topic dour and so folks if you want to start adding comments in the uh in the chat uh on LinkedIn then we'll get to those comments and we'll take those as we as they come. So without further ado uh I will how about let's see shrudy are you comfortable starting? >> Sure. >> Okay. So, so thoughts. So, starting off first question is just sort of uh set some ground level expect ground uh just content around you know what are what is your anecdotal experience you know so far with with AI and your thoughts about uh what everybody's talking about all the all the hype and and where the industry is headed. >> Sure. Um I don't know if I have an anecdotal experience. I can talk about my general experience. >> Yeah, sure. >> With AI. Um and my like view of how I use AI. Um >> yeah. >> So for as you said, I have about like three years of experience. Um and definitely for me, I use AI more as like a rubber duck or um something I go to to kind of like course correct or look at something that I'm not really sure about, right? Um to double check my work. I think I'm very conservative in using AI in in the sense that I don't want to become dependent on it and make it write most of my code. Um I think I'm like cautious and and more conservative in that sense. Um I don't think I'm like the best user of AI. I'm sure that there are people that use AI more effectively than I have for sure. Um I've played around with a bunch of different uh tools like cursor chat GBT. you know, I think GBT I use more for my personal um like make me itinerary for uh I don't know like if I'm going on a trip um but cursor or like um claude I'll use more for um coding and stuff like that. So I would say my view is more so I want to um focus on my coding skills um and then use AI more as a um a helper tool. Yeah, absolutely. No, that's that's a great uh I think that's a great outlook. So, I appreciate that. So, uh uh MMA, uh why don't you jump in and share a little bit about what your experience has been like? You're able to. >> Okay. >> Hello everyone. Can you hear me? Yeah, we can. >> Can you can you guys hear me? >> Yes. >> Yeah, we can hear you. >> My background is a bit, you know, noisy. So that's why I'm just looking for a very good place to be. >> So yeah, I'm a software engineer. Basically, I've worked I've worked more as a front end engineer, you know, integrating various kind of APIs, um, CMS. Basically, it might seem they actually believe in creating content for some other companies too. So, I do that and then also do some basic um integrations with AI. For example, we built a kind of like um online searching tool using um cloud's API. So um in my day-to-day work I use AI to kind of like enhance my performance. Sometimes to be very sincere I feel like h this thing is um kind of like making me feel lazy or I'm kind of like depending on it where I don't know like that's more reason why I'm also like very excited about this opportunity to um join this uh this thing to see how some other people are also like what is the attitude towards AI because I use it basically like um before AI or before the advent of chart GPT I think in 2020 or so when it came out uh we were using Stack Overflow and it wasn't really like that. For example, if you're working on something and then you have a bug, you want to go to Stack Overflow, you have to read through some other people's solutions and then um the iterative process there then you get a solution, you try to understand and then use it. But these days I think I no more use stack overflow literally when I have a problem I try to like um take it to um but I use cloud and v 0ero vz I don't know if um you know that but that's a versile solution that is very good with building AIS and things I mean UI UI user interfaces. So these these days when I have some development challenges and I want to check out for the reason for an error message I go straight to visero. I no more use stack overflow. In fact if I'm to plot the graph the usage of stack overflow has drastically reduced and sometimes when I want to like um if I have a feature I'm trying to work on and I need to read some documentation sometimes I brainstorm with uh AI. So basically I'm more like using it as a senior that I have in my office where I go I want to work on this or what do you think if you to do this how do you do this how are you going to approach this then from the response and then I try to like um add my own impute and then I get something out of it but I won't lie like sometimes I feel like I'm not supposed to do this this is going to make me weak this is going to make me become a weaker engineer these has some kind of internal battles that I fight while I use this thing. But honestly, it has really boosted my productivity by a very great margin because I I do things faster. I have you know complete I complete features faster but then I have that internal battle. So this why I I will be very excited to hear from other engineers and how they approach it. Thank you everyone and I will share more if there's any need for me to share again. But this is pretty much um what we can share. No, that's perfect. Thank you. And, uh, yeah, we definitely want to have people take questions as they come in. So, that's that's for sure that's the idea. Um, okay, great. So, uh, so William, let's hear a little bit about your perspective and experience with with all of it. >> Yeah, sure. So, I mean, where can I started? I started working around like a bit over a year ago full-time, right? A few industries before. >> Awesome. Um, I'd say AI is pretty central in actually everything we do. Specifically in my workflow, I use it pretty much all the time. So, um, whether it's searching internal docs, we have, um, you can use AI for example, searching internal docs. Um, searching your slacks as well. Um, we have I have a series of, um, automations as well. Um, different Slack channels for example to, uh, for AI to kind of like summarize things. um we're going to start meeting notes uh AI sum AI transcription summarize that um all the meeting notes um even software development in general so typically speaking I don't really I don't give much critical code when it comes to the um the LMS but when it comes to internal tooling I where for example we need to just host a web page debug pages I typically just v code it for example um when it comes to implementation ation ideas for example I like to bounce off uh conversations almost with the AI and then I like to go and you know cross check it talk chat it through with some seniors right but it gives me some good highle ideas that I haven't really thought of before and I think for me it's just a huge boost in unblocking me especially when that friction comes with information right so for me I use it as a very faster search tool say at least. >> Yeah, absolutely. No, I think that's a great way to think about it. Um, so I guess let's see in terms of uh you know the the rest of us, myself, John, Nick. Um I was trying to think like Nick, why don't you lead and share a little bit about what your your thoughts are here. >> All right, no pressure. No pressure. Um yeah, so I am an engineering manager at Microsoft. So I kind of give a couple