AI & the Future of Software Engineering: From “How” to “What”

Jake (00:02.104)
So, welcome back to the IoT Podcast. I’m Jake Gilbert, head of 5v Tech. And today I’m really glad to be joined by Mr. Samy Nada, Director of Software Engineering at Siemens EDA in the US. And Samy leads engineering teams working on really complex EDA tools used across the semiconductor world. And today our conversation is a follow-on from some that we’ve had privately, all around AI. Now that’s a pretty common topic with podcasts these days, but really we’re going to be having

quite a different conversation, not from the outside and from the hype, but actually really getting into the viewpoint of the user and more specifically the engineer of today. and Samy’s viewpoint from someone that’s actually leading engineering teams through this kind of adoption and really how they can make the most of it. So we’ll be talking about the phases that engineers go through, the challenges, the kind of wave that people ride to really get the most out of this, and where to start trusting. And

We’ll also get into what’s good today, what still needs work, why fundamentals are still really important despite what people might be saying and really what happens to the role of an engineer as these things progress. So, Samy, welcome to the podcast. Great to have you. Just I guess to start with, it’d be great if you could introduce yourself, your role at Siemens at the moment, and what kind of work you and your team are doing.

Samy Nada (01:23.661)
Okay, first thanks for the introduction and thanks for having me. so yeah, as you mentioned, my name is Saminara. I’m director of  engineering at Siemens EDA. so very briefly, what is EDA? EDA stands for electronic design automation, and it’s very briefly the software that all the chip designer, like electronic chip designer, use to build their tools, build all their ICs and chips and devices and all this kind of stuff. So this is a

All the software that is used for that. This process used to be done manually before. I don’t say before, I mean maybe 40 years ago. Right now, those ships are getting so complex, billions of devices, it’s almost impossible to do anything by hand anymore. And then you need tool for that. And that is what the industry I work on. It’s a very niche because you can imagine it’s not like there’s a million designer, million manufacturer out there. So it’s a very niche industry. so that’s

Again, we handle very, very complex workload and deal with demanding customers. So yeah, that’s this is basically kind of like what EDA and my role is I’m director of the software of the RN software engineering. I started as a software engineer, and about five, six years ago, become a first line manager and then transitioned to a director. so the basically we

write software, C and C and Python and take a whole bunch of other languages. And we develop those tools for the all the EDA customer. And yeah, and basically so the team, all the team I you work with or over C basically write those software. and again, as I said, mostly in C and C. yeah, and and it’s required not required, but it’s very highly recommended for all the engineers working on this team to have a little bit of background on electric

on what is electronic, what is design. So my background is computer engineering, which is a mix between software programming, like just programming C and C, but also hardware design. And a majority of the people who work for those for in our industry have a mix of both. I’ve been working in that field for over 25 years. Again, most of it was just like an IC through different roles from junior all the way to to senior.

Samy Nada (03:48.754)
and principal engineer and then again recently transition to management. yeah, that’s kind of briefly introduction what I do, what the team work on. I’ll be happy to go in detail or answer any questions. if more details are needed.

Jake (04:04.087)
So I guess the the good place to start for us after the conversations we’ve had then is at the start. So when is it that, you know, AI and its current form and the kind of tools that people have access to now for software development s went from being maybe a kind of interesting side project or something you’re keeping an eye on to something that you felt engineers really had to engage with day to day in their workflows?

Samy Nada (04:30.351)
Yeah, I think the jump, which is one the jump that I saw, I can I’m gonna talk about my experience and my team’s experience. I think the big jump happened when we started having tools that basically can view your source code and and and just make recommendations based on that. which is kind of like basically the start of what we call the agentic, what is called as the agentic flow. Before agentic,

and again, agentic, very simple. Agentic is just kind of loop. It’s just like you ask a question and then the answer gets feedback. So there’s no not a lot of magic into it. it’s just very simple. Agentic is just you think of it as just a loop. Take the output, put it back in as an input, and just have this thing kind of keep going up until it you so for agentich, you just give it a goal. For like I want this to work. And so it’s gonna keep moving until it gets to work. Try different things and get back to that.

so so I think agentich was a transition. So why? before because before AgenTec, people were just using all the open AI open chat GPT at that time. And they can w I’m here I’m gonna talk in the programming con programming field, like not of course anybody can use AI for anything. But for programming, mostly what people were using it for was just okay, write me that program, or maybe have a snippet of the code and say, spot the bug here. So the

You do this and whatever AI say something, I copy it back. So you people are doing all this manual work. it’s and and then after a while they realize, especially the more experienced people, like yeah, it’s finding stuff, but it’s just the hassle. I might just kind of find it myself. So the transition that I saw was basically the moment that we started having tool like like a cursor, Visual Studio Code, one server, like some of those tools where it can see your code.

Jake (06:16.781)
No.

Samy Nada (06:20.867)
And can start to interactively cut make a mid mo make a modification, test it, give back. This is what’s a transition. This is when everybody, yeah, this is different here. This is something different here. This is acting like kinda like a human that is actually looking, understanding, trying again. So yeah, short answer is agentic and tools like GitHub like GitHub Copilot and Visual Studio Code that mixed with GitHub Copilot or or Cursor or other tools or like cloud code.

codecs and all the tools that is now widely available.

Jake (06:53.463)
I see. And in terms of your team, you know, when that started to be something that you were maybe driving as as part of the workflows, what was the initial reaction? Was it is excitement, was it skepticism, a mix of both? How did that go?

Samy Nada (07:09.193)
It is a mix of both. So so my team, so a bit about our industry. So again, being a niche industry, we are we don’t work in web development, we don’t work in phone apps and stuff like that. it’s very complicated, and and this we are not the only industry is like this. So what I’m gonna say here applies to other industries as well, like financial risk analyst, if g gaming, embedded systems and stuff like this. that is niche, you can count how many companies work on it.

