E181 - AI and the Future of Developer Productivity - Viktor Jarnheimer

 In Episode 181 of the AIAW Podcast, we sit down with Viktor Jarnheimer, Founding CEO of Proxify, to explore how AI is transforming productivity inside modern tech teams. Drawing on his experience building a global network of remote developers, Viktor explains why AI-augmented development is becoming the new baseline—and what that means for hiring, performance, and competitive advantage. We unpack the widening productivity gap between teams that use AI and those that don’t, the critical role of data engineering in unlocking real AI value, and how companies can scale globally by embracing a borderless talent strategy. From balancing speed with code quality to redefining the role of human expertise in software development, this episode offers a practical perspective on building high-performing teams in the age of AI. 

Follow us on youtube: https://www.youtube.com/@aiawpodcast

Transcript

When Local Coding Hits Production

Anders Arpteg 0:00

But uh but you use cool code a lot then, right?

SPEAKER_07 0:03

Um I do. And every Friday afternoon on production. In production as well. So not directly in production, but do you also do coding yourself these days? I do. Um I know the RD team, they sometimes dislike when we from the C-suite are going in working directly in the code, not following the main processes. We are broken production twice so far.

Anders Arpteg 0:33

But uh you pushed it to the main branch as well, and yes.

SPEAKER_07 0:37

I can't blame actually we didn't. Um so I thought we were safe, yeah, but we still manage just on our local environment because we were rushing so fast to shut down elements in production. I thought we were safe. Yeah.

Anders Arpteg 0:56

Shut down all elements?

SPEAKER_07 0:58

Well, um, so in my case, I had a very old environment file, and this environment file triggered a production API. And this production API was really confident that our environment uh keys have been production keys have leaked, which they in some way had. They've been on my computer for four four years but never used. And that caused the email system to shut down. Oh, Jesus. In production, even though I was working locally. I can't blame Cloud Code for this, but I had such a good flow and maybe didn't think about all elements.

Anders Arpteg 1:43

Yeah, yeah. Uh easily happening. Well, you can almost blame AI, right? Yeah, yeah.

SPEAKER_07 1:49

I mean, of course I did to my colleagues.

Anders Arpteg 1:52

No. I mean, great to hear that you do coding as well. Uh that's it's so much fun. It's more fun nowadays than it I mean I've been programming since I was like seven, but I think it's just getting more fun when you can be so much more productive these days, right?

SPEAKER_07 2:08

Definitely.

Anders Arpteg 2:10

But do you use code or cloud code also for non-programming tasks, or is it mainly for coding?

SPEAKER_07 2:16

I view it as a challenge to pretty much use it in every case that I can. Um and um could be everything from writing a letter to anything else that I need to do in a repeated way.

Anders Arpteg 2:31

You know, we've been speaking a lot about you know um making PowerPoint presentations here or other type of presentation. Do you use Cloud for making presentations?

SPEAKER_07 2:42

I haven't so far, no. Uh of course, like getting feedback on presentation, etc. Yeah. Uh, but you use most mostly Google Slides, and yeah, it doesn't work that well with Google Slides yet. No.

Anders Arpteg 2:57

It doesn't. And I know you know this has been a pet peeve of mine, you know, that AI is uh really good at some things but horrible at other things. And in you know, something that so many people need to do is simply to you know make a presentation, but AI can't even do that. How can you claim that AI is so intelligent if it can't even make a PowerPoint presentation?

SPEAKER_07 3:18

If it's like just align the slide, please. That would be such a nice feature.

Anders Arpteg 3:22

Yeah. Of course it will at some point.

Goran Cvetanovski 3:25

I think we are getting there.

Anders Arpteg 3:27

We are, yeah.

Goran Cvetanovski 3:28

Yes, we are getting there. So if we but maybe not with one now, we need to have two. If you go to Nano Banana, yeah, it's doing quite good stuff actually. When a notebook LLM is doing quite good, uh, what is called representation of infographics and things like that.

Anders Arpteg 3:41

But that's images, it's a different thing from a presentation. Yes.

Goran Cvetanovski 3:45

But now if we have those images and you move them to Google Cloud or to um what is the presentation part, uh slides, right? Yeah, then in slides you can probably make a better design. So in two, three iterations, maybe it's there.

Anders Arpteg 4:00

Yes, and I think also Copilot now actually is starting to get better at Microsoft and it actually can create some type of presentation that it's not Microsoft anymore.

SPEAKER_07 4:11

We should probably name the date right now because in a few days it might change.

SPEAKER_00 4:16

Yes.

Goran Cvetanovski 4:17

Uh but Meta just released uh what is called like a new uh image model.

Anders Arpteg 4:22

Um the Muse models, yeah, exactly.

Goran Cvetanovski 4:24

So they were operational today, and um the imagery is now getting much closer to what the mid-journey has.

Anders Arpteg 4:31

Yeah, wow. And OpenAI just released you know imagine two as well.