of different thoughts here because I can talk about a sort of use case at work for me. Um and then how even I see my engineers kind of leaning into it and then I I do program outside of work but at work I I do not. Um so I can share a little bit about how I'm using it from a development perspective but um I like I work at Microsoft so co-pilot having summaries and things like that is like just tremendous. Um, we have uh some more advanced agents then that we can kind of like link up with co-pilot to to help effectively like research like and you've probably seen similar tools where it's not just like the chat model but you can ask it like hey I need deep analysis. So we have that kind of thing. So I've I've used that for doing things like post incident reviews just to collect a ton of data um kind of bring that all together. Um, it's I would say that it's not something that I just take the output of that. And this is a theme with like all of my AI usage. It's not just copy paste the output, you know, cover my eyes and hope it's good. But it's like it's it takes you like, you know, 80% of the way there. Uh, obviously I'm just kind of making up a number there, but it's far enough along where I can go edit tack on and it's enough of a time saver that way. Um, engineers on my team are using like C-pilot and their their IDE. So, you know, a lot of us are using Visual Studio. Having copilot there >> is awesome. Um, and kind of getting to the point now where a lot of uh tooling I would say we're moving in the direction of having like chat integration and things like that. So, um, in like it's a pretty common thing I would say especially at larger companies. You need help with something. Hey, like hey other team how does you know what's the status of this you know server uh service issues things like that. So, it's like an on call kind of helping tool. We're moving a lot in that direction, which is super cool. Um because if you've been on call like as a platform team and you're helping other teams, it's like the same types of questions come in and you're like, man, I wish I could just not have that happen and just give people all the information they need. >> Moving in that direction, which is cool. And then just quickly, a note for me in my development because I have been programming for a long time now. Uh, I would say I am using AI more to like vibe code through things or like sort of to plow through a lot of development. And if I'm moving into areas that I'm not familiar with, I will spend more time kind of like trying to learn a little bit more. But there are things like some front-end development, I'm like, I just need to kind of move through this. And I don't have a tremendous interest in in getting into the weeds. So sometimes some UI stuff gets vibe coded, but um like I have some agents running locally on my computer on a different screen right now trying to put some some stuff together uh like uh integration test frameworks for some stuff I'm building. But uh it's a lot of handholding and that's okay for me because I can kind of guide it to to go do what I need. So uh trying to use it a lot. >> Yeah, that's interesting. I I have to pause on that because that's so interesting to me. I can't help it. what when that's one area where so at first I was very reluctant about that sort of vibe coding sort of unsupervised and I found that I interrupt it and need to help it so much that I'm I'm really hesitant about multi but it sounds like you're doing multiple agents is that right I want to make sure I hear that >> yeah and so like depending on the scenario right um so at this moment I have uh like I was changing a bunch of code for like a personal project and I have a need for integration test suites to kind of get migrated over to a new pattern. But you're absolutely right, like when I especially with like an agent swarm, it's kind of like if you've worked with an agent running and you're like, "Oh man, it feels like it's not quite smart enough to do what I need." It's like that but times eight now. So it's not like uh you have like eight really really smart agents. It's kind of like eight agents where you're like h it's kind of rough. So they get going. But what's nice is that uh if I catch it early, like I can kind of pause it and say like absolutely don't do that. Save a memory of that >> and I'll continue on. So a lot of guiding though. >> Yeah, for sure. >> Yeah. Well, I want to go last though. So John, I'd love to hear your perspective on all of this. I know it's interesting to me too because you're in a teaching space and so there's that's really relevant here where people are able to use AI to like something that a lot of the folks have shared right is that they um they're able to use as a rubber duck they're able to use it to help explain things kind of highly contextual I mean to me that's a big advantage or a big unlock that we have you know with AI and then you're trying to get like people have said multiple of our folks folks have said they don't want to become too dependent on it. And so you're more in the in the business of getting helping people to, you know, unlock those skills where you're really getting into the the the depths of understanding things. So I'm curious how it all sort of fits together for you. Um, you know, love to hear your take. So to answer that question, I guess for most of the people I see that are using it and trying to do some of the coding challenges, >> yeah, >> it's stopping them getting that level of understanding. It's meaning they get a superficial solution that at first glance looks like it, you know, passes the test and meets the requirements, >> but they're not getting deeper. They're not getting that understanding. And quite often the code doesn't quite do what it should do. >> Mhm. >> So from that point of view, I think when people use it as a crutch, it is depriving them of the opportunity to learn. >> Now again, you know, we could level the same criticism of tutorials and just following along copying code from a book. If you don't take the time to understand it, you don't take the time to get familiar with the material, you won't learn it from a book or tutorial either. So, it's not that AI is bad for that. It's just that it can make it appear that you've done something and you've generated all this code a lot quicker and a lot, you know, more smoothly than if you had to copy it from a tutorial, copy it from a book. >> So, from that point of view, I'd prefer to be see people using it less when they're learning and focusing more on getting stuck in. >> Yeah. >> Um, the other problem I'm seeing is still an awful lot of people don't know how to start a project from scratch. they don't know how to sit down and think through a project. So again, if they use it as a crutch from that and it generates something, they don't really know then if what it's produced is good or bad. So they're not in a position to assess it and evaluate whether the outcome is good. Um so again, it's it can appear to be I think a lot more productive than it is for some people. I think if you've got more experience, it potentially has a chance to be uh far more powerful because you know what good