And there’s not a lot of open source code that’s about it there outside that AI can learn from. So there was initially big like people were like, I don’t know if this is even gonna work. But I think more than that, and I’m talking about kind of like the early phases, I think there was a huge AI stigma. this was not just in my team, this was across the whole again organization, and I’m sure so many other organizations. and also so that stigma.

Maybe I’m right now we don’t have that stigma. Actually, it’s reverse stigma now. but what I’ve seen also is different companies and different industries are going through this journey at a different time. So, so some started before us, some will start after us. So, what was the stigma? The stigma was basically if I use AI, I’m cheating. so what I’ve seen, and this is what kind of started the conversation, I was using AI, and I’m kind of like assumed people are using it. but then I was talking to one of the engineers of my team, and they were kind of like,

He he’s using it, but he doesn’t want to say he because he’s afraid that if he says that he’s using it, I’m gonna look at like yeah, he’s cheating. Like you are cheating, all this work is not credible and all that. This is what I started kinda noticing the first kind of sign of the stigma is people are afraid to come out and say, I’m using AI, because the moment you say that, especially to people who don’t use AI, and again I’m talking about the early time. Well, two years ago. Early time is just two years ago. It seems like ages now.

decades now. It’s it’s the the these thing they’re being labeled as sheeting. And if you do anything, fix anything, reply to any emails and anything that makes sense, then it’s AI. So this is the first thing I encountered is a stigma. And that stigma kept people from using it. That so that’s stigma was the first thing. I’m talking about the obstacles or the hurdles that people face. The stigma was one thing. The second thing, which is it was all corporation, where because it’s new, people don’t know what to do.

Samy Nada (09:37.05)
So so like, okay, okay, let’s ask the legals. Let’s check with this. Let’s check with that. And and the legal, they don’t know what is this. They know all the copyright and all this kind of stuff. So they don’t know what to do. So the easy answer is no. And and or they kind of leave it kind of wishy-washy. It’s not very clear what I should do. So this was kind of like the main two reasons that I saw that people want to use it but afraid. and so this is kind of like what I saw. And the first way to kind of overcome that was to kind of

Two things. First, remove the stigma by saying you f feel free to use it. Actually, I I see using it as a skill. Just make sure to mention that you’re using it. And there’s nothing wrong with that. So removing the stigma, I work on removing the stigma. And the other side work with management to kind of see have a clear usage guideline. Actually, this took a lot of time. but this is like if it’s a company that have some resistance to AI, I think this is the two things that you should need to work on is physically make sure there’s no stigma.

like people can use AI freely without being judged, without being looked at like the sheeting or anything like this. But also make sure that you have clear guidelines of what can be used, what models, what tools, what can you use it with. This is very important. And and again, legal and all this all this departments they are not aware. So it’s it’s on people who are in the management to kinda basically try to explain exactly what it is and try to say it in a language that the legal can understand.

Jake (11:03.661)
Yeah.

Samy Nada (11:03.849)
so that you can pull move the ball forward.

Jake (11:06.509)
Yeah. Well it’s interesting you say that around the stigma. you know, in our in our side of the coin, you see it a lot. you see companies these days, right? Like you said, removing that stigma and really championing and actually the other way around, actually the stigma, like you said, is actually on people who are resistant to it, right? Yeah, reverse stigma. and where we still see a lot of that and you hear a lot of that is actually in education, right? There is still

Samy Nada (11:24.478)
I said the reverse thing, but it’s the other one.

Jake (11:35.102)
I think a a lot of educational institutions is are still in that world of if you use AI to help you, you know, with some kind of task, then then you’re cheating, right? And yes, if you do an entire essay just fully off of a one line prompt, then maybe there’s a point. But I think there’s a nuance to it. And the problem is you then have people who grow up go through education getting that stigma and then it takes time for it to be reversed when you actually reach, you know, the professional world. So that’s interesting.

And, you know, you post a lot about AI adoption and and engineering behaviour and you have some really interesting takes. And I guess the question I had is, you know, why has this topic become so important to you personally? Is it a kind of experience you’ve had or or you know, way that you think it’s going?

Samy Nada (12:20.806)
Yes, it’s it’s it’s a mix of both. so when when when I graduated, it was right at the internet, not internet already came out, but I graduated around the time of the dot com bubble. And so so I was just new, so I didn’t know and I felt like I missed that that thing. By the time I was kinda like more experienced, all all the websites were created, everything was created, actually went through the boom and bust and everything, right? And then of course stabilized because internet changes the world.

And and then after that was a cell phone thing, which I was maybe seven, eight years after graduation. I missed this one too. Again, I was still kind of learning, understanding, and I was in a different field, which is EDA. It didn’t basically it didn’t get impacted much by all the apps and and and cell phones and things like that. But so when the eye came and I said, I’m not gonna miss this opportunity. It seems like it’s gonna change the world. And and I had the same feeling about the internet and and the phones, but again, I didn’t have I was not in the right position, I was not in the right technology.

So when AI came, I said, okay, this is looks like a good technology. But again, there’s so many technologies that comes in and might not prove itself. So when I started using it and I saw the potential, I I kind of like, I have to share this with people. I have to share this. it’s it’s again, I didn’t share it right away because it took me some time to get used to it. And I can kind of go this there’s phases when you go through AI, but once I started seeing it, basically

being can be very, very, very useful. This is when I start sharing my experience, whether my experience as somebody’s using it, or my experience as managing teams that use this, how how did I how did I use my position to basically kind of kind of move move things forward, like again, try to remove the stigma, try to talk to management. I wanted to share my experience. So other people who are in a similar position, maybe similar industries or similar roles,

can maybe use some of my exper some what somewhat of what I went through to that help help them in their in their adopting AI. So it’s a mix of both. It’s it’s it’s yeah. I think AI is gonna change the world. Maybe not exactly as everybody thinks. There’s clearly so many things that’s overhyped and and and I know that. It’s the hardest part for anybody in in doing or working with AI right now is kind of like realize the potential, where it can be useful, where it can be a waste of time.