SPEAKER_00 4:35

So yeah, but uh I was talking about OpenAI. Oh email. Yeah, yeah, I meant the OpenAI, yeah, in ChatGPT.

Anders Arpteg 4:41

Yeah, yeah. Yeah. That's amazing.

SPEAKER_00 4:43

Yes, uh amazing stuff that's coming up. Yeah.

Anders Arpteg 4:47

Anyway, uh, very welcome here, Victor uh Jean Heimer. Is that correctly pronounced?

SPEAKER_07 4:53

Yeah, yeah, close enough.

Anders Arpteg 4:55

CEO of Proxify, and uh, I'm very much looking forward to hearing more about what you do at Proxify and uh really who you are. We had a lot of discussions here about you know uh CEOs also working actually with Qualcode, and I think that sets really a great role model. I think everyone should. And uh yeah, so glad to hear that you are also pushing stuff in production, even though you didn't mean to. Happy to be here. So very welcome. Um, cool. But perhaps you can start by just giving a brief introduction to really who is Victor and how did you become the founder of Proxify?

SPEAKER_07 5:34

I've been programming pretty much every day since I was 11. Yeah. I did study finance, however, uh in high school in university, but I've been so interested and passionate about building things. This has resulted in a few startups, and probably because I was born in on Ireland where we were a small village. Yeah.

Anders Arpteg 5:58

I'm from Kalmar as well.

SPEAKER_07 6:00

Oh wow. It was hard to find like-minded people in a village of 350 people, so you had to go remote. Yeah. So remote felt very natural from the very, very beginning. And since then, I think I've been tinkering with a lot of websites, a lot of very high-tech products that we have been building, like an ed tech platform that is today used by about half of all the high schools. Today it's called Nook Flex, but when I run it, it was called Evuma. And AlbaCross is another company.

Anders Arpteg 6:46

Um so you started a number of companies. When did you first start your your like first company?

SPEAKER_07 6:52

Uh H of 15, Shirko Jack then, which is today the biggest booking booking platform for traffic schools, and uh they're they're doing well really well, but I can't take any credit for that. Um but I did found it.

Anders Arpteg 7:06

Oh, I mean, that's amazing. Yeah, that what year was that?

SPEAKER_07 7:10

Uh that's a long time ago, like 2006, probably.

Anders Arpteg 7:13

Uh-huh. Um then you started a number of other companies as well.

SPEAKER_07 7:19

Well, uh I uh these three were definitely the main ones. Um, but what they all had in common was that we had a team of remote engineers.

SPEAKER_04 7:30

Right.

SPEAKER_07 7:31

Especially in the last one with Noteflex. We had I I would I would argue he's still the best software engineer that I ever worked with. Uh his name was Max, he was in the same age. We were both 18 years old. I was working from my desk in on Ireland, and he was working from his desk in uh the middle of Siberia. Oh really, and we had such a connection. We both loved PHP so much, and we could I should maybe I shouldn't say that. Okay. It's way cooler nowadays, but um we we loved PHP. Yeah.

Anders Arpteg 8:21

Okay. I'm biting my tongue here. It's not the most beautiful language, right?

SPEAKER_07 8:26

Uh maybe no, maybe not back then, but it's actually got a lot better. A lot better. Okay, cool. Um I assume you're more towards maybe some JavaScript-based.

Anders Arpteg 8:39

Yeah, well, JavaScript is is also kind of nasty, but I think TypeScript is nicer. Right?

Henrik Göthberg 8:44

Yeah.

Anders Arpteg 8:45

And then you have a backend. Do you have a backend the favorite programming language? Yeah.

SPEAKER_07 8:52

Well, PHP along like Laravel and Vue.js on the front end if I had to choose.

The Max Story Behind Proxify

Anders Arpteg 9:02

Uh cool. And at some point you um came up with the idea of Proxify. How did that come about?

SPEAKER_07 9:10

Well, the the story with Max is actually kind of sad. His father, he knew that he wouldn't be able to afford university. So he took the money that he had for the university, saved so far, and he bought a computer to his son, who was then like 15, probably 13. Yeah. And he told his son, like, you need to learn coding. This is the future. And that's how he started to learn uh PHP. And when we were working together, and the business got acquired, he lost his job, and solely because he was working remotely. He then started university, and thankfully had saved up so much money when we were working together, so he could afford going to Moscow to go to university.

SPEAKER_04 10:11

Oh, right.

SPEAKER_07 10:12

When he completed university, he reached out saying, like, please, can you help me find a job? And I thought it was so insane that he wasn't able to find a job. And he said, like, no, well, they don't view me as senior because I just graduated. So um, my co-founder and I, Lee, we were then working in another business, and we told ourselves that all right, let's find Max a job. So we made a post on LinkedIn that got a lot of interest from other people. Like, yes, we want to work with Max. Like many of them knew that I was a programmer, and the fact that I vouched for another one telling that this is the best person ever worked with made it possible for him to get his next job. And then we just continued to help a few other friends of Max and ours to get a job. Yeah, and we helped the clients who sort out the payments, etc. And eventually we realized like this is a much bigger business than our current. So that's when we made the decision to make a switch.