looks like. You can direct it a bit better. You can be a bit more focused. You can pick out where it's providing value and where it's not. >> Yeah. >> And I think if you don't have that experience, you run the chance of getting something that looks good but is potentially quite bad or dangerous. And you know, I'm sure we've all seen the memes about things that you know, exposing all sorts of >> credentials and bits of security and if you you know, like clicking view source or go into the developer console on a web page and get everybody's secrets. So >> absolutely >> there's definitely a balance to be found it can be very very useful. I think one of the things that disappoints with AI because AI is of personal interest is this focus on LLMs and coding and as we've mentioned there's a bunch of other tools that are really useful like you know summarizing meetings and getting the key points for that that is very very useful because most people suck at meetings and don't record the minutes. So, if you weren't there or, you know, you were on holiday, you were sick or whatever, actually having that record or going back and and substituting that into your um ADRs or whatever record you use, so you at least got an idea of why a decision was made and when it's made is incredibly powerful. >> Yeah. >> Things like this and doing the podcast that you mentioned earlier, Brian. >> Yeah. >> AI is incredibly powerful for editing that and producing, you know, the podcast or producing useful material for tutorials. So there's an awful lot of stuff that AI can be really useful for in our field that isn't just producing code and we seem to kind of well the people pushing AI have glossed over that in that AIM will replace software engineers rather than all these amazing things that it can do to help us become more efficient at our jobs. >> Yeah. And I love Nick's example too about the on call scenario because we have that of course. So I'm a engineering lead at Salesforce and so you know we have that rotation as well and you know the interesting part there to me is you know every this is maybe unpopular because I'm ne I'm not I'm never a proponent of no documentation but sometimes a a heavy documentation culture eventually equates to zero documentation because if the doc if the docs are a thousand pages long then they might as well be zero Right? And so because nobody's going to know where to start, but AI is really good at finding context in that ocean of, you know, potentially relevant stuff that's in the runbook when you're on call and that sort of thing. So, you know, I think that's super powerful and I love Nick's example. So, uh, so I guess, you know, I'm for those who don't know, I'm a I'm an engineering lead at Salesforce. Um, easiest way to explain it is that my product is sort of a competitor to Nex.js. very popular if you're in the front-end world. Um, so I I'm the lead there on a Salesforce commerce specific version of an AWS Lambda runtime. So that's kind of me in my day job. On my on my on the side, I have a a company meet near me, which is a startup that I started and I have a few engineers. So I've had a chance to see up close some of the impact on uh newer engineers of you know with AI because I am actively mentoring and have a number of folks that are on my team. Uh and so you know anyway so all that to say um I started I took the same trajectory as the newer folks like when I started I was kind of early on this and very interested in it probably 18 month 15 months ago I don't know whatever it was you know a while ago and I was very active about um in the in the prehistoric days when we didn't have cursor and I had to actively put you know code snippets into chat GPT and see what it would tell me. And and so it was interesting because I I know even before cursor integration or the IDE level integration was really strong, I was learning Golang that way. Talked a lot about that on John's recent uh podcast on coding chats. And so I I had that kind of slower intro and I've only in the past I would say three to four weeks really been experimenting more with unassisted or un mostly unsupervised letting the agent just kind of go go nuts which is what it wants to do. And you know I find that the speed of comprehension for reading is absolutely critical. similar Nick has kind of started to say this. It's like, you know, I had yesterday it happened where it's like I had specifically quarantined this this this script which was like a subset of the functionality and I said go run this script and once we get this working we're going to fold it into the the main monolith API and it runs the script somehow misses that it gets a 500 back and then it starts to it's it's going to Okay, we're now we're going to rewrite your whole code base. Stop. Stop. Stop. Don't do that. Uh so, you know, anyway, um it I that being said though, it is uh it's super interesting because I I think that there's there's a study I saw recently. I didn't, this is more like I read the headline, but I'm really interested in in essentially people think that they're 50% more productive, but a study found that they were 20% less productive, you know, with AI. And I think that that really I can relate to that because I think a lot of it has to do with that sort of dopamine cycle of like feeling like you got a hit, you know what I mean? Like you're in the casino. It's like, oo, I got this working and I didn't have to understand it. And I think that that's our minds inherently avoid friction, which is exactly what John's trying to impart to people how to, you know, and and like all of Nick, John and I have all been in this 20 plus years. I'm assum I'm 20. I think Nick's probably longer. I know John's 30 or something like that. Nick, how long have you been uh in engineering >> professionally? Uh like 14 years. So >> Okay. Okay. Cool. >> Not not 20. Not for me. >> Oh, yeah. Sorry. Whatever. Well, that's okay. He's gone past me, so it doesn't You're How long doesn't matter. You know what I mean? Um, but that being said, uh, >> well, my my hairline my hairline shows it. So, >> yeah. So, but but anyway, I think like that's that's the big lesson, right? It for me being in this 20 years is that um there's always there's you're always going to spend time on the things you don't think you're going to, if that makes sense, right? Like it's the easy parts that are easy, you know? No, my I always tell people, imagine that all you do all day long is work with things that are broken and they're never fixed because as soon as you fix them, then you're fixing the next thing, right? There's no there's like if you're lucky, there's 30 seconds of celebration there. You know what I mean? And so when everything's always broken and you're always fixing it, you know, that's kind of um you know, that that's an interesting problem, I think. And so when you look at AI and the fact that you can quickly, you know, churn through lots of stuff that is subtly broken, um, you know, that's going to be interesting to see how that plays out with people who are doing that in production. So um, you know, I had a a big a post on LinkedIn that got a bunch of likes about like I had someone presumably vibe