In this episode, I sat down with Albert Tanure, a Senior Cloud Solution Architect at Microsoft, to talk about what AI is really changing in day-to-day engineering work—and what still matters just as much as ever. We got into the tension between speed and fundamentals: how AI can massively accelerate coding, dashboards, documentation, and learning, while also amplifying weak architecture, poor security, rising costs, and operational risk. Alberto shared how this shows up in the enterprise world, especially around Azure, resilience, disaster recovery, compliance, and the reality that AI still needs experienced humans to provide boundaries, context, and judgment.
We also explored the tradeoffs between senior and junior engineers in the AI era, why “vibe coding” can be both empowering and dangerous, and how teams are thinking about model choice, local models, hardware costs, gateways, and cloud versus on-device AI. Toward the end, the conversation turned philosophical: if AI keeps taking over more of the work, what happens to human purpose, attention, and our connection to the real world? It’s a practical conversation about architecture, tooling, and cost—but also a human one about where all of this is heading.
- Albert Tanure on LinkedIn: https://nl.linkedin.com/in/albert-tanure
- Code FC on YouTube: https://www.youtube.com/@CodeFC
- WALL-E: https://www.pixar.com/wall-e
[00:00:00] Barry Luijbregts:
Sales.
[00:00:07] Albert Tanure:
Sales. Hey. So how are you doing?
[00:00:13] Barry Luijbregts:
Yeah. I'm doing good. Yeah. Working a lot and learning a lot and struggling a lot as well. So pretty nice.
[00:00:24] Albert Tanure:
Yeah. So as we're recording now, maybe for the listener, can you introduce yourself and tell the people what you're working on and what your role is?
[00:00:35] Barry Luijbregts:
Yes, of course. So I am Alberto Tanure. I'm Brazilian, but I'm living in The Netherlands. Work at Microsoft as a senior cloud solution architect. I I work with... Basically, now focus on infrastructure, but my background came from the development world. I've worked with AI as well. I'm working with AI all the day. And then, yeah, I'm helping customers to not just using AI as the best way, but leverage the best resource we have at Azure, architecture design and approach, disaster recovery, many of stuff related to this high complex environment.
This is basically what I'm working on mostly.
[00:01:29] Albert Tanure:
Yeah. And and and your little AI friends are helping you with that work?
[00:01:35] Barry Luijbregts:
Yes. Of course. I believe, for me, the best the best users of AI on my daily basis, it's when I have to write down emails and then I just use AI as a cooperator to check my grammar or to check things and help me to make it fast. Of course, there's a ton of different ways to use it, but in my daily basis, emails, meetings, it helped me a lot to get context.
[00:02:03] Albert Tanure:
Yeah, get context and also just do the boring stuff, the simple things. Yes, yes. I believe
[00:02:10] Barry Luijbregts:
this is one of the great things about AI. Let's say I have data, I did it the other day, I have a lot of data and I want to combine it, not just in a dashboard, in a way that I can generate information. And it's very easy to do it, not do it by hand. So AI helps a lot with stuff.
[00:02:31] Albert Tanure:
Yeah, exactly. Do you even remember when we had to do all of that stuff by hand? Like, hey, I need a PowerPoint deck or something for this customer thingy. Let's make it all by hands. That's like a couple of years ago.
[00:02:48] Barry Luijbregts:
Yeah. Believe two years ago, I I would like to say, because I remember when I joined Microsoft, it was two years ago. We were using AI, but it's not like today. Right? Before I built a presentation for many stuffs, even for the talks that I was involved in the event. But nowadays, yeah, you can just ask AI, you can bring context, you can just share what do you want. And it's very easy to generate a presentation, documentation, whatever. You... Yeah, two years now, it's getting better. It's more advanced than before.
We can make videos now. So it's crazy.
[00:03:36] Albert Tanure:
Yeah. Absolutely. It's crazy. So basically every conversation I have about technology starts with AI nowadays. Exactly. Which is kind of strange, right? So you mentioned a bunch of other things as well. Disaster recovery, Azure, scalability, performance, security, all that type of stuff. So does that all still matter as well?
[00:04:03] Barry Luijbregts:
This is a good question, and I believe it still matter. Today, I was analyzing one report of a friend of mine. He was showing me the reports before AI and with AI. And then when you look into the report, you're gonna see, okay, I increased the number of pull requests. I increased the number of bugs. I increased the number of, let's say, comments and pull requests. And if you extract this information correctly, you'll see, okay, why this increase are so big than before AI. It's not just because you are doing more, but you are just accelerate the problems you had before AI.
So this is related. When we talk about Doctor, everyone said, okay, we have AI. AI can dealing with that for us. Like Azure SRE agent can do a lot. But you have to analyze the context that AI cannot predict. You have to look into the many different perspective. Then I'm not saying that AI is not able or capable to do it, but you need to understand what is the disaster, what is the the other stuffs. And then with that, AI, it plays a lot of things, help us, but it's required to understand the architecture design, disaster recovery, multi region, a lot of stuffs.
[00:05:34] Albert Tanure:
Yeah. Yeah. And this is kind of the crux, right? So I see this in software development as well. You see the more senior software developers are really good at doing, let's call it, agentic engineering. They can just steer the agent to do a lot of work in the right direction. And the younger developers that, you know, they kinda grew up with agents or, you know, kinda, although that's hard. It's kind of three years or, or, or only a year that it actually works for coding. But they have less understanding of the basics and architecture, So I, I feel like now, also for software developers, if you're a good architect, then you are also a good agentic engineer because you can kind of draw the bounce and architecture in which the thing goes and colors something in. The same goes for infrastructure as well. So disaster recovery, all those types of things.
If you don't know architecture and what could go wrong and how you would do that in the first place without AI, it's very, very difficult to steer the AI to actually do that, I think. Yeah. Yes. How do you think about that? Like, junior versus senior, know?
[00:06:54] Barry Luijbregts:
I believe it's a combination of things. Like, before AI, we had Stack Overflow. Yeah. I couldn't see that many, many developers wasn't able just to look into the foundations, but it was easy just to Googling and use such of, let's say, Stack Overflow coding and put those coding inside the production coding, and that's it. Nobody was interested in reading books or even to know more about some concepts. Then now we are accelerating. We don't need to know more about language because AI can deal with different kinds of languages. You can just ask, I want to build it on Go, I want to build it in C. It doesn't matter. You don't need to learn anymore. So I believe these younger people are learning by AI, doing things without foundation.