Anders Arpteg 11:28

And please do ex explain, you know, what's the business idea or Proxify?

SPEAKER_07 11:34

So if you were looking at all the software engineers in the world, um the difference between the very, very best one and the worst ones is insane. Right now, Stanford University, they have a data set of 120,000 software engineers, and they're looking at like how productive they are. And they have seen from this data, and we have seen from uh a lot of other cases from ourselves where we looked into team, and the top 25%, they are six times more, they're having six times more output than the bottom 25%. I believe there are very, very few occupations and professional professions where you have this insane difference, and that study also showed like if you're able to find the top 5%, they're usually having a higher output than all other people in the team combined. And I see this so many times. Every time we look into businesses and we are gathering this data, we see this insane differences in the team. Proxfy, we are driven by like how could we identify the very very best ones and give them the opportunity they deserve. It has been a huge focus on, I would say, the engineering side, like how can we get a job that they deserve? And it has resulted in a lot of companies coming to us asking for like who are the top engineers.

SPEAKER_04 13:26

Is it software engineers mainly, or is it other types of engineers as well?

SPEAKER_07 13:30

I would say any kind of role within a software engineering team. So, yes, it could be um it could be designers, QA sometimes, data scientists, and similar. But those kind of jobs.

Anders Arpteg 13:48

Sounds great. But uh please tell us, you know, how do you identify the top one five percent of the top uh people?

SPEAKER_07 13:56

It's certainly tricky, and I think we we are also having a lot of questions nowadays how this is gonna evolve. Today, where we're finding the highest um predictability is from pair coding sessions. It's like where you are two really like one very senior person talking with other engineer solving problems together. That that's an area where we find the biggest uh predictability. However, with that said, I would argue that there are no pre-hire evaluations that are having a very high predictability. It is hard to know without working with the person if they're good or bad. Today, Prox5, we have worked with a lot of engineers for years. And most of them, I would say like four out of five, they have had a job through Prox5 before. And if you have worked with them before, maybe tracked metrics, you know, the perceived perception from the client, then you have a very good understanding about the productivity. Um, but that requires you to work with them.

Goran Cvetanovski 15:20

Do you have any ranking or scoring models within your we do?

SPEAKER_07 15:25

And quite early, um, many, many years ago, we found a guy also remote who specialized in like how could you use AI to share pick candidates? So we built our own model where, and I would say this is our moat, which helps uh help us identify like which developer is the very best one uh in general, and also what software engineer is the very best one for a specific role.

Anders Arpteg 15:58

Based on the CV or based on some programming session, or or what?

SPEAKER_07 16:03

I I can give you an example. Like it's quite easy for you to find a model that evaluates your English level, for example. Like you're a C, for example, or C2. Um that's not the most, and we have a lot of third-party tools to help us evaluate software engineers. However, let's say you have a question to a software engineer asking, like, are you okay with working at the company that are not writing any unit tests? They're a big company, etc. And they say, like, yeah, no problem at all. How does that respond correlate with a future success at a job? How does that correlate with uh code quality as an example? Another example is like one software engineer is having an ask that is four dollars lower, but the English level is slightly light lower. Does that compensate for a slightly worse English level long term? Um and we are now having a lot of data like what is happening happening with that software engineer 18 months into the project.

Anders Arpteg 17:30

And what's the answer?

Remote Work Rules And Time Zones

SPEAKER_07 17:32

Uh I I don't know about I don't know about that specific trade-off, but uh in general, people are quite prends price uh in they're not price sensitive if the engineer is really, really good.

Anders Arpteg 17:47

Right. It's worth to pay for the talent, right?

SPEAKER_07 17:50

For sure.

Goran Cvetanovski 17:52

But Proxify is also a global company, right? You have you're represented everywhere, the pool of uh resources is worldwide uh offices. Uh how does that look like for you guys?

SPEAKER_07 18:04

We haven't really followed our own business model by building in a remote organization. We do have an HQ in in Stockholm where we're 30 people, however, um many of those people in that office they're remote, and then internally the other 90 people are remote. We are internally in almost 30 different countries, and it works great. Yeah. I'm surprised. Yeah. Yeah. But but but it's still, however, we have noticed when you're building a remote organization, we're not we're not so remote that we're willing to take anyone from any country. And we have two restrictions that have been driven by data. The first one is that it should be easier to meet. So there are it is you need to be, even if you're in the same time zone, there are benefits of making it easy to meet if you need yeah, three hours difference in a flight, or four hours or five hours, or something like that. We have some and when it comes to time zone. Now I'm talking about software engineers working for in a team. We see much worse retention if they are more than four hours west, no, east of the person, but only two hours west. So apparently the person that you're reporting to doesn't like if they are disturbed in the evening, yes, but they get a good feeling when they're waking up that they're a lot of work done.