code some you or like use some AI and not understand, you know, how an HTTP uh, handler in Go was returning nil. And so like I had this rare scenario where would return 500 and this was you know on a production project as my my side project which had real paying users and so an example of how it impacted right real users and a real business. So um you have to be really careful and it's really tempting to just allow for this large amount of code without really combing through it. Right. I I'm affected by that just like everybody else. So anyway, that's that's kind of my take. So let's go ahead and jump into questions here. Um, so let's see. Uh, looking for questions. Rubber duckings, great use case. Okay, so we've got, uh, Jose, I think Louise. Um he says, "I am a senior software and I think this is from Nick from your uh channel perhaps a senior software engineer with a strong background in C and I'm really interested in integrating it into my career AI into my career. However, I often feel lost about where to start. Do I need to switch to Python to be effective in this field? What should I focus on learning? Is there a path for someone like me who loves being a developer but doesn't want to become a data scientist? Right now I'm studying uh more the core concepts but I want to stay close to development and build practical AI powered applications. So Nick, we'll give that one to you. >> Sure thing. Yep. I put a response in the chat but I think this is kind of worth talking through because I think this is a really common one that I I see a lot and the way that I look at this uh is is actually almost answered in this person's question which was like I'm interested in AI. I'm interested in integrating it, but like I don't think I want to be a data scientist necessarily, right? And and I I don't want to oversimplify this, but I'm about to. So, uh the way I look at this is that if you want to be on the cutting edge of like creating new types of AI systems, right? So like the fact that we have people that are training models like basically building the parts of LLMs and things like that that the rest of us are consuming and using. I think if you want to be at the forefront of that, there's probably a lot more science involved. And I'm not saying that science isn't helpful in other aspects to this, but I think you probably need a lot more science to be pushing that. So for example, people will talk about LLMs having a limitation where they're saying like I don't think that's going to achieve Gen AI. like you probably need something else that can do that. I don't know what that something else is, but I think that if you were spending a lot more time in the science aspect, you'd probably be, you know, uh, a lot more familiar with some of the other systems that might be coming to the forefront or helping build those. On the other side though, like if you're interested in integrating AI into the things that you're building, the answer that I provided was like I think if we zoom out a little and think have like more systems level thinking, it's just like another thing that we can start to integrate. It's like when when should you be using cues? When should you have these other types of things that you might put into your systems? If you're like if I have some type of you know AI technology whether that's like uh like rag or some other system maybe you are building some search functionality and maybe that makes sense to now go integrate rag into your system so that it can effectively look for different types of content um like that would be one use case but I think the way that I look at this is that if you're trying to build AI into the systems that you're that you're using you don't need Python on necessarily. You don't need the science background. I have neither of those things. And I'm, you know, integrating AI into the stuff I'm building. Uh, can those be helpful? Sure. Um, I just don't think that it like is a a necessary requirement. So, you don't have to pivot your whole tech stack to go, oh, if I don't switch to Python, I'll never be able to integrate AI into what I'm building. >> And Go and Go's catching up. I've noticed they're they're behind, but catching up. like is like Python used to be all the tooling was there, but I found just in the 12 months I've been working with a bunch of this stuff that the tooling with Go is is coming up from behind hopefully, which I'm excited about. I know John's excited about that. Um, >> don't don't confuse AI with LLM. So, AI is a massive field and there's a lot more to it. And, you know, you can build a lot of AI technology in any programming language. >> Yeah. you know, we we started off things things like prologue or languages that support AI development decades ago. >> Yeah. >> Uh most of the production AI shipped over about the six years I did was in mat lab or C++. >> Yeah. >> Um or C and a little bit of Python. So there's you can build these solutions in anything again where >> you're in danger of narrowing AI down to just being using LLMs and it's it's a much bigger field. >> Yeah. I know that's that's a really good call out. Um well, so as we talk through that, I think this next question uh I want to jump into. So Stephen said, "It's an interesting question. How how deep down understanding is needed? Do we need devs still understanding assembly to be effective or are we okay with them focusing on the solution more than the process of being uh technically competent?" And I love this question because I I think like I want to actually go around to multiple people on this one and John like I think I'd like you to start because this is this is that hard question where like you're focused on getting people to understand all the building blocks but you know like I posted yesterday on LinkedIn uh you're never going to get down to the point of all right you know you're not going to like figure out how to turn sand into silicon probably but uh you So, how do you manage that? I'm curious, John, because when you're when you're focused on people mastering like what we've identified as core competencies, right? Like core systems, you know, that need that should be understood. It's like how do you go about ruling out certain things as not qualifying when you know that it's impossible for a human to actually learn all of it? You know, I'm interested in your thoughts there. So I generally think you need to go one or two levels below the level that you're dealing with. So >> yeah, >> you know, if if you're say a front-end developer, make sure you understand HTTP and that HTTP is built on TCP and that's built on top of IP. So >> yeah, >> depending where you want to argue, that's two or three levels below. But uh again the classic example I seen in the back of this is I worked with a team a few years ago that were building stuff calling an API right so restful API over HTTPS they could not understand that if they do a million queries some of those queries will fail and the system could be up and working >> because there's a network in the way and they're going over the public internet that is not reliable and fundamentally an unreliable infrastructure that's designed to be resilient on the