But there are good things as well, because they don't have these, like, blockers in their mind, and they are able to do a lot with AI, asking AI and let AI do such stuffs. In opposite of the the senior guys. For example, I had a lot of difficult when I start to work with AI because I know the concept. Always try to guide AI in a way that I want to build like that should be resilient, should be, let's say, blah, blah, blah. This is very important. But I I was trying to do, let's say, a step by step approach instead of accelerate things and then guide AI in a way that, okay, this is the way that you can go. So it looks like a palace of think. It is important to know this concept, but sometime this block me a lot to accelerate.
But from the opposite, from the general guys, they can do a lot because they can think about crazy stuffs with AI, but they are missing some foundations. So Yeah. We we should figure it out the balance.
[00:09:00] Albert Tanure:
Yeah. It's very, very difficult. And and, you know, my fear is now also as a CTO. I read the news today that there was a thing in PHP, and so people are taking down their websites just out of caution so they can update that it's... You know, this happens every week. We have a big thing now. And so if you have a bit of a... The younger generation, they might not think of these things. Security performance, disaster recovery, privacy, where does your data live, all those types of things. And yes, of course, AI probably can think of those things as well, but you have to guide it. You have to kind of tell it to also take care of performance security and make sure everything is encrypted, blah, blah, blah.
It's very easy to vibe code your way into big problems. And then boom, your company is done. So it's a crazy world. I see that at my company now as well. Non tech people, they just go on and they vibe code all sorts of stuff just for their own productivity. Right? They just tell, I call it the robots, they just tell their robots to make me a dashboard that does this and this and this. And then the robot says, cool. Can you just give me that data? Just plug in this thing and then they do. And then, you know, they have awesome dashboards. And then maybe they share it with somebody and hopefully they click the right button and don't share it with the world.
But that's very easy to do. So my work, which is now a startup, we have all these things where nothing is locked down yet. It's different in your world with enterprises, of course, where you have enterprise limits. Perhaps an S and P is not allowed for everything. You cannot just plug in everything there. Just thinking out loud. It is a very difficult and dangerous world now as well with AI. And we see that every day in the nearest. It's crazy.
[00:11:07] Barry Luijbregts:
I agree. I agree even because, let's say, cost, it's one of the things that is not considering by most of people, right? Because if you pay a subscription for whatever agent you want to use, yeah, you are limited by that one. But as soon as you are in a company like I am at Microsoft, I have a limited agent, okay, why not take care of the cost? Why not look into that approach? Because at the end of the day, if you don't have these AI gradities or limits, you cannot build anything else anymore because it's so expensive if you look at that. But you're right regarding this reflection.
People that... I I believe the greatest things from AI, it's like, we really need software at all to build software? This is my question because as you as you mentioned, okay. I can just build a dashboard based on the data. I don't need a developer to do it for me, which is good. It accelerates a lot. I can solve bugs. I can do whatever. But I'm thinking, okay, I can buy a software that can control my task and give me such of, I don't know, productive way to work. But at the same time, if I spend thirty one hour, two hours, I can build by myself without paying a subscription, of course, burning token, but I have my software to my own, and then I can use it. So it's easy to develop something, but at the same time, it's easy to open the doors for threats, open the doors for compliance issues and security issues. So we have to be very, I don't know, in control about it. But at the same time, we are learning.
Nobody knows how AI is working now for anything because we have responsible AI. We have a lot of topics. We have god rules. But at the same time, we are working. We are learning. And then... Yeah. It is a crazy world, as you said. Like, kids can build. All adults can build. Everyone now is developer, but everyone can open the doors.
[00:13:32] Albert Tanure:
Yeah. And control is a very difficult thing here. Because if for instance, as a company, I would say, alright. I want control, so I'm gonna close down all MGP. You cannot use it anymore. I don't know. If you go to chatgpt.com, I block that. I block all the things. Mhmm. And I just control. People will find a way to do what they want to do. So they'll find backdoors, they'll find something else. They use something on their phone, they start sending company data to their phone or something. Will find a way. So it's better to just give them something, a capability to do what they need to do and try to control it as best as we can.
Because people are clever. And especially with AI, the AI helps them to be clever as well.
[00:14:24] Barry Luijbregts:
It's true. But I like to do such exercises, the same as when you ask me, okay, let's say DDoS recover, it's a good thing to think now. My question for customer basically is what's happened if your server is not working that region anymore? Are you able to share or to serve the customers with your product? Maybe not. And what's happened if you don't have AI nowadays? Are you able to do something manually? Are you able to have an instruction that can give you a way to provide some services or provide some dashboard without the automation of AI?
This is a good question. So we have to keep it sharp and the topics that we are working on.
[00:15:15] Albert Tanure:
Yeah. Better. Well, Maybe we are the last generation of people that actually know something. We're It's all using this like the internet. I cannot live without you. And now if you turn off my AIs, I I don't know. And they still work? Probably. It's amazing.
[00:15:38] Barry Luijbregts:
May... Maybe I can... I I believe you can, but it's gonna take a bit more time than the other people with with AI. Right? Yeah. One of the good things we have in our favor, I would like to say, for example, it's very nice to learn with AI nowadays. It's easy. For example, the other day I was looking to an article, and I did a lot of stuff, of course, to to read and learn. But I did some tests here. I want... I have this article. I want to make a a video about it. In a couple of minutes, I got a a nice video with pictures, with use case, help me to understand this article and learn this stuff, and then it took me ten minutes to learn. So it's crazy. It is. And that's nice. Yeah. So so people can now learn how they want to learn. Yeah. And
[00:16:34] Albert Tanure:
I saw that in ChatGPT, for instance, it has this adaptive interface. And depending on the answer, it just shows you a table or something moving, a video, whatever. It just figures it out by itself. It's just... It's awesome. It's crazy, man. Yeah. It absolutely is. And so, you know, you said costs as well. AI costs money. Definitely. GitHub Copilot definitely costs some money. I know that because you pay API prices for it. Yeah. Yeah. To, for instance, Cloud Code where you get a huge discount in in their pricing.
Lots of enterprises are using GitHub with the planets, of course, because it also fits well within Versus Code and Visual Studio and GitHub itself in there. Do you see enterprises also experimenting with others like ClockCode or Codex?