Goran Cvetanovski 19:53

Interesting. But the the pool of resources that you have the pool of resources that you have uh that is international, right? That is global. You can get like a person from I don't know, New Zealand that is a super great uh uh okay. Maybe the time difference is not good, but you have resources and uh candidates from all around the world, right?

SPEAKER_07 20:11

Yeah, everywhere where the time zone correlates with the clients. Right. Yes. But we would never, or sometimes we do, but we would really, really um try to avoid, for example, placing a software engineer in Europe with a client in the States.

Henrik Göthberg 20:31

Right.

SPEAKER_07 20:32

Even though, yeah, even though both the client says like, ah, it's fine, don't worry about it, and the developer says, like, I'm working late anyway. We know that historically it dramatically impacts the communication.

Goran Cvetanovski 20:47

Yes, I agree. It's interesting. So um, so we have um we have offices in Dubai and uh India. With India, it's almost impossible to uh uh it's not impossible, but it's uh they're uh six hours away. Yeah. Uh and then you have Dubai around two to three hours away. Um and it's interesting for the Dubai people because when they are working, when we basically start working, they are already three hours in now. So they have already finished like 60% of the things and they're starting pinging everybody. So when you have like a like uh this global company and etc. and everybody is like interconnected and they need to talk to each other because the teams are made like that, it also brings a lot of stress to the teams that are actually not at work, which you said, like okay, even that uh the person from US and Europe said, like, ah, it's okay, I even work at nights. Yeah, after some time it brings uh friction a bit because it's here is like what 6 30? No, 5 30 when they are starting in Dubai right now. Or actually in the winter. So it's uh if somebody texts you and you wake up 6 30 and it's something burning, yeah, not a nice feeling. So it's a little Difficult, yeah. Super cool.

Anders Arpteg 22:03

Cool. I still think some people don't really understand what Proxify does. Can you explain who who should contact uh Spotify uh Proxify and how can they help some other companies?

AI Skills And The Widening Talent Gap

SPEAKER_07 22:18

In most cases, a company comes to us because they have a role to fill. Yeah. Usually a role that has been hard, hard to find a like a top-tier candidate for. It could be something small that needs to be fixed, maybe a bug, or it could be some long-term engagement. They come to us asking for a very, very niche um, or um, it doesn't need to be niche, but they're that they have a long list of requirements. Since we have had about 20,000 applicants to our network for many per month for many years, we have a massive database of software engineers mostly, not only software engineers, um, but talents, tech talents that we know we have evaluated them, that these are outstanding.

Anders Arpteg 23:19

Have you evaluated 20,000 applicants per month?

SPEAKER_07 23:22

We do, we do. With that said, not all of them have gone through every step in the funnel. Yeah, I would say 85% we don't have any interaction with, and it's 100% AI driven. Um but yeah, we it's an insane operation to evaluate those.

Anders Arpteg 23:43

So, first, how do you get 20,000 applicants per month?

SPEAKER_07 23:46

It took us a long time to get there. First, we probably had like 300 applicants a month for years, yeah. Then we had a few people that were in the marketing team that worked on the branding, um where we yeah, we I think we became quickly a known place for good jobs that are long term.

Anders Arpteg 24:21

Um built a brand and people thought that if they apply to you, and and basically do they get a full-time position then at Proxify or how does it work?

SPEAKER_07 24:34

Most of them are do, yes. Okay, we have very few people working as giggers, and so most talent platforms worldwide, they're focusing on gigs.

Anders Arpteg 24:48

Yeah.

SPEAKER_07 24:49

And yes, this is a common way for Proxify to get clients, but the average time or the median time a client is working with an engineer is 18 months, even though the median request is like, I need a developer for two months. Yeah, right. So they stay long. And we have many engineers that started seven years ago when we were founded, we're still in the same business.

Anders Arpteg 25:20

Okay, so you build up the brand, you get a lot of applicants, and um then somehow you need to filter down the the 20,000 applicants into a much smaller set of people, and you have some AI.

Jevons Paradox And On-Demand Software

SPEAKER_07 25:35

Yeah, that's our own, yeah. We we it's it's not like one model, but we call them together, Gandalf and uh Eros. Gandalf is responsible for not that that's the one rejecting the candidates. Okay. And it was, I would say like four years ago where the AI model became better than the the recruiters in this. And then we have the Eros model, and the Eros model is the one that is cherry-picking the right engineer for the roads. And I would argue that that one got better than a human matcher maybe two years ago.

Anders Arpteg 26:14

And I know it's your you know secret source here, but I'm still going to ask the question. Feel free to not answer. Uh but okay. So, how does Gandalf and Eros work?

SPEAKER_07 26:26

Um it's a very similar to machine learning. Um we have a lot of data on But you have like a questionnaire that you send out?

Anders Arpteg 26:38

Yeah, what do you do?

SPEAKER_07 26:40

We it we have a lot of different interviews steps, and we try to structure the data from those interviews.

SPEAKER_04 26:49

But automated interviews, right?

SPEAKER_07 26:51

Or yeah, a lot of it is automated, but it's also humans asking these questions, input it in a form, etc.