Samy Nada (14:44.37)
I think this is a key thing that and this is what I’m trying to in my LinkedIn posts on other yeah, to kind of try to get people to see it or at least share my how I see it with others and hopefully it resonate with others who going through the same experience, similar experience.

Jake (15:00.929)
Well, that leads nicely, and you mentioned on to the adoption phases, right? And the the kind of journey people go through when they they start and and get there. and we had we had a few conversations about this before, right? That as you mentioned, there’s that initial initial rise, that kind of excitement, maybe a bit of a tail off, and then that kind of stabilization. So you describe that kind of common pattern. Walk us through it.

Samy Nada (15:26.226)
Okay. So so first, of course, you do yeah, you like a lot of people who just getting newer to know AI, the the first start asking basic question and and and it starts answering, right? And then you start like when it comes to work area, you start look at that code, find a bug. You are asking a question you know the answer to, and it finds a bug. yeah, that is cool. of course, also at the same time, AI was progressing so fast.

And then you start kind of yeah, you doing this you again, this is back to the whole kind of chat GPT or open AI or whatever. You just put the thing. And then second phase, you start seeing, you put a a code in and you know there’s a bug in it, and you ask it. And that thing found that bug and found two other bugs that you didn’t even see when you look at the code. So this is where like, because this thing can see stuff I cannot.

So you start kind of like, okay, this seems good. Okay, try to fix it or something. Well, you know the right solution. And it comes up with way better creative solution than you are. And this is where you’re kind of like, there’s something here. And then when you start plugging into the actual code, big code and and and and huge code basis that maybe very few people, or I don’t think anybody knows the whole thing, but some people, everybody knows a little piece. And you start asking a question.

And like how this works and how this doesn’t work. And it comes back with complete like description, like of everything, the sequence of flow. So you start looking at this. Again, you are in the early. You didn’t use it to generate code, or very few, like it was script or two, but for the most part, you’re just reading. And then information that there is no way you want to know. In order for you to do this, you have to go spend two days with a person who wrote the code who might be not part of the company anymore. And and and it so.

This you see this and you think, okay, there is no limit to this. And then, but okay, it’s good in understanding. Can it actually again find bugs? find bugs in big code. So initially you thought it’s only work with small. Now it works with big code. Okay, can can it do instrumentation? Can it just maybe print add a couple of brand statements here and there? And you do it and it just goes to your code, follow the same way you print anything, or if there’s any kind of instrumentation, just does it there. Recommend the tool, recommend so.

Samy Nada (17:51.643)
This is kind of like fantastic. You start and can you write code? you it start writing code, fantastic. Every everything is going well. But because you’re still in this testing mode, evaluation mode, most you are you are guarding it. So you’re asking a question that you kinda know the answer for. You are asking to make changes that you kinda know how to do. so what happened after this is like now, okay, I trust this thing. And this this process could take up to six months.

And after six months, okay, now I trust this tool. basically you’re asking asking a question that you know 90% of the answers don’t know the 10% to give you the 10%, or it can just accelerate your work. Like it’s a work you can do, but it will take you two days, the whole thing takes it like 15 minutes. So now you establish a trust. And once you establish a trust, you can like this AI, and you see a lot of people. I I I can because I went through it, I can see when people are in this phase because this is the extreme excitement.

AI can do everything. We should kind of like, yeah, we shouldn’t be hiring people anymore. We shouldn’t be teaching people programming anymore. And you see this, and this is it. I can do the same work. I don’t need the team. I can two people be good enough. This is when they hit this, I can see. Then after that, which is what I called the dip. So what happened is once you start asking it or ask it to do things that you don’t hundred percent know. So when you knew.

Jake (19:03.841)
Yeah.

Samy Nada (19:20.208)
When you knew the answers, if the AI used to come back to you with answers as not kind of like 100% on, you kind of like you see it and then you correct it. Usually you nod it because you know the answer or you know 80% of the answer, right? You don’t feel you’re doing so. Yeah. So when you do this, it’s kind of like when somebody you’re training somebody or teaching somebody to do something and they make a slight mistake, but you correct them, right? I’ll do this. You don’t see yourself doing this.

So in your mind, okay, this thing is perfect, it can get it from the first time, and this little nudge, you just forget you did it. It just becomes like a second habit that you just do. But when you start asking it areas that you don’t know, or when you see other how other people are using it where they are not, they don’t have your same level of experience, or then the things start to go wild and it can go in the direction that you never this is completely wrong. Or I’ve seen this where I remember we were looking at one issue that either me and the person who was looking at

maybe instead of knowing 80%, we only know like 10%. So we kept asking a question, and I had an imagination of how what is how it works, and that other person has an imagination of how it works. And somehow, somehow, we had ended up using the same models, same tools, ended up with completely different answer. My answer was more aligned with what I thought it is, and his answer was more aligned with what he thought it was. And and this for me was kind of like one of the big wake-up calls where, okay.

Yes, it is good, but it’s not perfect. And and your input, consistent input to it, to nudge it and correct it, correct course, is key to it. A lot of people at that phase, that are what I saw, is because you go from this extreme excitement to this and and you might hit a failure. They might kinda like they were not very careful and actually added ask it to write code, they didn’t verify it very well, and it just kinda blows up in their face or a customer or something.