[00:17:28] Barry Luijbregts:
Yes. Based on my experience outside of Microsoft as well, companies are using cloud most of the time, much more than codecs, GitHub for sure. Because GitHub, at the end of the day, is not just the GitHub compiled, right? It's the entire platform. You have an enterprise organization, lot of stuffs related to the GitHub. And this is, it's well used by many companies nowadays. But at the end of the day, Claude, the model one, it's preferable by customers even outside of Microsoft because of the powerful of the model. So you can just host this model somewhere.
You can name it Azure, AWS, whatever, and use the potential of this model and have a better control about it. But for sure, most of friends of mine in Brazil, they are using a lot of Cloud, Cloud Code, but also a combination of Cloud and GitHub many times.
[00:18:43] Albert Tanure:
Also what your preference is, of course, and what you're used to. I was thinking about this and Code, I think, was launched somewhere early last year only. It's so new.
[00:18:58] Barry Luijbregts:
It's so new. But but for me, do do we know, like, it depends. You you mentioned about the ChatGPT where have these adaptive, I don't know, UI based on the answers, blah blah blah. I was thinking a couple of months ago where we have this bunch of new model. Basically, we are talking about GPT, COL, COLD, and Gemini, let's say the three biggest one. But you have a lot of different models. Right? And I was thinking, for example, discussing with a developer in Brazil, he was asking me, okay, what is the best model to use for coding?
And said, depends. It depends on the money. It depends on on the many stuff. But most of the time, it is not required to use Opus from Cloz or Astra from GPT because it's so costly. It's... It can do a lot. But do you really need this complexity? Do you really need to use this powerful frontier model for everything? But because of Claudine's offering Opus or Fable or whatever, most of guys and companies that I'm talking to them in Brazil, they are using the highest model. They are paying a lot because they don't have this, let's say, understanding about the usage of the best model for the same situation.
So it's difficult to have the discussion nowadays because everything new came up and people wants to use it without testing, without anything. Course is not an option in that case. But we have a lot of models, even open source model that we can use to do a lot. But the most famous always in the front of our computers.
[00:20:56] Albert Tanure:
Yeah. And honestly now... So I'm using flawed code now, and I don't wanna think about the model. I just wanna use... I use default Fable five one, and then it codes in sub threads of Opus five five. And, yes, that is probably expensive, but with Clotchef, like these subscriptions like Clotnex, for instance, you pay $200 a month and you get like €8,000 of token value out of that. It comes at a huge discount. Kinda like the GitHub Copilot PRU days almost. So with that, I don't really need to think about, do I use Opus or whatever? It just works, basically.
[00:21:41] Barry Luijbregts:
Yeah. It's true. It's true. It's a good one. But even because you said, if you have a max, like 20, I don't know, max, you can use a lot. You can do a lot. And now if if you know exactly what you are doing, it's good. It's it's not an issue. Right? But let's say if you look into the companies now, I was studying like gateways and Microsoft has a search of gateway, which based on your input, he can decide what is the best model that you can use. It's just as an example, right? But most of companies, when they start using AI, because of cost, because of security, because of compliance, they start to think about, okay, should we use the best one? Should we have a gateway for such a situation?
As example, my employee wants to build a dashboard based on a CSV file. Do they really need to use Opus? No. So if this request came from our employee, from the administrative part, let's bring GPT four point just as an example. Right? But because they are using a lot, they are looking to that perspective, mainly because of the cost. Yeah. You're right. If you know what you want to build, Fable is amazing.
[00:23:05] Albert Tanure:
It is. Yeah. And so you mentioned local models or open source models, open weight models as well. Mhmm. That's also very cool. I recently got a new laptop, which we're recording this on. Thing has a GPU, a GeForce RTX, and 12 gigabytes of RAM for a custom GPU. So I can also run local models kinda, and they're not super slow, so that's... Yeah. It's nice. So I kinda experiment with these things, Quinn and a couple others. There's there's a bunch of them that I can get through Olamo. Yeah. And that works. Right? I can block them into Versus Code, for instance, and they just work.
And I saw recently, Microsoft also announced new laptops where... Like the Surface... Is it the Surface line, I think? Yeah. And they have this huge amount of also shared RAM that also can be shared across the the chip. So now... And then GitHub Copilot then intelligently decides, should I use cloud or should I just use local or what's best for the... For this particular task? What do you think about that?
[00:24:19] Barry Luijbregts:
Yeah. I think it's it's getting expensive to have a laptop. Right? Which fits into this AI area. Because if we look into SurfaceSutra, it's around €33,000, if I'm saying. It's amazing, but at the same time, it's get expensive. Of course, from the technology perspective to decide whatever AI you want to use, or no local, it's expensive. Even you're not trying to use your local LLM in your laptop the whole day. So if you want to do it, you're gonna you're gonna spend a lot of money. Right? For example, I have a mini PC where I've bought, I don't know, one and a half year or last year for a thousand 500.
It's a 128 gigas of run. This RTX GPU, blah blah blah, prepare for AI. It can run models on it as well. But nowadays, it's more than €3,000 to have a machine like that. I have lucky to get that one. But can you imagine you you are running with a laptop like that? Have you considered to use that one? Or if you want to make a cluster of it, you're gonna spend a lot of money. So it's nice. But at this moment, my vision, I couldn't see, okay, why do I need a laptop like that if I have servers, if I have GitHub, whatever.
It's so expensive.
[00:25:52] Albert Tanure:
Yeah. It's not really worth it. No. Especially with the the RAM prices, it's nuts. Yeah. It's scary. Especially compared to, you know, if you get a subscription to something plot or a copilot or something for
[00:26:09] Barry Luijbregts:
It's crazy. I have some examples as friends. We have, like, building a local server for, like, the home lab. But most of the guys are complain, okay. I'm gonna start to use a LEM or any other model, but I need to spend a lot with RAM. Should I go to AI in my local home lab? Blah blah blah. If you have money, go ahead. But I believe it's too expensive. The machine is getting nice, powerful, okay. But if think about the use case, yeah, I don't think it makes sense.
[00:26:49] Albert Tanure:
No. I mostly use it to just transcribe things like an audio to text or something like that too.
[00:26:55] Barry Luijbregts:
Yeah.
[00:26:57] Albert Tanure:
But but it is... It's kind of fun to think about, especially if you think like a hand where also now these things that you can install in our local models, they are becoming smaller and simpler, where they need just less resources. And that's obviously very handy when we have all sorts of robots running around. Yeah. LLMs or whatever type of models in there. It's gonna be a very interesting, I don't know, five, ten years or something where we all, I don't know. Are you gonna get a humanoid robots?