Anders Arpteg 26:58

But you can't do that for the 20,000.

SPEAKER_07 27:00

No. Well, the first uh the the first 85, that's just a questionnaire. The ones that are like which there some of them are fast-tracked and they're getting a lot of human attention. Uh, but the ones that are like in the gray zone, they're asked to do more automated tests, self-assessments, etc. etc. And um so we oh no, we do not talk to 20,000, but we certainly have more than 600 interviews a month, probably like manual interviews. Yes, yes. There are about 40 people in the operations team talking to software engineers or tech talents, impressive.

Anders Arpteg 27:58

Okay, so then you have um like a pool of uh experts and talents, I guess, in some way, and then you get the requests from companies. How would you say that differs from like employing a consultancy?

SPEAKER_07 28:12

Do the company normally employ the person, or is it more of a two-month kind of gig that they um it's usually a relationship that is similar to an employment? Um they have the ambition to stay there for a long time. That that's the plan. And that's when But then they still get salary from Proxify, or they do, they do, and but that most like many of our clients are big businesses who don't want to have an employee, employer in a remote country. Right. They're they gladly help get help from us, and we handle a lot of the HR stuff.

SPEAKER_04 28:51

Yes.

Anders Arpteg 28:52

Oh, okay. I mean, awesome. Uh and and uh okay, so the steps basically you do some automated um assessment, so to speak, of the people. You have people that are then doing a lot of interviews as well, and uh, and then you have the people uh abroad. Do you have some kind of matching when some kind of customer comes in and they want to have help with some job? How does that matching work? Do you manually like try to see that this person should fit this, or how does that work?

SPEAKER_07 29:23

It is both these models are trained in the same way. We have we we have data that correlates with success. Um, for example, like how long did this person stayed at the position? And what was the rating from the client of the engineer after three months, for example? So we have a lot of uh signals that this is a good match or not.

Anders Arpteg 29:51

And I guess that influenced the salary that they get also.

Goran Cvetanovski 29:53

Oh, yeah, absolutely. And and the rating is coming from the customer, right?

SPEAKER_07 29:57

Or yeah, um as of today, that's how the model is trained, or if it's staying longer, but we are actually going more towards a more objective analysis of the engineers. There is a quite big difference among our clients in this case, how they perceive an engineer. Um, but there are much more objective ways to evaluate engineers, which we're now digging into more. So, like what's the actual velocity, etc. etc.

unknown 30:34

Yeah.

Anders Arpteg 30:34

But you have some automated way to do the matching as well.

SPEAKER_07 30:37

Yes, correct? That's fully automated.

Anders Arpteg 30:39

So so they okay. But they do do companies come into Proxify saying I want this type of competence, or or do you post like these are the type of of uh jobs that you you can help with?

Bad Enterprise UX As Opportunity

SPEAKER_07 30:57

We we don't post jobs really. We have a team that is working with the matching where they're getting a list of the potential candidates, and then they're going through this list trying to understand, like, all right, is there something that this AI model doesn't capture?

Anders Arpteg 31:16

Um But you do have an AI model that do the matching as well, without the team helping out if necessary.

SPEAKER_07 31:21

Okay. And I I think it's always gonna be human in the loop. The engineers in the in the marketplace, they're usually having a full-time position today. Otherwise, you should be worried. So we need to convince them to really embark this new mission. And having an automated email saying, like, hey, you should have this new job instead of a human. Um I go with a human.

Goran Cvetanovski 31:51

So basically, uh a customer comes and says, like, uh, hey, I need uh help, right? For two months, but it ends up with the 18. You're saying like these are the top candidates that we can offer for this position, approximately, right? And then the customer basically pre-qualifies those, let's say five and chooses one.

SPEAKER_07 32:09

We yeah, we usually present like one or maybe two. Okay, fine.

Goran Cvetanovski 32:12

But let's say it's one to five.

SPEAKER_07 32:14

Exactly.

Goran Cvetanovski 32:14

Two, let them be. Good. So you're doing also the pre-selection, you're the service is full, basically. This is the we did the pre-qualification, we did the matching. Here is two, choose two. Um right. Much easier to explain. We went too complicated in this.

Anders Arpteg 32:33

I think you mentioned also a bit about you know what kind of qualifications should a person applying to Proxify have. And I think you mentioned about you know if they should if they should use AI or not while working. Or what's your thinking there? You know, what what do you look at at a person applying to Proxify?

SPEAKER_07 32:54

Three years ago, we actually we made a rule that if you're not open to AI, you won't be allowed to enter the network. And uh basically, like we were we were making sure that we're having people that are open for change. And I think people being open for change is always good. Right. We want the curious people, etc. For the first years, and I would kind of argue till November last year, we didn't see a massive productivity increase from AI. Um, you did see from the data an increase in output, but you also saw other elements like you need to rework the code a lot, you need to go back, you saw quality issues uh in the code, worse than quality in many, many cases. So the net gain was much lower in the brownfield projects. And but since November, I would argue that that change. We see a quality in improvement in many cases for the ones using like clawed code in most in most cases, yeah. And we truly see a productivity gain, which we didn't really see before.

