[music] Gary. Google manager Gary, how's it going? How is your first time in Brazil?
Is the tech people in Brazil treating you well? How is your experience in Latin America so far? Oh, it's been great and uh thanks for having me.
Yeah, great to do my uh first podcast here. Uh I've actually been I've been to Brazil several times before actually. So, pretty much your first time.
The uh first time I was here was actually like uh right in the right before the uh do probably younger than some of the audience here, but I'm old uh the late 90s uh you know working on tech. So that was when you know tech was first becoming uh you know big around the world. Uh so it was fun to work with like uh young kids here, startups uh uh you know, back in the in the late 90s.
Um and then yeah, I've been back here the last few years. Um uh but yeah, everybody's great here. Um you know, meeting uh the technical communities have been great.
Um you know, work with a lot of uh good folks at Google. Um but just we've had like uh we did a CNCF meetup last year. Uh met a lot of like super super cool and interesting people.
And uh you know, great to see that uh you know, everybody's into you know, the tech community here I think is super strong. really good to know. So let's get to the questions first.
Uh the first question I have here is before we dive into Kubernetes and AI, uh everyone's talking about AI right now, so it's also a topic we'll try to to cover a bit. Uh when you tell your family or non tech friends uh that you are a product manager at Ga Gh at Google at Google Cloud, what do they think you actually do versus what you really do? That's a great question.
And I think uh I think it depends like uh you know my background was a little bit interesting because I've kind of done everything uh you know I was a you know hardcore like programmer uh done some coding worked on like government defense systems or whatever. So um yeah various I think from my family and friends what they think I do. Um uh I think most of them understand like the product manager role um you know sort of being like the leader of a of a of a tech business.
Um so you know many of them many of them sort of get that part. uh for my I guess my mom is the most interesting one. Uh she's always like, "Well, what do you do?
" Uh so like recently I send her uh I've done some videos at like Google Next. Um and I uh so I send those to her and she's like, "Oh, oh, now I get it. " So at least she knows what I kind of work on.
Um and usually it's more by example, right? So we'll say like we build the infrastructure and you know it ran like you know Pokemon Go or right or this is what like you know Marcato Libé like uses to like run their online stuff and she can kind of see the kind of back end and she knows that we deal with like coming up with all those requirements and making sure all that stuff happens together. But in the end, I'm not sure anybody really knows what we do.
But I think they at least get a get a get a good understanding that sort of product manager is the the sort of business combining the business side with the technology side, right? Um to to for outcomes. Okay.
Okay. Cool. Uh many students assume that technation is like 100% about writing code, right?
uh can you share a specific product decision where the right answer didn't come from peer software engineering but there from like understanding user operations and business uh needs. Oh yeah that's a that's a that's an awesome question. Um maybe I answer the question slightly differently but uh I'll answer the question for sure.
I think uh you know one thing to one thing people should always take note of is that there's a reason why sometimes the best coder in the world isn't always like a billionaire and the billionaire isn't always the best coder in the world. Not that becoming a billionaire is the only sign of success but there's a lot of I think you know business things or or understanding your kind of customers right um how you actually like choose to like go to market is extremely important from a from a product perspective so a lot of times engineers will be like this is a great feature we're going to do like xyz and we'll be like no like well why because we can't sell that right [clears throat] like who is your like producer or who is the person like actually like buying this feature you know how are we getting there. So, a lot of times I think you know that the decisions really have to come down to like knowing your your audience or or your consuming audience, right?
Um and on the infrastructure side, you know, your buyer um is kind of like, you know, it right. So, it's not always going to be about, you know, coding these people. It's going to be, you know, did you did you have the right like support terms and stuff, right?