[00:27:38] Barry Luijbregts:
I don't know. The the... I can't... I'm not saying that we cannot thinking about it. It's more closed than before, I would like to say. But, yeah, man, to be honest, if I'm thinking very in a simplest way, before in the past, I believe you're gonna remember that area where we have such disks to copy our data from one computer to another. Those were small disks, right? And then we have such a thing that compact the information to fit in that disk or even to split into more than one disk. It's like we have Wizard, Wizzip, whatever, seven zip, lot of stuff. You're talking about the discs and the floaters. These are the CDs. Yes. Yes. But, you know, I believe our challenge now for me, of course, the robot is coming, but the most important challenge now is how can you fit this big amount of data that is generated nowadays in our servers, data centers, whatever, because you want to make it performatic.
Want to make it manageable. But we need an infinity space to rest this data at all. How can we make it more performatic or more compact in a way that we can leverage it better? So as such optimization, I believe DeepSeek has done something about AI models where it's good. But for me, the next step before the robot will be this challenge about processing things, the memory and the data. I believe this is the most important now.
[00:29:36] Albert Tanure:
Yeah. Yeah. We'll probably need some more data centers as well.
[00:29:41] Barry Luijbregts:
Space? I don't know.
[00:29:43] Albert Tanure:
For the fact. Yeah. You think that's gonna happen?
[00:29:48] Barry Luijbregts:
I believe it must happen. Because if you look into our, let's say, the the resource we have by countries, of course, we have a lot of, it's not just have a, let's say, an area that we can build a data center. There is a lot of things around it, right? But even from computer, we need the memories. Memories are so expensive. This is gonna be, let's say, turned to customers like us. If you want to use AI, you have to pay more. But everyone wants to use it. Every company is using it. So we... Like before this AI area, computers were a bit cheaper than nowadays.
Memories was very cheaper. Now everything is expensive because it's required. And now we we must make it more, how do you say, accessible and then manage it more optimized way. And then we can have freedom and to onboard more stuff. Imagine if we onboard a lot of IoT devices. We already have this one, but imagine robots in every human place. It's gonna be crazy, man. You... You... Manage those kind of complexity and memories and data, whatever.
[00:31:10] Albert Tanure:
Yeah. Yeah. Let... Let's see how that goes. I I don't know. I don't know. Yeah. That's gonna be... There's so many crazy problems coming.
[00:31:21] Barry Luijbregts:
Yes. Yes. It's a good one. We're gonna figure it out. New things, basic quantum computing. Most of those stuff is gonna be, like, important and very, very soon. So, yeah, I believe we're gonna move for the the nice area.
[00:31:41] Albert Tanure:
Yeah. Well, let's hope so. So it sounds like you are positive. You're not an AI doer. So what was the latest thing a couple weeks ago? You know, all of these weeks feel like years nowadays in AI land. But I think Didn't in tropics say, there's a Yes. More than... Or less than... Something like 10% chance that AI will kill us all in the future, something like that.
[00:32:09] Barry Luijbregts:
Yeah. We never know. What do say? Yeah, we never know. I don't know. The other day my friend said that we are rate neutral. I said, I don't think so, but I can say that it will not happen because AI have a lot of God rules, but someone is controlling this one. Right? Yeah. Every provider is controlling that one. What's happened if someone just remove one God rule? What's happening with that one? Like Tony Stark did. So we cannot predict that one. I believe AI has a great potential for many stuffs, accelerate many innovations.
Before we have, let's say, one version of a system a year or in two years. Nowadays you have one system per second. It's a lot. Yep. Every time when you open a Visual Studio Code, you have a new update. Every time you have something new. This is great. But there's, I believe, more innovation is coming. My concern... Did did you watch a Disney movie called Wall E? Yeah. Yeah. I really watched it, like, a couple weeks ago. Yes. And then you see everyone was just sit down in front of a TV, not doing anything, just eating and watching.
This is my concern, because if you look into how people is dealing with AI, how people is generating content, I'm I'm not saying everyone is not reading the content anymore. No. But the majority of people is just building things, let AI the power to generate something and just share into the network. We are create a surge of zombies, starting from technology people. And then my my my concern is in the future. Are we able to walk? Are we able to just breathe in nature without AI? This is my concern.
[00:34:16] Albert Tanure:
Yeah, I share that concern. I don't know. It's gotta be something. So if AI powers the robots and they do everything, what will people do for purpose? Think that's already a difficult thing for people. And we are already super disconnected from nature, which we are. We are nature, of course. Like you said, so we have to rediscover our connection to everything. That's gonna be difficult for most people.
[00:34:49] Barry Luijbregts:
Yeah. It's funny because it's a philosophical thing, we are in an area that we are very connected. Very, very connected. I can talk to my mother in Brazil. Yeah. When we never had that before. But at the same time, we all are disconnected.
[00:35:09] Albert Tanure:
Yes, very disconnected, yeah. It's funny how these conversations often end up into philosophical discussions about life and the universe and all that type of Because that is kind of how profound this technology wave also is, how important it is, that it touches all those things very soon and already kind of.
[00:35:35] Barry Luijbregts:
Yeah, but I believe this is the purpose. Why are we building software to, I believe, to provide a better life for people, right? In hospitals, in the banks, in every place. So we are somehow trying to provide the best environment for people, and technologies bring that facilities to us. Like it's easy now to, I don't know, just transfer a bunch of money by an app or even call to your mom. Technology provide that for us. But at the same time, the Ultron isn't there. Yeah. Well,
[00:36:21] Albert Tanure:
let's see. How was Ultron defeated again?
[00:36:24] Barry Luijbregts:
I think so. I don't know who is gonna be the Avengers. I believe the guy who was just turning off the things. Let's Yeah.
[00:36:35] Albert Tanure:
That's what it is. You can still turn it off. So Yeah.
[00:36:39] Barry Luijbregts:
But we're not gonna because we love using it. Yeah. But the question is remaining. Right? Are you able to survive without AI?
[00:36:49] Albert Tanure:
Yeah. I think that's a nice thought to to end it to. Yes. Where can people find more about you?