Anders Arpteg 34:22

It's just the quality of Claude, or what that would be my guess, yes.

SPEAKER_07 34:29

Going from mostly like an autocomplete or a separated chat to a more integrated way that suddenly doesn't feel like talking to a five-year-old.

Goran Cvetanovski 34:40

You mentioned something before, which was like um the the differentiation between a good and bad is basically six times more.

SPEAKER_07 34:48

Yeah, and that's like pre-clawed data. Yeah, exactly.

Goran Cvetanovski 34:51

How do you see the changes right now? Is it harder even to evaluate uh the level of people?

SPEAKER_07 34:59

I I spoke with um one of the researchers from Stanford last week, uh, who's also a co-owner of Proxify, and they have seen that the spread is now increasing. So instead of 6x, it's gonna be more from now on. And we're having a few cases of people that are extremely productive.

Goran Cvetanovski 35:23

All right, wait, wait. So uh it's increasing in the gap between the two.

SPEAKER_07 35:27

Yes, the spread is increasing, yeah.

Goran Cvetanovski 35:29

I thought it's gonna equalize.

SPEAKER_07 35:31

Uh yeah, uh I would argue I I would probably think that too. But it makes sense, yeah.

Goran Cvetanovski 35:37

Because the person who actually knows how to how to uh to code and uh evaluate code, he can be now instead of six times, it can be 10x, and 10x or 1 is still 10 and uh 10x or whatever it is of six is a little bit more than that.

Anders Arpteg 35:53

Now I mean I I think it's it's a very interesting in you know conclusion that we we need to have people that really know how to use AI properly, yeah. And it's really back to the original kind of quote from some oldman like uh uh one or two years ago saying, you know, AI won't replace people, but people using AI will replace people not using AI.

SPEAKER_07 36:14

Yeah.

Anders Arpteg 36:14

Right? That's very much true, right?

SPEAKER_07 36:17

Yeah.

Goran Cvetanovski 36:18

But still, I've I I'm I haven't thought about this. This is great because uh, you know, everything that you read from BCG and all the AI, uh what is called uh index and etc. It says like the gap between the uh the seniors and the juniors is actually like um uh decreasing, which means that right now using the junior using uh AI can actually be as equivalent or productive as seniors. So that was uh interesting.

SPEAKER_07 36:43

No, but I I do think like the the junior they've had a massive gain. Like the junior using these tools, if if they have the right other skills, they're gonna get a lot out of this. But with that said, it's gonna be a group that's gonna be extremely good using these workflows, etc. And it's truly an art. Another thing that I find extremely interesting in this study is the fact that it's not just if you're using it or not, they see a big difference how long time you have been using it. So the companies that started using AI in the workflows, etc., a long time ago are now much, much better than the ones that didn't. And there's a big pool of companies that started to use AI maybe a few years ago. They didn't see the productivity gains, so they stopped waiting. And they are now the ones that are starting to lag behind even more. So they see both the spread among the individuals, but also in the ones, the teams that are not using it.

Anders Arpteg 38:03

So it wasn't a you know super easy start, I mean, so to speak, in the previous years in using AI for development. But if you were to give some advice then to companies that want to hire engineers and they have to choose, you know, either I you know pay a small salary for junior people that use AI, or I you know pay higher salary for more senior people. What would be your advice there?

SPEAKER_07 38:29

I would definitely go with the more senior. Um, but it depends. If you're gonna start a greenfield project um and you want to get some kind of proof of concept, a junior with a low pay, yeah, go for it probably. But when I'm looking in the organization and you're building things that require a lot many people, you will meet challenges that are complex. Um challenges that senior developers are facing a lot both today and before, that AI so far does not solve. Right. Um I would definitely go with the senior. And given the spread that the spread is increasing, I would say speaks for it.

Henrik Göthberg 39:22

Yeah.

AI News Musk Cursor Mythos

Anders Arpteg 39:23

I guess some kind of you know argument for why that is could be that AI can work more or less like a junior developer, but it can't really work like a senior developer because a senior developer can do the high-level reasoning that an AI really couldn't. And that's why you know, if you put a senior developer together with the AI, you get you know the best of both worlds. Yeah, yeah.

SPEAKER_07 39:46

Yeah. But I can also imagine that that the developers that or yeah, the software engineers that were the very, very best ones yesterday might not be the same that are the best ones tomorrow.

Anders Arpteg 40:01

Oh, that's interesting. Can you elaborate? What do you mean? So I I I can see what you mean here, but but just to try to understand why that is, how would you describe it?