So, a classic agree argument that we always have is how long are we going to support something for? Are you going to support it for 10 years? Are you going to support it for five years?
um what's your sort of support paradigm and these are like important like decisions right that actually affect whether or not uh bunk Brazil is going to buy your software versus like you know some startup right so do you guys actually plan how long you're going to support uh specific technology from the start or it's something that you deal with later how does that work exactly both um you know at Google we're slight I guess uh at Google we we definitely have a minimum of what we have to support which is which is pretty good um Um and then I think as things go along we kind of change that. Uh in my previous roles at at different companies um you know I think we had much longer uh you know support things. The cloud has changed things a bit because I came from the more traditional like software world right where you install stuff on computers and servers and that you know people expected they have to manage it themselves right so in that case people didn't want to be like changing like every two years right like you build a system for a bank I mean the stuff that's been running the banking systems for the world on the main frames has been running since like the 1970s on on the same systems right versus you know the ability to change things in the cloud it makes it a lot easier for us to handle those sort of upgrades on your behalf Um but but that overall like support and life cycle um is still very important.
Okay, cool. Um now for people starting out in the the cloud world in general like modern cloud infrastructure can feel invisible right for most people. So if you had to explain the engine room behind massive events like Black Friday or hightra global apps running Kubernetes, how would you describe how clothing infrastructure actually works under the hood?
Yeah. Um that's a that's a tough one. Um the uh you know I guess on on the sort of you know pure you know technical side right?
Um I guess who is my audience if it's a more technical sort of audience right? Yeah. Yeah, I mean at the end of the day people come in and they say here's what I want my you know you want to when you build your application you want to say like you know I want to scale to you know a million transactions per second you know a million users um whatever it may be right you know behind the scenes um you know what do we have to deal with from an infrastructure perspective it's and if you look at it from the sort of outside in you have to deal with everything from like the networking layer right so can your like load balancers right actually like handle the network traffic and bandwidth right on the sort of response times at Google that's usually not a problem because we have a good like you know overall global network right then we have to make sure that there's actually uh in this day and age the tricky thing is that there's actually capacity um I shouldn't really reveal this on here but there are capacity constraints for uh compute in the world these days uh uh we'll talk about AI a little bit later but like everything from like memory chips right the little all the pieces right to put together all the materials to put together memory have all gone into building billions of GPUs right so we have like you know kind of a memory shortage Right?
So we have to make sure that these systems can all sort of scale up, right? That the compute actually exists. So from you as a cloud user, you just see like a VM or hopefully you don't maybe you don't see anything, right?
Your application just scaled up underneath the covers. There still is actually physical infrastructure. Yes, we have physical servers in physical data centers in physical locations.
Um so we have to, you know, make sure all of that stuff is highly available, highly optimized. uh you know make sure that we have uh that we can properly share that out across all the different customers um who are coming in right we have lots of big customers Marcato Libre Shopify all all the big ones um you know who are running this for like especially for BFCM type events so what most people don't see it's actually a lot of people working on that it's not as simple as people might think right it's a huge team it's a big team I mean you know a lot of this stuff is work that goes on like all the time like during the here, right? You know, making sure that obviously everything is like scalable underneath the covers.
There is some level of uh planning for the big customers, right? Just to make sure that we're going to that we can handle, you know, all their load. Um but yeah, I mean it's it's constant work.
Um and there's a lot of automation, but there's constant optimizations going on, right? Because if you look at it, you know, you take a company like uh we'll just take locally, you know, I'll just you can pick another one, but like you know, America Libé or whatever, right? Hey, you know, if you look at their scale in like, you know, 2000, right?
What would happen on like BFCM compared to what it is now, right? Like it's not even this, it's not even the same, right? They continue to grow and grow and grow, right?
So, we have to continue, you know, optimizing like every every single year, right? Adding more infrastructure, optimizing the compute, how do we make sure that our infrastructure like Kubernetes uses less memory, uses less resources. um how do we make sure we can scale out across multiple regions and data centers right for sort of fall tolerance.