[00:36:59] Barry Luijbregts:
Well, I have my LinkedIn, Albert Anur. You can find me in there, and we can talk to me and by chat. I always answer or even on my YouTube channel called FC. I will share with you the link later on. I'm planning search of local models and stuffs that I want to share in the the the series. But, yeah, LinkedIn and called FC YouTube channel is the way people can talk to me.
[00:37:31] Albert Tanure:
Excellent. And all those things will be in the show notes for you, dear listener, that you can just click on and connect with Albert. For now, thank you for listening, and until next time.
Sales.
[00:00:07] Albert Tanure:
Sales. Hey. So how are you doing?
[00:00:13] Barry Luijbregts:
Yeah. I'm doing good. Yeah. Working a lot and learning a lot and struggling a lot as well. So pretty nice.
[00:00:24] Albert Tanure:
Yeah. So as we're recording now, maybe for the listener, can you introduce yourself and tell the people what you're working on and what your role is?
[00:00:35] Barry Luijbregts:
Yes, of course. So I am Alberto Tanure. I'm Brazilian, but I'm living in The Netherlands. Work at Microsoft as a senior cloud solution architect. I I work with... Basically, now focus on infrastructure, but my background came from the development world. I've worked with AI as well. I'm working with AI all the day. And then, yeah, I'm helping customers to not just using AI as the best way, but leverage the best resource we have at Azure, architecture design and approach, disaster recovery, many of stuff related to this high complex environment.
This is basically what I'm working on mostly.
[00:01:29] Albert Tanure:
Yeah. And and and your little AI friends are helping you with that work?
[00:01:35] Barry Luijbregts:
Yes. Of course. I believe, for me, the best the best users of AI on my daily basis, it's when I have to write down emails and then I just use AI as a cooperator to check my grammar or to check things and help me to make it fast. Of course, there's a ton of different ways to use it, but in my daily basis, emails, meetings, it helped me a lot to get context.
[00:02:03] Albert Tanure:
Yeah, get context and also just do the boring stuff, the simple things. Yes, yes. I believe
[00:02:10] Barry Luijbregts:
this is one of the great things about AI. Let's say I have data, I did it the other day, I have a lot of data and I want to combine it, not just in a dashboard, in a way that I can generate information. And it's very easy to do it, not do it by hand. So AI helps a lot with stuff.
[00:02:31] Albert Tanure:
Yeah, exactly. Do you even remember when we had to do all of that stuff by hand? Like, hey, I need a PowerPoint deck or something for this customer thingy. Let's make it all by hands. That's like a couple of years ago.
[00:02:48] Barry Luijbregts:
Yeah. Believe two years ago, I I would like to say, because I remember when I joined Microsoft, it was two years ago. We were using AI, but it's not like today. Right? Before I built a presentation for many stuffs, even for the talks that I was involved in the event. But nowadays, yeah, you can just ask AI, you can bring context, you can just share what do you want. And it's very easy to generate a presentation, documentation, whatever. You... Yeah, two years now, it's getting better. It's more advanced than before.
We can make videos now. So it's crazy.
[00:03:36] Albert Tanure:
Yeah. Absolutely. It's crazy. So basically every conversation I have about technology starts with AI nowadays. Exactly. Which is kind of strange, right? So you mentioned a bunch of other things as well. Disaster recovery, Azure, scalability, performance, security, all that type of stuff. So does that all still matter as well?
[00:04:03] Barry Luijbregts:
This is a good question, and I believe it still matter. Today, I was analyzing one report of a friend of mine. He was showing me the reports before AI and with AI. And then when you look into the report, you're gonna see, okay, I increased the number of pull requests. I increased the number of bugs. I increased the number of, let's say, comments and pull requests. And if you extract this information correctly, you'll see, okay, why this increase are so big than before AI. It's not just because you are doing more, but you are just accelerate the problems you had before AI.
So this is related. When we talk about Doctor, everyone said, okay, we have AI. AI can dealing with that for us. Like Azure SRE agent can do a lot. But you have to analyze the context that AI cannot predict. You have to look into the many different perspective. Then I'm not saying that AI is not able or capable to do it, but you need to understand what is the disaster, what is the the other stuffs. And then with that, AI, it plays a lot of things, help us, but it's required to understand the architecture design, disaster recovery, multi region, a lot of stuffs.
[00:05:34] Albert Tanure:
Yeah. Yeah. And this is kind of the crux, right? So I see this in software development as well. You see the more senior software developers are really good at doing, let's call it, agentic engineering. They can just steer the agent to do a lot of work in the right direction. And the younger developers that, you know, they kinda grew up with agents or, you know, kinda, although that's hard. It's kind of three years or, or, or only a year that it actually works for coding. But they have less understanding of the basics and architecture, So I, I feel like now, also for software developers, if you're a good architect, then you are also a good agentic engineer because you can kind of draw the bounce and architecture in which the thing goes and colors something in. The same goes for infrastructure as well. So disaster recovery, all those types of things.
If you don't know architecture and what could go wrong and how you would do that in the first place without AI, it's very, very difficult to steer the AI to actually do that, I think. Yeah. Yes. How do you think about that? Like, junior versus senior, know?
[00:06:54] Barry Luijbregts:
I believe it's a combination of things. Like, before AI, we had Stack Overflow. Yeah. I couldn't see that many, many developers wasn't able just to look into the foundations, but it was easy just to Googling and use such of, let's say, Stack Overflow coding and put those coding inside the production coding, and that's it. Nobody was interested in reading books or even to know more about some concepts. Then now we are accelerating. We don't need to know more about language because AI can deal with different kinds of languages. You can just ask, I want to build it on Go, I want to build it in C. It doesn't matter. You don't need to learn anymore. So I believe these younger people are learning by AI, doing things without foundation.
But there are good things as well, because they don't have these, like, blockers in their mind, and they are able to do a lot with AI, asking AI and let AI do such stuffs. In opposite of the the senior guys. For example, I had a lot of difficult when I start to work with AI because I know the concept. Always try to guide AI in a way that I want to build like that should be resilient, should be, let's say, blah, blah, blah. This is very important. But I I was trying to do, let's say, a step by step approach instead of accelerate things and then guide AI in a way that, okay, this is the way that you can go. So it looks like a palace of think. It is important to know this concept, but sometime this block me a lot to accelerate.
But from the opposite, from the general guys, they can do a lot because they can think about crazy stuffs with AI, but they are missing some foundations. So Yeah. We we should figure it out the balance.