SPEAKER_07 40:10

But I think there are some skills that are truly, truly amplified with AI. Yeah. To give one example, um if you have a good business sense, you can choose the direction where to go. Uh, should not how fast can I go from A to B, but maybe we should go to C instead. That group that has a strong feeling of picking the right direction doesn't need to go to the person asking them for the task or the people they're building for for feedback as often, which will be a bottleneck. And maybe yesterday, the ones that was the winner was the one that wrote the most beautiful syntax. Uh, of course, like the very best had all these traits. Yeah. But um if you are really good at understanding the business side, that's a skill that's highly amplified.

Anders Arpteg 41:11

Um, that's that's I think that's so interesting and uh very profound, I would say, to have people that understand the domain, so to speak, the business, and also perhaps people that can communicate in some sense. Oh, yeah, right? So you can have these kind of people that before wrote really, really beautiful and complicated algorithms and code, but they may not be the best in understanding the need on the business and and be able to work in teams to communicate properly. They may not be the most relevant ones in the future.

SPEAKER_07 41:43

No, right. And I I know uh we help a lot of companies start to measure the productivity in the teams. And one of the most common reflections back in the days is like, ah, we don't want to measure velocity because the most important part is what they build. Like, are we building the right things?

Anders Arpteg 42:06

The right things and the quality of that, yes.

SPEAKER_07 42:08

And I certainly agree that that's true. Like, it doesn't matter how fast you build something if you build the wrong thing. However, nowadays, sadly, the software engineer doesn't choose what to build. They're getting tickets assigned to themselves. Of course, there are things they can choose within that, but it's extremely common, especially in larger companies, that you have a clear specification what's going to be built. And in that case, it's not the software engineer by themselves who choose the direction where to go. That group, I mean that group, if you're really good at choosing the direction, might should probably change workplace. Because that's a new skill that's going to be so amplified if you're working, for example, in startup and you can go a little bit more nuts on your own.

Anders Arpteg 43:03

So if you have a way of working where you give very specific demands, do this. Yeah. That could not perhaps be the best working environment for a person who wants to choose.

SPEAKER_07 43:17

No. No.

Anders Arpteg 43:18

Yeah. And perhaps that's better suited to AI to do.

SPEAKER_07 43:22

Yeah, certainly.

Anders Arpteg 43:22

Yeah.

SPEAKER_07 43:23

They will use AI a lot, hopefully.

Anders Arpteg 43:27

I think I think you mentioned also the the No, no, sorry. Um you mentioned uh something that we mentioned here in the pod a couple of times, also uh the Jebon's paradox, uh, meaning you know, a lot of people are a bit afraid that AI will replace software engineers, um, and then perhaps they will not hire as many software engineers, but at least to what I've seen, it's the opposite that we're seeing right now. I can also mention that uh but that is very interesting.

Goran Cvetanovski 43:54

Uh the question, because uh we were speaking with Lee uh um about this and the same thing.

Anders Arpteg 44:00

That is true what you're saying, but what are they hiring and who and which levels are it may change a bit, you know, who they're hiring and what they are working with, but the need for engineers is still very high. They have this, you know, Andrew Carparthes is a very famous person in AI, and he he actually published um a web page uh called something like the US Job Market Uh Visualizer or something. And he takes statistics from the US uh Bureau of Statistics or something, uh job market, and then he put it um and tries to see how will AI impact like 300 different types of jobs. Wow. And he can see some jobs, you know, uh they're really uh negatively impacted, like um receptionists or even customer support potentially. But actually, if you look at software developers, uh they are actually positively impacted, which is kind of surprising to a lot of people.

SPEAKER_07 44:54

I'm happy to hear that. It feels good.

Goran Cvetanovski 44:57

Yeah, but uh the there is a research um that is looking at uh if you look at the amount of people higher than uh let go in the past couple of years, is actually pretty much stable. However, there is a restructuring of which type of uh level companies hire, which is actually aligned with what you said before. So, like hire a senior person. Um, most of the companies are hiring a senior senior person, and then you ask yourself why they're they're hiring senior persons, because if you are doing AI uh uh coding with uh let's say uh with the claude or whatever it is, you need to have a senior person who understands if the cloud has done any faults or not, um, which makes mostly of the organization a lot of organizations to um change the organization of the teams from a little bit more pyramid shape to a diamond shape. Pyramid shape means like uh you have on top a senior who is living, uh the same is happening in the legal uh uh industry right now. The the former pyramid was like you have a senior who is running, let's say, five, ten uh juniors, right? But now, as you have like AI and etc., uh you hire less juniors, you hire uh good seniors that have basically agents or they can actually clothe with agents, and then you have a couple of seniors that you uh juniors that you want to then promote so they become seniors and etc. Uh, how do you see? I mean, you have firsthand insight in uh how people hire, what do they want, and et cetera. Does this make like um sense, or do you have like a completely different um picture in mind when it comes to reshaping of let's make it very simple? How do you see the organization of teams today?