So all of that stuff is like always going on um you know across across the years and um as customers grow we have to continue to grow as well. Yeah, you brought a interest interesting point like it's possible to automate a lot of things these days but your kind of work still demands a lot of manual work manual work right I think I think a lot of um manual planning um not necessarily most of the most of the most of the systems work we have in place for like an automation perspective but we do have to figure out things like the right levels like allocations quotas so there's there's definitely like the inputs into to the automation I would say you know still have some still have some planning um I guess I don't want to say that like you know cloud's not truly elastic but I think it's not always like at at massive scale okay for massive people right we still do have to do like some level of planning um we're trying to avoid that you know over time but things like these crunches um you know definitely definitely make it harder okay cool um okay next question and Google originally created Kubernetes right Kubernetes Kubernetes Kubernetes yeah and donated to the open source community right uh how has GK GKE evolved evolved uh from a container management tool into the backbone of modern platforms and automation and what is the core engineering philosophy behind GKE oh yeah um right so like yeah I mean Kubernetes uh we celebrated the uh I think the 10th anniversary of Kubernetes in uh was it so 2014 2015 uh 2014 for Kubernetes 2015 for uh for GKE. So just over sort of uh you know 10 years u you know I think if we look back when we first started right containers orchestrating containers you know there's VMs right virtual machines were big then you know docker came around and we had this like orchestration of kind of container orchestration I think over the last 10 years that really started to become kind of mainstream right um that it is fundamentally kind of the core infrastructure that big big enterprise sort of deployments kind of use um you know from a you know the other thing that we've had to do is look at how do we take on you know more and more sort of workloads over time right so when Kubernetes first came out it was great for not that it wasn't good at other stuff but like you know stateless web applications right scale up scale down right you know here's my container I can have you know I need one of them now like more load comes in I can easily spin up like 50 of them right and then spin them down right and that's pretty easy when you don't require any state right?
It's just app you know things going into them. Um over time people like hey this is great can I run databases here right or I guess like you know no SQL or uh you know caches right so we had to add you know more and more kind of capabilities for running stateful applications and that was in the kind of core open source side uh on the Google side um you know from a Google kubernetes engine side we p primarily figure out how do we map that best into the sort of Google cloud infrastructure right um so making sure that we have the best the fastest scale up right? Um the best use of our different storage, right?
Uh the highest optimizations for sort of networking. Um and the other thing that we specifically focus on on GKE was, you know, as we talked about in the previous session was how do we make it easier to, you know, operate GKE? So, can we Google take on more of the role of operating Kubernetes for you?
So, we've definitely been doing that. We had a technology version called like GKE autopilot. Um and it it makes Kubernetes seem more serverless.
you kind of just deploy your applications. We generally handle most things for you. Um, you know, so I think that's been what we've tried to do like more automation, day2 operations.
And then, you know, what was it two years ago? Time flies. Then all of a sudden um and you know one smart thing we actually had done on on on GKE a while back was we actually started supporting things like GPUs uh accelerators and even like TPUs uh Google's version of uh you know AI accelerators you know a while back clearly uh that's become a huge thing now as we know with all these LLMs that are out there.
Um, so I think we've pivoted pretty hard in the open source Kubernetes side on making sure that not only can Kubernetes be the orchestrator for general compute and applications, but also the orchestrator for this complex AI infrastructure that we have to put out. Um, and then from a GKE perspective, um, you know, we've been, you know, doubling down on that to to figure out how do we push GKE beyond the bounds of scaling. And and what do that mean?
We have customers that need to run like hundreds of thousands of GPUs um and nodes, right? And so how do you now make Kubernetes, its core infrastructure, be able to manage, you know, 100,000, 200,000 actual nodes. Um and there's like lots of little things that go into it.
I don't know how much we want to go into it, but [snorts] you know, if you think about it, you have in in G the way Kubernetes works is there's this sort of control plane, right? It's a centralized kind of control plane. It knows where all your nodes and applications are.
Um, clearly that's fine. You you have a, you know, 100,000 whatever nodes. You're just hitting a normal kind of API.
All was good. Now all of a sudden we go up to like 250,000 nodes. Now I've got 250,000 things pounding on this API server, right?
And then all these pods pounding on it. So how do we, you know, make sure that we're resilient, you know, at that sort of scale? And so they're pretty pretty interesting, uh, you know, problems to uh to to try to deal with.
I think that's where we've really tried to do from a GKE perspective is become the most bulletproof uh and the best for you know quick scale up and and and mass scale to take on like any kind of workload. Yeah, resilience is pretty much the most important thing in this type of business. Right.