basis of of you know things changing on the network and uh potential impact by outside forces and if you make millions of HTTP requests some of those will fail doesn't mean the end service you're hitting is down it's just the nature of networks but they hadn't understood that you know the code they were using under HUD it looked like an API to them was making HTTP calls and go over a way and they needed to build a bit of resilience in for that. Now sure they don't need to go and understand or IP they don't need to understand HTTP TCP but they need to have enough of an understanding that hey this API is making calls over HTTP that's running over a networks are fundamentally unreliable. We cannot assume every call succeeds. >> Yeah. >> So >> there there a couple of levels believe where you are so that you can understand the failure modes and so that if something goes wrong you can start looking into that and start debugging it and understanding it. Yeah. >> Not not down to assemble for most people. Most of us aren't going to need to go that low. >> Yeah. Well, I want to I want to give more uh space here for the the folks like Shudy and Will. I'd love to hear from from you two on this one as well in the sense that uh specifically I'm curious, you know, as you approach your own learning journey, right? And there's I'm I'm assuming you're constantly confronted like me for the last 20 years with things that you don't understand. Um you know, how do you narrow down like where you want to focus and then like you know, how do you assess sort of like where you want to dig in next? I think like what's that process been like for and and specific technologies is like great, you know, speak from where you are, right? like you know if that's a specific tool or technology that's totally fine. I'm curious to hear what you guys think. I think just to clarify I guess right so I mean going deeper into what exactly you're talking I don't know like technical concepts or >> I guess I I mean like I'm I'm more thinking about the dayto-day of the the work of engineering and I'm thinking um you know you're I don't know are you more backend are you more front end what language are you in and then you know there are frameworks that sit on top of languages and like where have you found that um how do you go about the process of deciding kind of like where like John would say be curious you know what I mean it's like where where how do you how do you determine where to let your curiosity guide you in terms of sort of like where you want to go branching out beyond what you're currently doing I guess. Oh yeah like I work a lot on back end but at the same time it also involves a lot of like algorithmic development right so a lot of times um again I use I think the contrary I use AI to go deep actually in a lot of cases I'm having a full conversation and trying to go deeper why is this why is this the case why is this the case >> and I just keep asking it questions over and over until we keep getting deeper and obviously you you ask for citations or you just straight like Google it because it makes up some fake content here and there but I use it kind of like a schizophren IC like chat pretty much as keep going deeper and deeper like um one example is optimizing the um the runtime for some critical code right um you put it this way we need it under maybe like under a second and it's currently at 10 seconds so >> and just think of okay here's what are some general ways to improve it and then kind of go deeper why do you think this is the case why why why should we why should we change the code to this and kind of go down deeper another case too is using it for surveys Um, so you can use AI to get a huge breath and then you can use that breath to select some certain concepts. So say I want to create a u a search across different um many many different strings, right? You do something like a try or you could uh you can look up some other cooler structures like BK trees, right? So I I use it to kind of understand like what are the trade-offs here and there and then I go and like look at some sources on Google afterwards. >> Interesting. So a little bit uh if it's not if it's not private your work like when you talk about deciding between data different data structures in that example there what's the customerf facing use case or the impact what what what is it that you're trying to accomplish there? I'm not sure I can't say but um >> okay that's >> you say say at least it's um you can think of it like translating between two different lexographic um like two different languages almost in a sense so it's something you see dayto day and for some reason translation is not as clear as you think >> okay y makes sense cool uh shudy what are your thoughts uh I'm curious how you you know think through this sort of like approach and like what you're going to learn next and and sort of how you you talked a bunch earlier about how you don't want to sort of become weakened or dependent on it, right? You know, so what's that look like in practice? >> Sure, that's a good question. I think this is something I think about um as I am like in the earlier part of my journey as an engineer. Um definitely like depth first or like we talk about like T-shaped engineers um about where I want to find like my niche. Um I definitely haven't found it. I am a fullstack developer. So I my approach to learning is learning by doing. Um I enjoy like getting my hands dirty as they say and um trying on new technologies and seeing what works for me. Um, and I think for me it's learning by need. A a lot of the times the skills that I pick up is like, hey, this ticket requires me to go deeper into this technology or um I'm curious about how this is working if there's something I'm using and I'm like, okay, like how do they do that? Um, how is that working in the back end? So, it's majority of the time it's curiosity um or if it's a need for the ticket I'm working on. um and interest on in certain things like AI if I'm using it. Um and like Will said um is it will sorry um yeah so like I will use um AI and like chat GBT to kind of learn more and go deeper and ask the wise um there is a little bit of like distrust in me in in in the answers they get back. So I always verify. I think that's something that like is because of the hallucinations that I've um seen. Um so I'm like always double-checking trust but verify is like my approach to using it. It's not like I trust a senior engineer if they tell me to try something more than I trust check GBT when it tells me to try something. So I think I prefer if it's a human being telling me like hey like I've tried it it these are the caveats. Um but it is good to use Chajip even if I if I face the caveats I think I'll learn more I think. Um for me it's like failing trying that's where I'm learning. >> Yeah. Yeah. And I think the potential here you know when we think about rubber ducking like you're saying sort of both of you are are sort of highlighting that you know using it as a conversational tool almost a sort of rubber duck um to sort of you know ask a lot of questions. Why this and then why why that in in response and so on. You know I that that