[00:09:00] Albert Tanure:
Yeah. It's very, very difficult. And and, you know, my fear is now also as a CTO. I read the news today that there was a thing in PHP, and so people are taking down their websites just out of caution so they can update that it's... You know, this happens every week. We have a big thing now. And so if you have a bit of a... The younger generation, they might not think of these things. Security performance, disaster recovery, privacy, where does your data live, all those types of things. And yes, of course, AI probably can think of those things as well, but you have to guide it. You have to kind of tell it to also take care of performance security and make sure everything is encrypted, blah, blah, blah.
It's very easy to vibe code your way into big problems. And then boom, your company is done. So it's a crazy world. I see that at my company now as well. Non tech people, they just go on and they vibe code all sorts of stuff just for their own productivity. Right? They just tell, I call it the robots, they just tell their robots to make me a dashboard that does this and this and this. And then the robot says, cool. Can you just give me that data? Just plug in this thing and then they do. And then, you know, they have awesome dashboards. And then maybe they share it with somebody and hopefully they click the right button and don't share it with the world.
But that's very easy to do. So my work, which is now a startup, we have all these things where nothing is locked down yet. It's different in your world with enterprises, of course, where you have enterprise limits. Perhaps an S and P is not allowed for everything. You cannot just plug in everything there. Just thinking out loud. It is a very difficult and dangerous world now as well with AI. And we see that every day in the nearest. It's crazy.
[00:11:07] Barry Luijbregts:
I agree. I agree even because, let's say, cost, it's one of the things that is not considering by most of people, right? Because if you pay a subscription for whatever agent you want to use, yeah, you are limited by that one. But as soon as you are in a company like I am at Microsoft, I have a limited agent, okay, why not take care of the cost? Why not look into that approach? Because at the end of the day, if you don't have these AI gradities or limits, you cannot build anything else anymore because it's so expensive if you look at that. But you're right regarding this reflection.
People that... I I believe the greatest things from AI, it's like, we really need software at all to build software? This is my question because as you as you mentioned, okay. I can just build a dashboard based on the data. I don't need a developer to do it for me, which is good. It accelerates a lot. I can solve bugs. I can do whatever. But I'm thinking, okay, I can buy a software that can control my task and give me such of, I don't know, productive way to work. But at the same time, if I spend thirty one hour, two hours, I can build by myself without paying a subscription, of course, burning token, but I have my software to my own, and then I can use it. So it's easy to develop something, but at the same time, it's easy to open the doors for threats, open the doors for compliance issues and security issues. So we have to be very, I don't know, in control about it. But at the same time, we are learning.
Nobody knows how AI is working now for anything because we have responsible AI. We have a lot of topics. We have god rules. But at the same time, we are working. We are learning. And then... Yeah. It is a crazy world, as you said. Like, kids can build. All adults can build. Everyone now is developer, but everyone can open the doors.
[00:13:32] Albert Tanure:
Yeah. And control is a very difficult thing here. Because if for instance, as a company, I would say, alright. I want control, so I'm gonna close down all MGP. You cannot use it anymore. I don't know. If you go to chatgpt.com, I block that. I block all the things. Mhmm. And I just control. People will find a way to do what they want to do. So they'll find backdoors, they'll find something else. They use something on their phone, they start sending company data to their phone or something. Will find a way. So it's better to just give them something, a capability to do what they need to do and try to control it as best as we can.
Because people are clever. And especially with AI, the AI helps them to be clever as well.
[00:14:24] Barry Luijbregts:
It's true. But I like to do such exercises, the same as when you ask me, okay, let's say DDoS recover, it's a good thing to think now. My question for customer basically is what's happened if your server is not working that region anymore? Are you able to share or to serve the customers with your product? Maybe not. And what's happened if you don't have AI nowadays? Are you able to do something manually? Are you able to have an instruction that can give you a way to provide some services or provide some dashboard without the automation of AI?
This is a good question. So we have to keep it sharp and the topics that we are working on.
[00:15:15] Albert Tanure:
Yeah. Better. Well, Maybe we are the last generation of people that actually know something. We're It's all using this like the internet. I cannot live without you. And now if you turn off my AIs, I I don't know. And they still work? Probably. It's amazing.
[00:15:38] Barry Luijbregts:
May... Maybe I can... I I believe you can, but it's gonna take a bit more time than the other people with with AI. Right? Yeah. One of the good things we have in our favor, I would like to say, for example, it's very nice to learn with AI nowadays. It's easy. For example, the other day I was looking to an article, and I did a lot of stuff, of course, to to read and learn. But I did some tests here. I want... I have this article. I want to make a a video about it. In a couple of minutes, I got a a nice video with pictures, with use case, help me to understand this article and learn this stuff, and then it took me ten minutes to learn. So it's crazy. It is. And that's nice. Yeah. So so people can now learn how they want to learn. Yeah. And
[00:16:34] Albert Tanure:
I saw that in ChatGPT, for instance, it has this adaptive interface. And depending on the answer, it just shows you a table or something moving, a video, whatever. It just figures it out by itself. It's just... It's awesome. It's crazy, man. Yeah. It absolutely is. And so, you know, you said costs as well. AI costs money. Definitely. GitHub Copilot definitely costs some money. I know that because you pay API prices for it. Yeah. Yeah. To, for instance, Cloud Code where you get a huge discount in in their pricing.
Lots of enterprises are using GitHub with the planets, of course, because it also fits well within Versus Code and Visual Studio and GitHub itself in there. Do you see enterprises also experimenting with others like ClockCode or Codex?
[00:17:28] Barry Luijbregts:
Yes. Based on my experience outside of Microsoft as well, companies are using cloud most of the time, much more than codecs, GitHub for sure. Because GitHub, at the end of the day, is not just the GitHub compiled, right? It's the entire platform. You have an enterprise organization, lot of stuffs related to the GitHub. And this is, it's well used by many companies nowadays. But at the end of the day, Claude, the model one, it's preferable by customers even outside of Microsoft because of the powerful of the model. So you can just host this model somewhere.
You can name it Azure, AWS, whatever, and use the potential of this model and have a better control about it. But for sure, most of friends of mine in Brazil, they are using a lot of Cloud, Cloud Code, but also a combination of Cloud and GitHub many times.
[00:18:43] Albert Tanure:
Also what your preference is, of course, and what you're used to. I was thinking about this and Code, I think, was launched somewhere early last year only. It's so new.