And and this is where the kind of man, I thought AI is good, it’s all not good. And so it’s just kind of like move on. I’m not gonna use this again. It’s good for a small little script, it’s not again. I call this the dip dip. And the only way to get through it is to kind of keep at it, just persistent, like success or understanding. Just stay persistent, and when you stay persistent, after some time, you will develop a very good understanding of what is limitation. So

Samy Nada (21:41.786)
You know you’ve you’ve seen the capabilities, you’ve seen now the more you do it, you see the limitation. And this is when you kinda once you hit the limitation, this is kinda like for me, I think, yeah, you know what it can do and what is the limitation. This is the start of you the new learning, which I call realization.

Jake (21:59.726)
I see. And the funny thing around that as well is I was having a conversation with someone recently, and it’s kind of on that same line, maybe not so much from a software development perspective, but from general users as well, which is too many people try and treat these language models like a search engine. So it’s a question and answer, and if it’s not the right answer, then it doesn’t work, rather than treating it like a something that you actually kind of have to train.

To give you the best possible context and answers, and especially in the software development environment where you can’t just go from a one-line prompt to a a solution or to a program, right? You have to layer in that context, you know, the same kind of process you have to go through with a new employee or or a or a whole team around really being explicit and where you’re trying to get to, what we can and can’t do with it. and I think, you know, like you said, maybe it’s that persistence. but is there a

Is there a kind of mindset shift that people need to consider to to really really get out of that dip, do you think?

Samy Nada (23:04.966)
Yes. And I think you kinda hit the nail on on that one. Because the problem is, especially in the software industry, is people are used with very deterministic tools. Like you have a script, you write the script, and the script is always consistent, always gives you the right answer, knows what it’s doing. And with AI, it’s not like this. W so is before you are dealing with a computer. With AI, you are dealing with a computer, but actually the engine is not on

It’s not like a real computer. It’s exactly as you mentioned, it’s kind of like a new hire or maybe a junior hire. And the shift has to come from you treating AI as such. This is imagine a person who needs learning, you need tuning. You can ask him for something. He might misunderstood what you said because whatever you said, because a lot of the time we think what we said is clear. But yeah, somebody will take the thing literally. And and and he did if especially if somebody was not experienced, you can tell them,

Look at look at this code and you can look at the code. Or like which code? Which file? I didn’t know that I should follow this. I didn’t know I should do that. I I don’t know. So the sh main shift would be stop treating it like a computer program that is deterministic, gives you the right answer. to like this this is kind of like an intern or a junior engineer that they need some calibration. You have to tell them what is the boundary, what to look for, what is success, what is failure, what is

What is a good output? What is a bad output? So you have to define to them basically how to like what the input and the output. It has to be very clearly defined. And even with that, even with that, expect that they will come back because of misunderstanding or because of something. Maybe they didn’t see everything, they didn’t read read everything. You told them to look at everything, but what is everything? A lot of what we say is subjective can be open to interpretation. So we know how to do this when we’re dealing with human.

The trick part here is when you look at AI, I’m I’m dealing with it like I’m working with a computer. So if the shift happened to your point, if you work with it like you’re working with a junior, then it’ll be the best. And expect from it what you get from a junior or an intern. That’s what you expect from it. Don’t expect to get like a like a perfect answer. You’ll get good answers. But if your expectation is like the moon, somebody who’s like 130 years of experience and doesn’t make mistakes and knows their stuff, you will be disappointed.

Samy Nada (25:28.068)
It’s it’s a right now, the way I see it right now, it’s a it’s a an engineer with maybe five, six years of experience. And if you give it the wrong input, it’s gonna give you the wrong output and it will it that person again, imagine a junior who is very confident, a very confident junior who’ve been in the company for three years. He thinks he knows everything. If you switch your mindset to work with a person like this, I think this there will be the way to get you there.

Jake (25:44.865)
Yeah.

Jake (25:50.862)
That’s interesting. And I think that goes for any kind of use of those things, right? Even in not a kind of software development context. I think it’s it’s broadly the same anywhere. you have to really train it to do what you want to do in the same as a person. It’s just that the potential that it has to become productive is is probably a much shorter ramp up time, but it still needs that ramp up time, right? so okay, well that leads nicely on to I think, you know, once we can get over that, over that dip and people can

Samy Nada (26:00.627)
Hundred percent.

Samy Nada (26:13.084)
Yeah.

Jake (26:20.641)
get that mindset out the way and really find that stride with it. With where it’s at today, where you see it in your team and in an engineering context, what are the kinds of things today that it can become fantastic at and that it can, you know, save time and and resources?

Samy Nada (26:35.984)
Yeah, so it is still developing. what I’ve seen to be great help and is explaining huge code bases and coming up with recommendation or planning for what to do next, how to address those problems. so and also so this is from one side. This is you look at a huge code base and ask it, how does it work? And you keep kinda asking it back.

Okay, how does it work? can you verify that that your output is correct? can you create a disk case that does this and this? okay, can you lock look at at that input and see if it makes sense? Can you analyze this user case to see if this is if there’s any similarities there? So the the ability to deal with huge amount of data. So again, I get back to the junior engineer. if you go to the junior engineer and ask him to run look at the code, it will take him two years to look at the code. So this is where AI kind of helps. Where

It can do the same thing, but it doesn’t take two years. It can do this in 15, 20 minutes. So it can do look at a massive amount of data, summarize it and make sense out of it, whether it’s code base, whether it’s whether it’s like input data, whether it’s user data. So where I see it strong right now, and this is what’s used used the most in in my team and other teams, is understanding or being kind of a more of a data analysis.

And and no, but just not just an analysis. Analysis and coming about like an expert person looking at the code base, looking and trying to come up with finding that that that it’s clear. The findings are clear. That’s the finding is not very sophisticated, but it there’s I don’t think there’s any human have the capacity to look at like a a a a a million line and then come, there’s a pattern, right? It can do that. It can do that, it’s strong. Writing code, yeah. If you’re writing simple code, like if a web b website or something which is not performance critical.