I had that experience like I said learning Golang and you know from zero and it was like it worked really well for me. I think that the more the more you're on the paved path, you know, the less it the less it's an opinion and the more it's like, hey, here are the here are the data structures, right? Like, you know, data structures are a tool, right? And like pick one of them, you know, and uh here's what all the tools in the toolbox do. You can ask all the questions about the tools. That's like a well- paved path. And then as soon as you get into like gray areas where it's like, well, do I want like a document database or do I want a SQL database? You know what I mean? Do I want NoSQL or SQL? And it's like, well, you know, you're probably always going to have things that push you toward both, right? You know, and it's deciding on which is the more important factor. So those kinds of things are where you know you it requires some intuition I think because it's like well which is more important to the customer you know is it is it like like how is it reads or writes like what's the use case you know what I mean which is more important just in in that example you know so anyway but um cool okay let's uh I think it was good to camp out on this I wanted to you know get a couple takes so let's keep rolling through uh some other questions. This makes you most of the time is wasted debugging generated code. Yeah, that's happens. Uh fun fact, we built F5. Okay. Hey everyone, I'm looking for Feel free to call them out if other folks have specific questions. I can find them and highlight them. Uh okay, here's one. So web developer with six years of experience here thinking to uh switching into a field that would get less impacted by AI. Do you think that AI had or will have impact on embedded software development? Oh, embedded. Okay, nice. Uh all right. Well, anybody can just jump in here. I think it will have a lot less impact on embedded software development, at least from the code generation point of view and people claiming to replace them because you're far more hardware specific. You're you far more niche and there's a hell of a lot less training data. >> Yeah, >> you there's a lot of embedded developers out there and still number in the millions, but there's many many more millions of web developers out there. So the training data that's available for LLMs to genite code is much bigger and there's far less there's far less common stuff in the embedded things. There's a lot of common patterns in the problem solving but there's a lot less common stuff in the code and the solutions that are being built. >> Yeah. Yeah. Absolutely. Uh if other folks have have something want to jump in feel free. That being said, also it's like I don't want to comment too much on on embedded. I don't know about others. John might be the only one today who I don't know if Nick do you have much embed I don't have any embedded >> just from early in my career from internships but uh yeah like my my comment that I added uh back to this individual was essentially like I think uh I I don't the framing of the question and sorry I'm not like trying to to pick on the question. I don't this comes up a lot to me. So I don't like the framing of the question only because it seems to imply like this is my take when I read this kind of thing is that like every like I'm trying to look for the the one spot in software engineering that's that's going to be safe because all the others are going to be replaced. That's generally how I interpret this kind of thing. And uh I know it's not exactly how it's said, but I I feel like this is the mindset that we're trying to look for a safe haven. And my take on this is like I think that all of software engineering is going to be impacted basically like every single role in the world will be impacted. But I the point that I added back to them was that impacted does not equal replaced. >> And I just want to keep reminding people that because I know that what we see like in media you'll see you know tech executives kind of it seems like the narrative being pushed is like >> we don't need junior developers. Look, all this AI is writing all this code. Like it's just like it's far too much um like I don't know it's just like a big media push I find and it ends up being like if you if you look at how like YouTube headlines work or any artic sorry YouTube titles or article headlines or anything it's very sensationalized because it works. It makes you click it. It makes you watch it. It makes you read it. it makes, you know, it's appealing to the to the right audience because they're getting people to go look at their material. And to me, that's kind of dangerous because it continues to push this narrative that um that I I just I don't think is quite accurate. Do I think that there are parts of software engineering that will become more effective to have AI and LLMs do? Of course. Um, but I personally I think the stuff that it's going to be especially good at is all the stuff I certainly don't want to do. I've already mentioned like I'm having AI right now as we're talking do test suite migration. >> I absolutely don't want to do that. I was happy to be part of the creation of what that framework looks like. I don't want to go migrate all these tests. It needs to get done. I don't want to do it. >> Yep. Yeah. I think the the thing I would I agree and when you talk about people worrying about so a couple things here one um the idea of so I have it on I have it on good authority which I won't disclose how that when people say AI is writing X% of the code what that means very literally is that in cursor there is instrumentation that logs the number of lines of suggested code and how many times you click accept or reject and that's the calculation. So when I hit accept and I delete 70% of what was there ai no longer wrote 70% of that code right. So when so that's a that's a big agenda that um you know I happen to know that that's how these companies are measuring and it's uh you know obviously not quite accurate it's it's marketing headline you know and so that is that's the first thing so then the other part that I would just add to that similar to what Nick was saying I think like also the the The bottleneck will always be human comprehension. Like that's not going away anytime soon, right? You can generate large massive amounts of code that have been poorly reviewed and that will have a direct impact on the fate of your company, right? Like if you let AI generate 100,000 lines of code and you don't take seven days to read that, you know, then you're not going to make it. Uh your company is not going to make it. Um, so anyway, all that's just a long way of saying that I think that it's a very very strange I I didn't anticipate this and in my opinion, science fiction didn't really anticipate very well the idea when they talked about robots and you see C-3PO can speak thousand languages. There was just no time spent on the fact that like the sheer amount of data is is is like incomprehensible. No, no humans can speak a thousand languages. And so your bandwidth as a human being, like what's the most languages anybody speaks? There