Yeah. How are you going to handle like all the impact all the people that are using your service? Right.
You mentioned autopilot. Can you talk more about that from a client uh perspective? How it works?
Yeah. and how it makes it easier for the client. Yeah, because I think you know in the in the more in our traditional mode of like Kubernetes and most of the managed Kubernetes service there was most customer most cloud providers will run that control plane part right the the brain but then there's a shared responsibility model of where you're you know all where all your workloads run like all the nodes and those kind of you know workloads and you're kind of responsible for those and it's you're still very you're still exposed to the compute part and I think from an autopilot perspective we said Hey, you still want the power of the Kubernetes API um because it can do a lot of different workloads.
Um also I think people like the Kubernetes API because it is portable. So if you didn't like Google's GKE, apparently there's other cloud providers that have Kubernetes that have that have Kubernetes distributions you could move, right? So people wanted to go with the Kubernetes API, but they just didn't want to deal with all that gorp.
So our idea was just expose the API to you. you tell us how much resource you want for your particular workload. Don't worry about the compute underneath the covers and we'll help you know we'll scale that up and down behind the scenes for you.
Um so that was really kind of uh kind of autopilot. We also of course because we control a lot there we're able to build in all the best practices for kind of security. Um so now we have we're responsible or accountable for security of the nodes.
uh like a lot of that underlying maintenance and management um right that you can kind of say like it's not as much of a shared responsibility it's more of like Google responsibility I'm only responsible for getting my workloads kind of deployed okay cool um next question uh everyone is talking about jai models these days but few discuss the massive infrastructure challenges required to train and serve them right yep uh how are GKE and Google cloud adapting to manage heavy workloads to make AI scalable and viable in production. the uh yeah so um you know I think we talked about a couple of these things but uh the like I said the some of the major issues that that we sort of found were was um there was two things there was one the massive scale of the infrastructure um and I think we sort of covered that right so making sure that we can scale up and scale down um to like these massive amounts of accelerators that you need but then you have to get with the AI infrastructure things become a little bit more like challenging uh Um, you know, for example, um, in like super large training workloads, you actually have to use multiple G multiple GPUs or TPUs actually have to work together. And by work together, I literally mean they have to they have to be more or less like physically interconnected from a sort of networking perspective.
Pass work from one to another from one to another or sometimes the work is actually so on on training um, they divide up the work, you know, across them, but they have shared data that has to be passed across them. Okay. Um and then from an inference perspective sometimes the models like I think you know people uh you know don't always pay as much of attention but like when they talk about usually people talk about like the parameters right uh you know 300 billion parameters 450 billion parameters right how long was a window or whatever that the models were trained with underneath the covers what well what does that end up meaning is it means that that thing is going to need more GPU and more memory and so there become physical limits to how much actual uh NV RAM or or or GPU memory is actually on a card and so if if my model is bigger than one card now we have to figure out you know how do we do that so there's underlying technologies like the model serving engines like you know VLM etc can help like shard that model but you have to make sure that when we deploy our nodes and our infrastructure that we've set up all the interconnects correctly right so how have we wired together all the GPUs is and we want to do this and make this like seamless to the end user.
The end user just wants to say, "Hey man, I have a 450 gigabyte, you know, Quen model or Gemma 4 or whatever. I want to use your whatever B200s or your TPUs. I'm going to divide it up on four.
Please provision the infrastructure correctly for this to happen. " And uh there's a lot of little pieces that go together to kind of make that happen. Um so those are the those are the tough things I've been working on.
low-level high-speed networking. There's just there's a like I won't go into all of it, but it becomes uh it was I would say this, it was also very interesting, I think, for a lot of our kind of like PMs and engineers. Um because um there's a lot of like low-level stuff that goes into like GPUs.
Um the uh for old guy like me, I'm like, "Oh, we did some of this technology way back when like high frequency trading and other things like that. " Um but yeah um and and then we've even changed like a lot of things like added new new ways of how do we make how do we make GPUs and accelerators first class citizens in in kind of Kubernetes right they used to be a little bit secondass um now we have this thing called dynam what it's called dynamic resource allocation if you anybody wants to look it up um but how do we allow you to like allocate when you deploy your workload properly like allocate uh GPUs and TPUs or any type of accelerator or networking as kind of first class resources Okay. So, a lot I guess is what we've been doing.