Jake (28:17.997)
Yeah. Yeah.

Samy Nada (28:27.746)
it’s it’s fantastic. That’s why it’s taken over the web development, taking over the the app where simplicity works and and performance usually is not a key is a problem. In in niche industries with performance and high performance computing and stuff like this, I don’t think it is there yet. it can be used for small little thing parts, but to rewrite the whole thing, it’s not there yet. And and a huge part of this is there’s not a lot of training material out there that they can train on.

especially in niche fields, right? Especially in niche fields. Like if you work in an operating system, it is a very complicated thing you work with, but there’s so many operating system codes out there. so you kind of have to think about it. It’s again it’s a this engineer who’s still very good, but he can only judge on work from experience. So he will judge based on what he saw. So if your code base or the style is kind of for foreign to it, it will need more learning and more time to come up to speed.

So it’s basically code exploration, data analysis, code generation for complex code. It it’s improving at a significant rate, but I don’t think it’s there yet. I don’t think we’re ready yet to kinda like, yep, you take over. Not yet, at least in that part.

Jake (29:37.911)
Yeah. And and that’s also a a really interesting point around you know, is it only a matter of time or or are there structural limitations to how well it can move from those very well documented areas to ones that are where the knowledge is much more centralized and maybe proprietary, right? And I think that’s one of the main things people don’t think about is if the knowledge

Of doing something very niche and specific is is proprietary or it’s behind a locked door or it’s in the heads of a very few people in the world. It doesn’t matter how competent you can make a model, if it doesn’t have that available context or or data, then it there’s only so good it can be in those areas. and I think, like you said, that’s where in those high performing computing or whether you’re working very close to hardware or or working on a niche topic,

there’s still probably quite a a way to go before there’s enough information out there for models to be as competent they are at doing things like web development, than they would be, especially in the EDA world or compiler engineering and things like that. so bit of time. how long, you know, realistically do you think until it could be as competent writing EDA tools as it could as a website?

Samy Nada (30:55.29)
It’s yeah, it is kind of hard to guess. And and I’ll tell you why. If if if if there is no hurdles, and I’ll kinda talk a little bit about the hurdles later, I think it will take four or five years because it’s con that’s it’s just convenience. It’s gonna for the top management they will for them if if you have an AI model that is either the frontier model that’s so good that they can handle information that they don’t know and they can just deal with it good, or people or some teams and companies start kinda

Taking some of those good really good models. So models right now are changing every every day, right? In a in a year or two, I think it’s gonna stabilize. And then the open weight models, which is models that you can download in your computer or your network, are gonna be as good as those ones. Again, not now, maybe in two, three years. And and once you do that, now you can train those model. You can train this in your code, you can train this in your all the knowledge, you can all the documents, all the emails. so I do think if if if there’s no if nobody’s gonna slow this down.

I think regardless of which technology it is, regard it convenience will win because this is a huge money saver for the corporation for everyone. And and instead of it’s gonna be instead of me kinda saving the code or the knowledge in a code base, it’s gonna be like I’m saving guarding my models or guarding guarding the data that the model has access to. So it’s it’s just gonna be a similar thing. It’s not like you’re gonna leave it all in the open. So my guess four to five years, to for it to handle stuff in my work kinda like it handled with design right now, right? but

Ha reaching there doesn’t mean that it’s gonna be fault. It’s still with whip development, you still know people. But I’m sure I’m saying it’s gonna get to the same level of work development. So what is the hurdles? The hurdles could be a lot of political. And I I personally do believe there is more people rooting for AI to fail than people rooting for to succeed. And the people who are rooting for AI to fail are not very subtle about it. So they don’t say that. But there are other person who are like, Well, I tried it and didn’t do that. I they just point the bad stuff.

And and and those guys can and and they can slow down. And and I’m not saying they’re being disingenuous or anything like this. It’s just they either don’t believe in it, the the the think or they think that okay, this is it’s gonna take over my job and it’s just subconsciously just kind of product. So there’s a I do believe there’s a lot of people rooting for it, they want us to fail because they f the fears of their success could could could have an impact on them. This the more people you have like this in your organization, and especially closer to the decision making.

Samy Nada (33:21.134)
the the the slow the slower the process will take but i do believe that those who are gonna fail and be slow to adopt this are gonna lose and and and ai I do believe in in any industry it’s a winner to call because those who manage to figure it out before others and when I say figure it out not just use AI where everybody’s using AI right now in the right way with the right guardrail with the right judgment for no one to do it they’re gonna win and nobody can stop them. So again this is a thing so I would say four or five years

It depends on how much politics this five years could end up being ten.

Jake (33:56.072)
And, you know, if we look at that kind of future state, right? So when we had a a conversation around what the role of the future looks like for this, and where you can see that in other industries that have automation come around, you you use that phrase of the kind of the five percent role, right? Where what currently might be a hundred percent of your role will end up being about five percent of the rest of it is all around oversight, judgment, context, accountability,

like you might have in a factory right now, where you have one person overseeing, you know, an entire floor of machines. So how do you see that moving in software development?

Samy Nada (34:33.638)
Yeah, it’s it’s gonna be very, very similar. I a couple of weeks ago, I live I live here in Oregon, US America. And I was we were famous for having some big cheese factories. So I went to she’s factory and they actually show you the assembly line from like you c from a deck you can look at it. And I saw and almost everything is automated. It’s a big machine, and I’m sure this machine has existed for I don’t say a hundred, but at least at least over a hundred years. And and even with that, there was about twenty people on the floor.