[00:18:58] Barry Luijbregts:
It's so new. But but for me, do do we know, like, it depends. You you mentioned about the ChatGPT where have these adaptive, I don't know, UI based on the answers, blah blah blah. I was thinking a couple of months ago where we have this bunch of new model. Basically, we are talking about GPT, COL, COLD, and Gemini, let's say the three biggest one. But you have a lot of different models. Right? And I was thinking, for example, discussing with a developer in Brazil, he was asking me, okay, what is the best model to use for coding?
And said, depends. It depends on the money. It depends on on the many stuff. But most of the time, it is not required to use Opus from Cloz or Astra from GPT because it's so costly. It's... It can do a lot. But do you really need this complexity? Do you really need to use this powerful frontier model for everything? But because of Claudine's offering Opus or Fable or whatever, most of guys and companies that I'm talking to them in Brazil, they are using the highest model. They are paying a lot because they don't have this, let's say, understanding about the usage of the best model for the same situation.
So it's difficult to have the discussion nowadays because everything new came up and people wants to use it without testing, without anything. Course is not an option in that case. But we have a lot of models, even open source model that we can use to do a lot. But the most famous always in the front of our computers.
[00:20:56] Albert Tanure:
Yeah. And honestly now... So I'm using flawed code now, and I don't wanna think about the model. I just wanna use... I use default Fable five one, and then it codes in sub threads of Opus five five. And, yes, that is probably expensive, but with Clotchef, like these subscriptions like Clotnex, for instance, you pay $200 a month and you get like €8,000 of token value out of that. It comes at a huge discount. Kinda like the GitHub Copilot PRU days almost. So with that, I don't really need to think about, do I use Opus or whatever? It just works, basically.
[00:21:41] Barry Luijbregts:
Yeah. It's true. It's true. It's a good one. But even because you said, if you have a max, like 20, I don't know, max, you can use a lot. You can do a lot. And now if if you know exactly what you are doing, it's good. It's it's not an issue. Right? But let's say if you look into the companies now, I was studying like gateways and Microsoft has a search of gateway, which based on your input, he can decide what is the best model that you can use. It's just as an example, right? But most of companies, when they start using AI, because of cost, because of security, because of compliance, they start to think about, okay, should we use the best one? Should we have a gateway for such a situation?
As example, my employee wants to build a dashboard based on a CSV file. Do they really need to use Opus? No. So if this request came from our employee, from the administrative part, let's bring GPT four point just as an example. Right? But because they are using a lot, they are looking to that perspective, mainly because of the cost. Yeah. You're right. If you know what you want to build, Fable is amazing.
[00:23:05] Albert Tanure:
It is. Yeah. And so you mentioned local models or open source models, open weight models as well. Mhmm. That's also very cool. I recently got a new laptop, which we're recording this on. Thing has a GPU, a GeForce RTX, and 12 gigabytes of RAM for a custom GPU. So I can also run local models kinda, and they're not super slow, so that's... Yeah. It's nice. So I kinda experiment with these things, Quinn and a couple others. There's there's a bunch of them that I can get through Olamo. Yeah. And that works. Right? I can block them into Versus Code, for instance, and they just work.
And I saw recently, Microsoft also announced new laptops where... Like the Surface... Is it the Surface line, I think? Yeah. And they have this huge amount of also shared RAM that also can be shared across the the chip. So now... And then GitHub Copilot then intelligently decides, should I use cloud or should I just use local or what's best for the... For this particular task? What do you think about that?
[00:24:19] Barry Luijbregts:
Yeah. I think it's it's getting expensive to have a laptop. Right? Which fits into this AI area. Because if we look into SurfaceSutra, it's around €33,000, if I'm saying. It's amazing, but at the same time, it's get expensive. Of course, from the technology perspective to decide whatever AI you want to use, or no local, it's expensive. Even you're not trying to use your local LLM in your laptop the whole day. So if you want to do it, you're gonna you're gonna spend a lot of money. Right? For example, I have a mini PC where I've bought, I don't know, one and a half year or last year for a thousand 500.
It's a 128 gigas of run. This RTX GPU, blah blah blah, prepare for AI. It can run models on it as well. But nowadays, it's more than €3,000 to have a machine like that. I have lucky to get that one. But can you imagine you you are running with a laptop like that? Have you considered to use that one? Or if you want to make a cluster of it, you're gonna spend a lot of money. So it's nice. But at this moment, my vision, I couldn't see, okay, why do I need a laptop like that if I have servers, if I have GitHub, whatever.
It's so expensive.
[00:25:52] Albert Tanure:
Yeah. It's not really worth it. No. Especially with the the RAM prices, it's nuts. Yeah. It's scary. Especially compared to, you know, if you get a subscription to something plot or a copilot or something for
[00:26:09] Barry Luijbregts:
It's crazy. I have some examples as friends. We have, like, building a local server for, like, the home lab. But most of the guys are complain, okay. I'm gonna start to use a LEM or any other model, but I need to spend a lot with RAM. Should I go to AI in my local home lab? Blah blah blah. If you have money, go ahead. But I believe it's too expensive. The machine is getting nice, powerful, okay. But if think about the use case, yeah, I don't think it makes sense.
[00:26:49] Albert Tanure:
No. I mostly use it to just transcribe things like an audio to text or something like that too.
[00:26:55] Barry Luijbregts:
Yeah.
[00:26:57] Albert Tanure:
But but it is... It's kind of fun to think about, especially if you think like a hand where also now these things that you can install in our local models, they are becoming smaller and simpler, where they need just less resources. And that's obviously very handy when we have all sorts of robots running around. Yeah. LLMs or whatever type of models in there. It's gonna be a very interesting, I don't know, five, ten years or something where we all, I don't know. Are you gonna get a humanoid robots?
[00:27:38] Barry Luijbregts:
I don't know. The the... I can't... I'm not saying that we cannot thinking about it. It's more closed than before, I would like to say. But, yeah, man, to be honest, if I'm thinking very in a simplest way, before in the past, I believe you're gonna remember that area where we have such disks to copy our data from one computer to another. Those were small disks, right? And then we have such a thing that compact the information to fit in that disk or even to split into more than one disk. It's like we have Wizard, Wizzip, whatever, seven zip, lot of stuff. You're talking about the discs and the floaters. These are the CDs. Yes. Yes. But, you know, I believe our challenge now for me, of course, the robot is coming, but the most important challenge now is how can you fit this big amount of data that is generated nowadays in our servers, data centers, whatever, because you want to make it performatic.