Yeah. [laughter] Something something I'm I'm curious about. Does the GPU shortage actually affect you guys in Google or parts in general?
Uh or is it something you guys have enough optimization to not really care about? Uh it would be everybody cares about it. Um I think you know we have a pretty good you know we we've been doing some pretty good planning for it um in terms of being able to allocate to our you know some of our largest customers.
Um and then some of our we we've have some interesting technologies that we put in place for you know we we find for example that not everybody needs to like train their model like right now right or needs to run their training job like exactly precisely like this moment. Maybe it's okay that it needs to run like you know in the next like seven days. So we've we've added a few things like things like uh flexible scheduling right which allows us to make greater use of these.
So instead of having to like dedicate all the GPUs to a specific customer, this workload, we're like, hey, you say, you know, you're like, hey, I want mine to run in the next seven days. Gary says he wants to run his, you say want here. We can now kind of slot those across.
Okay. So it's basically we're doing almost like managed batch processing. Um, so we have like some optimizations like that, but I would say everybody is uh I mean there's no cloud provider out there that's not not thinking about that.
That's not thinking about that. and uh and it's you know and I think there's uh the interesting constraints are it's not only the physical GPUs and TPUs and the ability to produce those power power is like a massive thing the amount of power that these uh data centers take it's just something crazy right it's crazy right and um yeah the uh and so and it becomes like interesting right because then people will be like well can we get more in this region and we'll be like well it's hard because we can't get power like in that region right in certain areas. So um I guess it it does affect everybody.
I think that we're doing our best to like manage it at kind of a high level. Um and then I think some of the optimizations in place should help for you know the um more people to be able to uh to better access the resources. Okay, cool.
Uh, next question we have here is the cloud native ecosystem contains hundreds of tools and difficult words people that just don't know about, right? Uh, practically speaking, how can a student build real hands-on experience to demonstrate they know what they're talking about without getting overwhelmed by trying to learn everything at the same time? Yeah, I think you know the the I mean the easiest uh the easy advice to give people maybe sometimes the hardest to follow is um but this is what I've you know always done I think over over the years of my sort of like you know career of building stuff was always like um you know come up with some type of problem that you want to solve right um and honestly maybe I mean as a side thing you can almost use like AI to help you brainstorm like a problem that you want to solve right um and then I think you once because it's always easier to constrain yourself a little bit early on, right?
Like so you just say, you know, I want to build uh you know the best uh you know online like curriculum, right? Or I want to build uh uh you know, something for gaming or whatever it may be, right? Just come up with some idea whether it be a true business or a personal or consumer.
Um and then you'll start to look at like, okay, what's the first key thing I want to do on my infrastructure perspective? If you said, "Hey, I want to run this on like, you know, Kubernetes," you say, "Okay, well, we're going to use, you know, Kubernetes for this. Here's my basic workloads.
They're going to be kind of, you know, web- based, maybe a little database. " So, to me, it's always you kind of like, you know, incrementally build out. Um, and then you're going to naturally find problems as you go along.
And as those sort of problems happen, then you figure out like what's the appropriate thing to big build in. Um, my my classic example is if like, you know, how do people how did caching coming along? So in the in the in the biggest days let's say you just wanted to build an app application and you're I build a web application and you're like okay I was storing everything in local JS you know uh you know uh browser local storage but you know it gets it gets erepped you know so you're like oh now I need to store state so now I add a database and you're like okay this is great [sighs] but now performance is kind of struggling a little bit and I have a lot of read you know I'm doing a lot of reads not a lot of writes and performance struggling like oh somebody will say oh you should add caching okay well let's go look at a caching solution.
So maybe I now like add reddis in, right? And so it kind of becomes this like um you know progressive discovery like incremental [clears throat] sort of imple you know integration. I think it's very hard if you go and say like I want to have these you know I learned these 28 buzzwords from the CNCF thing and I want to build a project that sort of does all of those.