Doing some kind of visual inspection. they have all this machine with lasers and quality test, but there were still few machines, few people, sorry, looking and inspecting the output. And they can look at it, hold it or something, and just either pass or fail. Or maybe something going on a conveyor belt and the person just nudge it the right way, make sure the thing kind of get it, make sure there’s like when the when when the packet when it gets packaged that there’s nothing extension or nothing sharp. So with a huge advancement we have over.

the last few decades in and and how good machines are or is everything we still need human. We still need human to do the final verification. I am sure if I went to this same sheets factor and this sheets factor has been there forever. So maybe before all these machines were available, maybe one or two machines, it it would there were like maybe 10x the number of people doing the job. So it’s it’s it’s it’s definitely what the change will happen is gonna be less people doing the work.

and and actually you’re gonna be two kind of division here. So there will be people who do the work, like for say coding. Coding the writing code is gonna be the technical per technical guys, and you you’re not gonna need a lot of those guys. And whoever is left is gonna be the person who’s in this floor trying to make sure, does it pass the quality? Does it look right? Does it do this? which is maybe will be five percent of what we have today. The other side of people or will be the person who decides what needs fixing, like person, the person who designed the machine, right?

He looked at the problem and said, I can have a machine that does that. Right now, yes, those people exist, but you need to have huge teams to kind of to to to have anything up and running. With AI, yes, have the idea and and and and and AI is gonna do it for you. So I always look at this and the how, the what and a how. So the what is the idea, like the machine that does all this. The how is actually how to build a machine or even how to how how to run it.

Samy Nada (37:01.458)
for software engineering we because again it was more of a manual like somebody has to write the number of people doing the how is a lot compared to the number doing the what I think what’s happening what’s gonna happen is the number of what are gonna stay because you need the the what you need to know what you need to fix but the how is gonna shrink. So if you are the person you’re focused in the how just programming understand your domain know what you’re doing understand what the customer needs understand the problem because this is the skill that’s gonna pay off.

knowing what need to be fixing, need it knowing the problems that customers are facing, so you can’t come with solution. Just knowing how to fix the problem or the right code to fix the problem, I think by time, we’ll need less and less. We still need people to do so, but it’s going to be less and less and less in the next in the next decade or so.

Jake (37:49.85)
you know, the the thing about that is and I guess where probably some of the resistance comes from, it’s probably the same kind of resistance that people would have had, you know, around the time when th the same kind of automation was coming in other industries. It’s that maybe it’s that kind of yes, there’s the obvious concern of will I have a job or not, but I think maybe what’s the emotional pull for software engineering is maybe a bit of a loss of identity. It’s like well

to a lot of people that that’s what the job is, right? It’s manually producing all of it yourself, it’s having the craftsmanship to do it. And it’s maybe that sense of, well, if that’s not really my job anymore, then then what what does it become, right? Where’s the identity? you know, where do they associate their value if they’re more on the reviewing side and the and the problem side than actually the the doing and and the how side. D do you see that being something

Samy Nada (38:47.154)
Hundred percent. Hundred percent. And and again I liken it to the again, I was not alive in the early industrial day, but I imagine it was like this. And people used to do manual, and all the all they had is this physical power and and the craftsmanship where I can have like a nail that looks like this. And and and at that time, imagine the early industrial day. I’m sure the machine was not the best. And maybe it wasteful and and early on, imagine the early on. So those people had of course did they have this identity, but also the machine because they were not perfect.

It was like, those stuff not the best. They will never take over. They will never do this. They will never do this. So so the hat the exactly everything you said, but again, I don’t think it’s by any way malicious. It’s just it’s just this this whole thing subconscious. They just kind of feel like they have to resist it. And also that the industry itself, the technology or whatever, is still not there. It still produces work that’s not a high quality. And

Right now, yeah. Right now, again in this she’s factory, you needed five people. I am sure early on maybe you needed more people to verify than people that are actually working on it. And so this is gives a material or at least reasoning to for anybody who’s objecting to this and don’t don’t think it’s gonna be right the right solution to push back. see? now everybody yeah that’s why how we do know people who are against this. people thinking we are saving more money. Now we’re spending more money on token. This is how I hear and understand when I see anybody looking at

Like the cost, the this and this. But we know technology. We know technology. It gets cheaper. It’s what happened is the barrier go low. It gets cheaper, it gets more accessible, it gets better. It has always been like this. It has so if it is even 10% now, if it’s zero percent, yes, okay, there’s no go. It’s fake. If it’s 10% now, we’ve seen it over and over and over. We’ve seen how the internet started, right? It was dial-up. You wait and you and and it’s just like all this noise, and you cannot.

If you want to picture it, it takes forever to load up. Look how it is now. Everybody has a phone that you can look at everything, watch videos and while while you’re in an airplane. I mean an airplane or a car. And I’m nobody would have even thought this. And I’m not saying 200 years ago, just 20 years ago. If you would have said this to somebody, they would no. This is what technology always does. It gets better, it gets cheaper. So I think should people should bid on the future. This is what’s gonna happen. We’ve seen it happen before, it will happen here. But people who are get stuck right now.

Samy Nada (41:10.808)
and and and and the the easer again they div the identified themselves as I am the best coder. Of course they have to resist it and and again those guys have to change into from just focusing 100% on the how on focusing on the what understand your domain understand what the customer pain points are you need to do that anybody who wanna suck who wanna survive this thing have to under still we right now we still need the how and we will need it for at least five not ten years so don’t give it up up on it.

But start transitioning to the what. Start transitioning to understanding the pain point and come up with solutions or what’s the problem are and how can you come up with solutions.

Jake (41:51.006)
And when it comes to that kind of shift, right, and both in how people feel about it and also how they’re applying it, what are the best ways that kind of end engineering leadership and and whether it’s kind of leads or whether it’s, you know, real kind of high level managers, what can they do to help their teams adapt to that and and take those steps?