you go. You know, 10, I don't know, 20. So if there's a thousand languages, which I don't think there are, but that's Star Wars. you know that you're going to run into this problem where if AI can generate, which it can, more output than all of humanity has generated since the dawn of time in 12 months, you're still going to have the same problem, which is essentially like it's now you've just move you've moved the goalpost so that it's all just review now, right? Like humans don't get to write code anymore. Now all we do is sit and review code. And so, you know, and you can't review code effectively if you never wrote code, which is what John's doing, you know, he's trying to teach people anyway. So, it, you know, it it's really it's really interesting because I think that um I wasn't I didn't have this on my bingo card before it happened. the idea that like, oh, we have this new problem, like if you let this thing loose on, you know, your codebase or the world or whatever you want to call it. Um, you now have this massive amount of generated stuff, but like you have no idea. It doesn't mean anything. You know, it's like people I I saw a a startup that was claiming that they're doing fully un unreeded like they're just doing specificationdriven development and that the AI is writing like I forget if it was 300,000 or 3 million lines of code a day and I'm thinking to myself, okay, so is it 99.9% repetition of the exact same code blocks or is it like 99.999% exactly identical? because the functionality needs don't require 3 million lines of code. So, what the hell are you doing? You know what I mean? Like, but anyway, I don't know. That's that's my like long feel free anybody to >> There's a couple of things I like to add on this one a different point of view. >> Um, as you say, I've been doing this for nearly 30 years. I've been writing code for over 40 years and at least once a decade since I've been interested in computers and writing software AI has been or some other technology has been about to replace software engineers. >> Yeah. >> Now at some point that may be true but so far all the hype and each decade that we've had that hasn't happened. Instead a lot of these tools have increased the demand for software engineers. So I'd hypothesize it the other way. You know, in in 2020, 2021 during COVID, we had a big boom in demand for software engineers. Most people couldn't hire enough software engineers to do all the projects they had. >> Yeah. >> So if AI is making people 30% more productive, demand for software engineers went up by about 50% in those years. Why wouldn't we be redeploying all that 30% and the extra software engineers to build all these extra projects that we want that all these extra things we had? We would be leveraging this productivity that 5 years ago we had all these ideas for that had a potential for positive ROI. We'd be building those. So makes me cynical about the claims of the boost in productivity and also cynical about the claims that AI is why we're laying people off. Yeah, >> I think it's more interest rates have changed and finance and >> we'll come the full circle. We'll get out of this economic cycle. >> People will be looking to build and invest again. >> Whatever level of genuine productivity we get from a boost of of AI will be used to make people more productive and then we'll hire more software engineers and the market will grow because there'll be more things we want to do. You know, if AI ever gets close to being truly intelligent, there's going to be a lot of money made in things like curing cancer, and there's going to be a lot of software and hardware needed for that. >> Yeah. >> Somebody's going to be involved in building all that. So, you know, if we get out of this economic cycle and it provides any productivity, we'll still be using it and we'll be using it to grow grow the industry, grow what we can do, and be more productive and create more value. >> Absolutely. And I have a feeling we might actually uh I'm surprised at how little people are talking about this, but we may see hiring pick up I think soon because everybody talks about the AI doom and gloom, but in fact I think a big driver is section 174 for anybody who's not familiar. um you know is hey uh if you spend a million if you make a million dollars as a startup and you spend a million on 10 engineers making $100,000 forgive the numbers and you get to write off 100,000 or 200,000 of that and your tax bill is now hugely increased versus what it was which is what se section 174 did then surprise surprise you know I heard that Microsoft paid billions extra in taxes or something like that. And so that being said, uh that just got reversed. And so I think that that might see some some uptick in uh in hiring. And I have a feeling that nobody's been talking about that for the last two years, but that's falls right in line with the valley that we've kind of been in for the last three years or so, you know. So >> to be fair, there have been a lot of ummemes in the US talking about it and there's been a lot of discussion things like pragmatic engineer covering it. And I'm sure it's had a massive impact as you say it does mean that you are paying tax on a whole bunch of money you spent which for a startup you spend that money to then pay tax on it as well. >> Yeah. >> It massively impacts your cash flow and will be very negative. >> Uh and for us startup plowing overseas again it was even worse. It was more like 15 or 20 years. I can't remember the numbers. Doesn't impact me being in the UK but it's a significant impact on US startups hiring overseas. So that's kind of impacted that. >> Yeah. >> Um and again interest rates as we say between those two things it has a massive impact on how affordable is to hire people, >> how affordable it is to invest in new projects. >> Absolutely. Yeah. Um let's see. Let's Is there any Does anybody see any questions that they want to highlight with We don't have a ton more time, but we might take just one more question real quick before we call it for today. Um, there was one from Jonathan Baron. I'm not sure if uh that one comes through properly in the chat, but >> it should. Yeah, >> one it said, "Wondering what economic or legislative Yeah, there we go. factors you think most significantly impact the cost of building, deploying, or scaling AI systems." And when I was thinking about answering this one, I wasn't sure. Uh like I immediately I jumped to like infrastructure on this >> and I was like that's actually outside of my wheelhouse to know like there's like at Microsoft there are lots of other people scaling out the actual infrastructure and I am sure there is a ton of uh like factors like that especially when you're thinking scale of like across the planet. So like for example, you need to go build a data center. It's not just okay, where is there a warehouse that we can go lease or something. It's like >> you're, you know, clearing land, you need to go think about power now. Like there's, you've seen talk about like companies like buying like nuclear power plants to go like power new data centers and stuff. So there's a whole bunch of stuff with like with respect to