It's much easier to start out with a problem and then as you build it out, there's going to be issues that come up and you'll start to bring in, you know, the appropriate pieces in the ecosystem and build things out like actually try to learn and use it when you actually need it, right? Like when you see the need for it. Yeah.
And it for from a hands-on perspective, I think in a general perspective, you know, I would go like CNCF, you know, for example, go through the ecosystem. you know, they did a they did a good job, I think, of categorizing things, right? Like here's your infra stuff, here's your DevOps stuff, here's your logging, right?
Here's your, you know, I think they even have some blockchain, here's these. So, you can kind of learn the categories and maybe just get an idea of like, you know, what those um groupings of technology are so that you know that they're available when you need them. But I wouldn't necessarily get into like, you know, all thousand of those sort of products underneath.
Leave that for when when you need. Okay. Cool.
Cool. Uh, next question we have here is looking back at your career, what was one of like a major mistake or unexpected product outcome that changed how you approach problem solving and technical leadership and for students and professionals who want to go deeper, what are the best ways to learn GKE? Great.
Um, so I I always I have a very good example of uh, you know, I think something that's helped guide my you know, career. So a long long time ago in a galaxy far far away, I was uh I was working for a small company and uh we had this job to uh rebuild a system for uh for this customer. Okay.
And uh and you know so I talked to a few people there and they were like here's what we need the system to kind of do um and here's what it kind of does. I'm like okay great. And then I being like, you know, in my head a hot shot technologist, I went and found like all the latest and greatest like sort of tech of like how I would build this application to be the most like technical appealing application in the world and it was pretty cool at the time.
I mean it was for for the record it was like 1996 or seven something like that. So a long time ago. And then I took that application and by the way as I'm building this application I basically like built it in my basement or you know in the basement small consulting company right and I occasionally talk to whatever then it comes time for hey I'm done let's go try it with the actual users at this company who are going to use the system.
Mind you I had never really tried it with them before. I had only talked to like their leaders. So they go in and they start trying the system and they don't like it.
And I won't get into like the reasons why, but like there was like, you know, weird. There were valid reasons. [snorts] And then I realized to myself, I'm like, I never I'm like, I built this thinking it's like cool tech.
I thought it was like awesome. It looks great to me, but I never actually went and sat with the person who was going to actually use this system. only thinking about tech other tech people, right?
Yeah. And I never thought about what the person those people doing their job was. The people, by the way, were like were data entry clerks, right?
Okay. Um and and and so they liked parts of it, but they were used to their system. And then I also learned a lot about like how people type weird combinations that people do, like how how data entry works, hotkey, like there's a whole bunch of things you learn.
You're like, "Oh, I could have when we fixed most of the problems, right? " But if I had just gone in like at first and literally just sat with one of them for like a day even and watched exactly what they do on a daily basis with their current system. [clears throat] Then I could have thought well okay here's what the improvements I'm going to bring but I'm also going to make it way more user friendly and tight to that.
So to me this all was notion I think people call them like you know uh you know customer empathy uh customer user journeys customer stories um you know user design there's all kinds of you know terminate terms for it but I guess the the expression is like know your users and know know the business know the business know the stuff u because it's going to be super hard to just go and think you're going to build like great tech. Yeah. um without like ever having had somebody try it out or the people who are going to use it try it out.
So I think I've always sort of been you know in that boat. Um and then I've always tried to be carry that along. I've always tried to be a user of the stuff that I build as well.
Um so like with like GKE and Kubernetes, I used to use Kubernetes before I started working for Google. I still use Kubernetes at Google like whatever just build all my stuff on it. And I think you have much more better empathy.
uh it allows you to relate to customers better and uh and and work on the tech. So to me that's the biggest lesson I think you learn in tech is know your users use empathy have empathy for your users um and it'll make you people will like your products a lot better. Yeah, I'm sure they do.
Like there's a massive impact not only for I think that's in general a really important thing, right? Not just for uh this kind of business. Yeah.