Samy Nada (42:15.216)
Yeah, what I can do is first make it very clear what can be used, what is the limitation, and and make it available. So lower the barriers. Basically make it easy for people to use it, make sure there’s no stigma or nobody’s being ridiculed or harassed or something just for using it by by by senior engineers who might be against it. so basically try to remove barriers as much as you can.

whether by making sure that again as I said the clear what you can use what you cannot use and also make sure that nobody’s being look treated badly just because they use they use AI. But at the same time you have to kinda go and make sure that there’s some good guard lens in place to make sure that you don’t end up with some junk in in in your production code. So so that’s kinda like this is kinda like the whole plan. So for management

Remove the barrier, but at the same time have a very well-defined guardrail and make sure that everybody just go through this. I do think what’s happening right now is a a lot of my senior engineers are spending the majority of their time instead of writing code, they’re verifying code. And and it might seem like a tedious job initially, but this is gonna be the new role. This is it it is shifting. It is shifting right now. Maybe you get to a point where you don’t need don’t need to spend much time on it, but it is shifting. So as management, you have to know that this is happening. Happening in terms of

Okay, you have to tell the people that this is happening, right? And you also and so and so we have to tell the the the the the the management we need more people to do this. We need to invest more in testing, we need to teach people more about how to verify AI generated code because some people might have experience in verifying code written by humans. Now we have this junior engineer who might know the problem when he wrote the most sophisticated code that so this is kind of like it’s new something new. So we have to teach people that.

And you have to go through the process, but also don’t even though I’m I’m a big believer in AI, but I’m not gonna be like, everything is right. We feel free to yeah, it’s it’s to give yourself break, to get back and say, let’s adjust this, let’s look at it a different way. Maybe I need to talk to this person and say, Well, I I know use AI and I encourage it, but you have to verify the output because even though it’s AI generated, it has your name on it. So we still when you get to the point where AI get the code and fix a problem.

Samy Nada (44:39.164)
But it still has your name on it. You are the person who’s gonna sign it off. So don’t use AI as j as a as a as a excuse to give me stuff that is not quality. It still has your name on it, still verify it. So again, it’s multiple things, but mostly just removing, lower the barrier, but mean while maintaining the guardrail and make very clear, don’t make it ambiguous what what what we should do, what is the process, and and and how how to move it forward.

Jake (45:07.743)
And and within those kind of teams and and maybe within yours, right? how do you see the differences between the differences or the tensions between maybe, you know, veteran, senior engineers and the younger, what people might call AI native, right? Engineers and and that kind of term.

Samy Nada (45:23.9)
So I do think everybody brings something to the table here. So I’m gonna talk first about people who are really good at good at it. So that the the native AI, AI native, those guys know the tools. They know the ins and outs, they know the best tool, the best context, they can optimize the model for the best thing. They know everything, but they don’t know the problem. They don’t know what needs fixing. They they they know the tool, right? They know the hammer, but they don’t know it was just kind of like you just swing it right or left, you don’t know what to do with it. The veteran engineers, they know the problem.

And in my team, those out of the veteran engineers who started to kind of use AI heavily, of course, it’s expected the veteran engineer would have some resistance because it’s a new thing, it’s new to them. And of course, you look at the junior engineers, like do you know everything and talking about this model there, this thing there, and the it’s kind of like they speak in a language they don’t understand. But once the veteran engineer gets into, because those guys know the problem, they know what needs fixing, this is the power. I do think that the veteran

Jake (46:12.428)
Chao.

Samy Nada (46:22.0)
Veteran engineers who use AI are so powerful. The sky is the limit. They know the problem, they know the calibration, they know when to adjust it, and it’s a multiplier. AI is more of a multiplier. So if you are good at what you do, it’s gonna be a multiplier of all the good things. If you, if what you produce is not the best, there’s gonna be a multiplier of junk. So so this is for people who are good, yeah, like really into it. The other side I see too is some veteran. like I don’t see any young or like junior engineer who is just like an anti AI.

But what I see a lot is like sooner who are are against it and they kind of pick and hole and everything. And for those pe yeah, I I have to work with them and said, AI is coming. So this is not gonna, if you’re gonna come to me and tell me, yeah, that’s the AI is junk and it’s not producing, I said, Yeah, everybody will use it. And if I told them not to use it, they will use it without telling me. So we have to adjust to the world we live in right now, make sure that we have the right expectation, defining what is correct, what is not, and help those engineers, young engineers, to to be able to

No, what is good and what’s bad. I have to tell them that. And now they can use AI and judge if it’s good or bad. Have them push the judgment to them. Teach them that. This is what you need to teach them. What is good? Teach them to to tell what is a good from from from bad when it comes to code or product or output or any of this kind.

Jake (47:39.436)
when we think about, you know, where where it is at today, its usefulness, the whole kind of debate about tokens and token costs. you know, when when we’re looking maybe three years, five years down the line, you know, where do you think those kind of things are gonna get to? W what what’s gonna change?

Samy Nada (47:55.663)
gonna free i think this kind of stuff is gonna be free i don’t think so this is why even though or every other organization including mine have to imply have to have very cost saving measure in terms of how many tokens and all this kind of stuff and and adjust this context do this do that fine tune it i i believe all this is gonna be gone in the next three years next three years everything is gonna be almost free and the models are gonna get smaller you can run them in smaller machines

you you’ll be able to run very powerful models on on machines like this. Again, it might take more than three or four years, but but but it’s it’s it this is gonna happen. So optimize, but keep in mind one thing. Don’t spend too much time over optimizing because the more optimizing you do, unfortunately, unintentionally, you are putting some constraints on the future model that can come and can solve them. Like the first few models, I have to be very careful what to tell it you are an experienced engineer, do this, do that. I’m kinda like

Very careful was a prompt. I think we’re getting to the point where prompt doesn’t matter anymore. That this models, like I was asking to do something, and it created like seven agents. Basically, five agents to do the job and five agents to criticize the job. I didn’t tell it to do any of this kind of stuff. And it’s every every co every claim, there was like two, three agents trying to see why it doesn’t work. All of this is happening. I don’t control it. So the more control and guard you put that make it permanent, you have to be just careful of it.