infrastructure that honestly I cannot answer because it would be so wildly complex depending on on where you're talking about in the world. So I didn't know how to answer that part. I don't know if other people want to jump in on that. And then the other part was kind of what I felt like was maybe more in our control as software developers like you're integrating AI into the systems you're building. I think the thing that stood out to me the most in terms of scaling or or like you know a legislative factor if if if I can even call it that would be more around like data privacy. Uh like how are you storing data? How is it being transmitted? Uh like you know if I was building something running it on my desktop I almost don't care. I hope no one has access to my own desktop but whatever. You know I don't need to encrypt things at rest or anything like that. But all of a sudden, if like John and I were building a company and we're like, "Hey, like you know, John's across the planet from me and he has different laws and legislation over in the UK versus me in currently in the US." If if we were like, "Hey, like we're going to have a server in both of our countries and like or you know, I'm like, "Hey, no, John, we got to keep it in the US." And John's like, "No, we should keep it in the UK." Like these are decisions that we'd have to go figure out for what that means for other people's data. Um, and even that like I don't know all the details around that, but fortunately like working at Microsoft, I can go consult the right people that would tell me like can't do that or here's how that needs to work. So, uh, it's a pretty complicated thing for sure. >> You have to think of your customers as well because again, if you're doing it B2B and your customers are business that are putting in a different jurisdiction, you may have to think about how you comply with their laws and how they're impacted. >> Yeah. So I've worked in telco several times and this is a great example. Every country has different laws for telco and how you handle their data. So even if you and I were doing this Nick, I'm in the UK, you're in the US, if we've got a a customer that's say in Germany, you then got to think about how you put out that data and keep it within within the borders of Germany for example. So that's a whole big issue to think about. the cost of building, deploying and scaling your AI systems or any data center. Again, hardware not my area, but I do believe that Microsoft and AWS has some great videos on this that I think public. They certainly shared them to organizations I've been working with that take you through the cost and the involvement and the effort that they go to in sourcing a place for data centers. Thinking about how safe it is from natural disasters, access to water for cooling, access to power, access to just transport and logistics because you need to get a lot heavy equipment there. You need to get compute there. You need to get replacement compute there. You need access to people to support it. all the environmental concerns. Um, so I believe that both those videos are shared on on YouTube. So I would go and have a look for those because they do a good job of explaining it all. >> Nice. And this is a little bit this is a little bit of a tangent I suppose but I'll just briefly say that that we talked earlier about embedded you know and that's not something that I'm an expert on but the other the other you know a lot of people are in software but what we're seeing right with Nvidia being the what most valuable company in the world is that there's a need for hardware. In fact, funny story. I, you know, I lost like uh not a lot, you know, a couple hundred dollars investing in Intel cuz I hoped that they would catch up, right? You know, because surely they've been around long enough like they can figure out how to do this and like no, I I'm not confident in that anymore. But that being said, um you know, this whole race where all these companies that are starting up, if you look at Cursor for example, they're losing how much for every every individual that's licensed, right? And so I'm I'm pretty sure that everyone is betting on figuring out the the hardware problem, right? essentially driving the cost of hardware way down the cost to serve for these for these LLMs and tools like cursor that that use LLMs and you know I mean and Microsoft's involved in that with open AAI and like you know that's public knowledge but anyway I think that that is really significant here when you think about like historically you know everything's been about software for so long this is a really big shift to me in the industry where what we're going to see is hardware is going matter more than ever. Like power plants now matter because you have to drive down the the the cost to serve by making the hardware more optimal and meanwhile you're floating the whole thing with more power and and spinning up new power plants and stuff like that. really big change, you know, in the industry because historically, you know, I'm I've done a lot of front end and it's like, oh, if you don't carefully architect your front end, then it won't work on a slow 3G device that's 10 years old, right? You know, so we've had some of that, but now it's way more relevant because of just how things are are shifting in the industry. So I think for me, you know, that's super interesting to think that being a hardware engineer and bringing there's no way Nvidia is going to have the market capture on this forever, right? So bringing those kinds of hardware designs to other companies, you know, power plants, these things that we haven't talked about or I haven't much in my 20 years in the industry are now really going to become relevant in a big way. So that's where I see a lot of opportunity happening. you know, it just in my opinion. So anyway, but yeah, any final thoughts uh because we're a little over time, but uh I think we can leave it there. If anyone has final parting thoughts, we can uh share that. But yeah, I think that's probably a good place to leave it and that's totally fine. So, thank you everybody for for joining and thanks uh Will, Nick, Abime, John, Shrudy. I really appreciate everybody being here and and uh jumping into this conversation because I think it is a massive pivot and uh it's it's really exciting and confusing at the same time and uh you know the job has always been the same I guess but my parting advice to folks who are confused or thinking that the industry is going away it's like you have to find leverage it doesn't matter if you're a software engineer or what other job you have right and so leverage might move around you know where are the advantages is might change but um you know adapting is I think key and uh you know that's why I love what John's doing with coding challenges to help people level up. Maybe you are a hardware designer, maybe John runs a hardware cloud. No, I'm just kidding. But uh but anyway, no, I think that uh it's interesting times ahead. So, thanks everybody and uh don't quit your job just because some CEO told you that you're redundant. All right, we'll talk to you later. >> Cool. Thank you. >> Yeah, thank you everybody.