Want to make it manageable. But we need an infinity space to rest this data at all. How can we make it more performatic or more compact in a way that we can leverage it better? So as such optimization, I believe DeepSeek has done something about AI models where it's good. But for me, the next step before the robot will be this challenge about processing things, the memory and the data. I believe this is the most important now.
[00:29:36] Albert Tanure:
Yeah. Yeah. We'll probably need some more data centers as well.
[00:29:41] Barry Luijbregts:
Space? I don't know.
[00:29:43] Albert Tanure:
For the fact. Yeah. You think that's gonna happen?
[00:29:48] Barry Luijbregts:
I believe it must happen. Because if you look into our, let's say, the the resource we have by countries, of course, we have a lot of, it's not just have a, let's say, an area that we can build a data center. There is a lot of things around it, right? But even from computer, we need the memories. Memories are so expensive. This is gonna be, let's say, turned to customers like us. If you want to use AI, you have to pay more. But everyone wants to use it. Every company is using it. So we... Like before this AI area, computers were a bit cheaper than nowadays.
Memories was very cheaper. Now everything is expensive because it's required. And now we we must make it more, how do you say, accessible and then manage it more optimized way. And then we can have freedom and to onboard more stuff. Imagine if we onboard a lot of IoT devices. We already have this one, but imagine robots in every human place. It's gonna be crazy, man. You... You... Manage those kind of complexity and memories and data, whatever.
[00:31:10] Albert Tanure:
Yeah. Yeah. Let... Let's see how that goes. I I don't know. I don't know. Yeah. That's gonna be... There's so many crazy problems coming.
[00:31:21] Barry Luijbregts:
Yes. Yes. It's a good one. We're gonna figure it out. New things, basic quantum computing. Most of those stuff is gonna be, like, important and very, very soon. So, yeah, I believe we're gonna move for the the nice area.
[00:31:41] Albert Tanure:
Yeah. Well, let's hope so. So it sounds like you are positive. You're not an AI doer. So what was the latest thing a couple weeks ago? You know, all of these weeks feel like years nowadays in AI land. But I think Didn't in tropics say, there's a Yes. More than... Or less than... Something like 10% chance that AI will kill us all in the future, something like that.
[00:32:09] Barry Luijbregts:
Yeah. We never know. What do say? Yeah, we never know. I don't know. The other day my friend said that we are rate neutral. I said, I don't think so, but I can say that it will not happen because AI have a lot of God rules, but someone is controlling this one. Right? Yeah. Every provider is controlling that one. What's happened if someone just remove one God rule? What's happening with that one? Like Tony Stark did. So we cannot predict that one. I believe AI has a great potential for many stuffs, accelerate many innovations.
Before we have, let's say, one version of a system a year or in two years. Nowadays you have one system per second. It's a lot. Yep. Every time when you open a Visual Studio Code, you have a new update. Every time you have something new. This is great. But there's, I believe, more innovation is coming. My concern... Did did you watch a Disney movie called Wall E? Yeah. Yeah. I really watched it, like, a couple weeks ago. Yes. And then you see everyone was just sit down in front of a TV, not doing anything, just eating and watching.
This is my concern, because if you look into how people is dealing with AI, how people is generating content, I'm I'm not saying everyone is not reading the content anymore. No. But the majority of people is just building things, let AI the power to generate something and just share into the network. We are create a surge of zombies, starting from technology people. And then my my my concern is in the future. Are we able to walk? Are we able to just breathe in nature without AI? This is my concern.
[00:34:16] Albert Tanure:
Yeah, I share that concern. I don't know. It's gotta be something. So if AI powers the robots and they do everything, what will people do for purpose? Think that's already a difficult thing for people. And we are already super disconnected from nature, which we are. We are nature, of course. Like you said, so we have to rediscover our connection to everything. That's gonna be difficult for most people.
[00:34:49] Barry Luijbregts:
Yeah. It's funny because it's a philosophical thing, we are in an area that we are very connected. Very, very connected. I can talk to my mother in Brazil. Yeah. When we never had that before. But at the same time, we all are disconnected.
[00:35:09] Albert Tanure:
Yes, very disconnected, yeah. It's funny how these conversations often end up into philosophical discussions about life and the universe and all that type of Because that is kind of how profound this technology wave also is, how important it is, that it touches all those things very soon and already kind of.
[00:35:35] Barry Luijbregts:
Yeah, but I believe this is the purpose. Why are we building software to, I believe, to provide a better life for people, right? In hospitals, in the banks, in every place. So we are somehow trying to provide the best environment for people, and technologies bring that facilities to us. Like it's easy now to, I don't know, just transfer a bunch of money by an app or even call to your mom. Technology provide that for us. But at the same time, the Ultron isn't there. Yeah. Well,
[00:36:21] Albert Tanure:
let's see. How was Ultron defeated again?
[00:36:24] Barry Luijbregts:
I think so. I don't know who is gonna be the Avengers. I believe the guy who was just turning off the things. Let's Yeah.
[00:36:35] Albert Tanure:
That's what it is. You can still turn it off. So Yeah.
[00:36:39] Barry Luijbregts:
But we're not gonna because we love using it. Yeah. But the question is remaining. Right? Are you able to survive without AI?
[00:36:49] Albert Tanure:
Yeah. I think that's a nice thought to to end it to. Yes. Where can people find more about you?
[00:36:59] Barry Luijbregts:
Well, I have my LinkedIn, Albert Anur. You can find me in there, and we can talk to me and by chat. I always answer or even on my YouTube channel called FC. I will share with you the link later on. I'm planning search of local models and stuffs that I want to share in the the the series. But, yeah, LinkedIn and called FC YouTube channel is the way people can talk to me.
[00:37:31] Albert Tanure:
Excellent. And all those things will be in the show notes for you, dear listener, that you can just click on and connect with Albert. For now, thank you for listening, and until next time.
Alberto Tanure on Azure, AI, and architecture
How AI helps with daily work and communication
Why architecture and disaster recovery still matter
Junior versus senior engineers in the AI era
Security, compliance, and cost in vibe coding
Enterprise model choices and AI tooling tradeoffs
Local models, hardware costs, and home lab limits
Data centers, memory pressure, and the road ahead
AI risk, human purpose, and staying connected
Where to find Alberto online