So when you come maybe a new model comes in, have a way where, okay, for this model, this is a garden, new model comes in. Okay, let’s start from scratch. I know it’s a lot of work, but it’s a new technology. It’s a new technology, and you don’t want to get stuck. You don’t want to get stuck at a point in time. And and and yeah, this new working with new technology means that you have to be quick to adopt and adjust your style as you as it comes with it. Again, my my my forward look in this tokens are gonna be free. You you that there is a less and less and less babysitting. Again, think.

It’s basically it’s gonna get it without from an intern to junior to a senior engineer. Still makes mistakes, but needs way less oversight and doesn’t yeah, and and the difference here is not gonna cost as much. So senior engineer cost and and this is not gonna it’s AI is not gonna cost as much. So the actual cost is gonna go go go low. But in terms of the experience, the quality of the work you get out is gonna be like a senior engineer.

Jake (50:16.028)
And if we look at today then, what are the kinds of things that engineers should do now to prepare, to make best use of this and to start their journey if they haven’t really started it already?

Samy Nada (50:27.44)
Yeah, so I it depends on where you are, but if you never use AI, I would I would just basically for programming, I would start by using it to write very basic scripts and see if it works or not. Once you can make the write the basic script, maybe you can introduce it to other script you wrote and have it explain it, find bugs in it. once you go through this, expose it to the larger code, have it explain it. Once you are comfortable with the answers there, start to have it the pro profile.

And maybe after that you can start adding, ask it to add an instrument code or maybe add basic functionality. and after that, you can even ask it to plan big features. This journey could take a year or two. Take it at your own pace. This is what people that that that in programming and and need to understand. Once you get this, now you know the limitation, what it can do, what it cannot do. The next thing you need to work on are people who are already there or are close to the end, the focus has to be in the domain knowledge, understand.

Understand your industry. If you’re working in the EDA, understand design. If you’re working in in financial risk management, understand the financial. Understand next time you there’s a customer coming up with a complaint. I know it might be coming like through marketing and other people who handle this kind of stuff. Try to understand what is the problem. Don’t just kind of look at the curated, okay, just go change this line or fix this bug or something. What is the pain point? What is the problem that the customer is facing? Right? What is it, their day to day struggle?

What is the thing that the eventually wanna fix? R and D engineers, right now, if they just focus on this, you’re gonna you you again, you’re gonna get stuck in in in the how. And the how we will need less and less. This is this is no doubt about it. We’ll need less and less people eventually that doing the how. You have to transition to the what? Use the time right now. watch YouTube. use AI itself to give you some sort of curriculum. What what how my what are my customer pain point? Try if you can engage with the customer to understand what’s going on.

Please do. You have to understand the domain. If you don’t understand the domain, you’ll be left behind. Maybe not tomorrow, maybe not in five years, but certainly in ten years, this will be the case.

Jake (52:33.462)
Well, I think that is a a really good point to to wrap on. And and maybe the last the last point I always like to add into these kind of conversations is what what do you think is the biggest misconception, biggest myth that you hear that you just wanna kind of bust and and leave on a on a note today?

Samy Nada (52:54.386)
yeah, the I think the biggest one right now is the people who think that AI as it stands today is basically can fix everything, can fix all all problems and it can do everything. I know not everybody has this misconception, but some people are, and and I this is not correct. We are not there yet. We are not there yet. It’s it’s are we getting there?

maybe again we still need human inter like interactions and review and stuff, but we are not there yet. And if you think we are there yet, you don’t know, you haven’t played with AI enough. You the more you go, the more you go deeper into it, you will understand the limitation and the boundaries, and there is still a lot of boundaries. Then these boundaries can be removed and get better, yes for sure. But we are not there yet. So if you think it is it is it is there, yeah. Think twice, read more.

again, if you work in an industry where it’s very simple, maybe that’s already kinda cover it, but the the more niche environment you work on is not there yet. Maybe five years from now. So

Jake (54:01.836)
Great, well look, Tana, this has been a a really useful conversation and I think also quite a unique insight from someone that’s directly seeing and and leading this kind of transition, you know, in a industry context, right? Especially in something as as niche as EDA. And I think the key takeaway, you know, correct me if I’m wrong, is is probably that AI is not simply about replacing engineers or creating instant, you know, geniuses. It’s much more about changing the shape of that work as

all technologies right do over time, right? And I guess more of the raw generation, the the how maybe automated, but the ability to frame a problem, the what, guide it, validate it, becomes even more important. And so I guess for anyone listening who wants to follow Sami’s thinking on AI adoption and that kind of engineering leadership, I’m I’m guessing the best place probably to do that is your LinkedIn, right? Well now.

Samy Nada (54:55.782)
For now, yes. Yeah. LinkedIn is definitely the best way. Feel free to connect and give me a comment. Look at my post, give me comment, give me ideas if you want me to cover or or share my experience.

Jake (55:06.582)
Well look, Sammy, it’s been great. Thanks again for joining us on the IIT podcast. And to everyone listening, if you enjoyed this episode, follow the podcast, connect with us at Five V Tech, let us know who you think we should invite on the next show and we will link Sammy’s LinkedIn profile and Siemens as well. but Sammy it’s been great. Thank you very much.

Samy Nada (55:25.627)
Likewise, thanks for having me. It was a pleasure.