Yeah. In general, it's not in general. It's not the it's it's pretty much the the same.
It you could be a product manager for anything or uh yeah, anything, right? You could be building a car, selling something, building a business, uh building your right, building your you can do whatever it is. But always knowing your user is I think the the the important lesson there.
Cool. Uh we're done here with the questions, but I would like to know like you have a specific talk topic you want to talk about or like send a message to our students at Impacta. Yeah, I guess uh you know the message I would send is I think uh super interesting times right now.
um you know uh like like I I look back you know on me I've been around for a while but I I was around uh in the various stages of technology right like mainframes were still around when I first started they're still around obviously uh client server technology and people can look this up if you don't know it I know these are old data terms right was around distributed computing was coming uh the first e-commerce right the first websites you know I came from a world where there wasn't web like I was around there uh the age of like you know middleware and Java, the age of cloud, right? There's all these sort of like cycles and and I thought those were always interesting, right? To be on the cusp at those times, but now like we're at a and and I think maybe when the internet the dawn of the, you know, the internet and e-commerce, that was probably a big, you know, big uh changing point, right?
Turning point. Right now, we're at another big one. So this is like these rare opportunities I think you know that you that you that you that you come across.
They seem to be faster I guess now in in life but like AI is you know AI is here. Um and uh I think you know figuring out how to like uh embrace it um and best leverage it for what you want to do is going to be I think it's going to be some trying times for some people right um but I think how do you you know how do you figure out how to work it into your workflows? I think you don't want to necessarily be dependent on like AI.
Um I mean for me I love it. Um you know but I came from a big systems background right so like I understand kind of the systems and how they all kind of work. So coming into AI it becomes like a nice assistant right.
Yeah. Um for being able to help with that sort of design. So you'll have to find the right balance I think to people right now like learn some core stuff.
uh you know like do build some few things yourself so that you understand like the sort of basic concepts don't worry so much about like learning like code syntax as being like the I don't think you have to be like the best coder and know like every algorithm anymore you should conceptually like you know understand that type of stuff I think AIs are going to be yeah especially these days that AI write really good code right yeah they write really good code I think you want to be the person who knows what code to ask the AI exactly and then how to put that all together. So that to me is like you know this is a great opportunity right to get on that side so you can get some level of like domain knowledge or specialization in whatever it is you kind of want to build maybe towards the user empathy side learn enough about the tech that you know what to ask the AI to do. Um but like it's a it's a great I think the the role of uh the the technologist is is changing.
Um, and I think more people, there's always been a maybe a difference between like product managers and engineers. Not not a difference, but they, you know, people separate those roles out. I think you're probably going to see the people who are most successful moving forward.
And and again, success, just define it however you will are going to be the people who blend both of those. Okay. Um and so to me that's what I would advise like students today like do learn like you know about the product side how to build the product side as well as you know learn the low-level tech and uh don't be afraid to like you know prototype try and fail um and you know the AI is going to do your be your coding assistant but you got to be the person who's the business leader and the sort of technical lead um you know pushing it and I think if you can like you know put all that together those are going to be the the people who uh you know that that you'll do you'll do you'll do well you'll do fine yeah these days it's a lot easier to learn stuff uh if you have the discipline and know how to answer how to ask questions to AI right so that helps a lot with the process right yeah I think um yeah it's just uh yeah I mean you can always work and I think you don't always it's not that you can't and again to me it's not that you don't want to it's not that you can't trust the AI uh the thing that will bring you need to go deeper right yeah you need to go deeper and you need to so now you need to have time for more I guess you know critical thought right the thing can like converse with you and spit ideas back fast but you have to like take a look and be like yeah does this actually make sense like maybe also go to the old school and you know turn off AI mode in Google or the code or whatever and look something up right and reference just so you kind of can go back and forth between them but uh don't blindly trust don't blindly trust um but uh but yeah I mean these are these are exciting times like uh I'm Super excited.
I mean, I I I I I'm getting a lot done these days. Yeah, me too. [laughter] It helps a lot.
I can't say otherwise. Yeah. Yeah.
Exactly. Cool. So, I guess this is it.