SPEAKER_07 46:55

First of all, like I'm zero worried that it's gonna be few engineers. Um our finance department are building products today, ink lawed code. They never wrote the single code before line of code before. Our recruiters are building tools. Uh, and these are cases where business are going into the engineering side. We're also gonna see how the engineers are going into the business side. If you're working on a standalone product, I think for quite some long time, um yeah, like in those cases, like it's an internal tool, 100% that those are gonna be built by the power users, they know the context so well, so they will be able to build it so quickly. However, there are gonna be applications used by many other users, etc., where there are other complexities where it's probably gonna be the other side where the engineering team goes more towards product, where it's easier to find the engineers that have the product sense or the business sense than teaching the business side, the engineering problems.

Anders Arpteg 48:23

It's like who should upscale? Should it be the domain experts or the engineers? Oh, yeah, and probably both. Yeah.

Goran Cvetanovski 48:28

Yeah, right. I'm not sure whether we answered that question, but uh where do you see most of this? Uh is it more seniors that are getting hired or more juniors? Do you see any shifts in the teams of your customers?

SPEAKER_07 48:41

Like uh I I I I don't have a helicopter perspective on that since we only have senior engineers. All right. Um and there's yes, there's an increase in demand for senior engineers for sure. Yes, right now.

Anders Arpteg 48:57

But getting back to the Jebon's paradox, I don't think we explained it properly uh really. And and we got a bit into it saying if you have uh the finance department now building things, yeah, they actually become engineers in some way, right? Meaning you have more engineers, probably. Or or how uh you know about the how would you explain it?

Data Engineering Demand And Moats

SPEAKER_07 49:20

I I I think the best example that I've come across is from my grandma in her diary. When she was 15 years old, she wrote it was a page where she's telling that she met this friend, and in their living room they had two lamps. And basically, what she wanted to say, but not write, was like they're so rich because they have two lamps in the same room.

SPEAKER_04 49:49

Right.

SPEAKER_07 49:49

Then the price of electricity went down, uh the price of lamps went down a lot, and the result of you ending up with lamps everywhere. Um and the uses the usage of electricity has just gone up, even though the price of it has um increased. Yeah, it yeah. And we we see we saw it was a huge worry in during the industry uh revolution revolution revolution when the turbines got twice as more productive and they were so worried that now they're just gonna buy half of them. But instead, since they got so much more productive, they increased the usage. So they sold even more. Um AI has increased the value of an engineer so much that I mean personally, I just want to have more of them, and that's what I'm feeling most companies are thinking.

Anders Arpteg 50:59

I think it's it's very true. Um just to uh to use an example of this, if we take the concept of uh on-demand software, uh this was actually from Sam Altman. Yeah, he mentioned you know, there there may be a point in the future where we do not develop software as we do today. Today we have software development, meaning we define, you know, we need to have this problem, we set up a team working on something, or even a company building some piece of software, and then a year later or a month later, you have something that you release, yeah, and then you continue to build on it. But it may be a point where you simply say, I right now, for this minute, need a software to bake a piece of pie. When I ask an AI, build a piece of software for me, yeah, I use it and I throw it away. So it's never even reused ever again. So it's a one-off kind of software, yeah, it's on-demand software. Uh-huh. And if you think about that type of future where software is so easy and cheap to produce, then of course the demand will go up a lot. And if you think about, you know, if you can change in Excel or in some kind of horrible SP product, um, to say that I can actually change the interface or the functionality, or just add this kind of automation to the software by simply telling the software, please add this functionality, which is software engineering, yeah. Then of course the increase of engineering needs will increase a lot because it's so easy to do.

SPEAKER_07 52:34

Oh, yeah. And talking about terrible SAP products. Um recently a colleague and I we were going on uh to Amsterdam for for a conference. And we were standing in the line for checking in our luggage, and you know how you're standing there and just typing and typing and typing, you're like, what are they doing? Like they're checking in so many people. Why isn't just a button to do whatever they want to do? I wanted so badly asking what they're doing, but since I said I'm in Swede, I didn't. However, when we came to the hotel, exactly the same thing happened. We were asking for the room, and he was um we asked him, like, would it be okay if we take a step and look at your screen? And he's like, Well, I'm I'm can ensure you that I'm doing this as fast as I can. And then I told him, like, we're actually working with these kind of products, and I want to understand what it looks like. We are software engineers, yeah. And it does let us go around and look, even though before it says like if you see any like personal data, you need to erase it from your mind because it might be sensitive, so really GDPR compliant. And while we're standing there, like I am I'm blown away about how standardized this product is. Just to give a few examples, like it was one like pop-up that came up asking, is this person smoking or not? And he pressed like yes immediately. And I asked him, like, how do we how do you know we're not smoking? And he's like, Well, it doesn't matter because every room in our hotel chain is a non-smoking room.

Goran Cvetanovski 54:30

So why do we have to do that?

SPEAKER_07 54:31

Yeah. And he also needs to say check in a separate system to see if the room is clean or not. So it doesn't give a tick to uh the key to a room that isn't clean. And he's doing this back and forth and like and um it it took forever. If AI, if if and if that this was today, I think they're actually approaching a stage where they're not buying this hospitality tool from SAP, which every other hotel is doing, and um they will build it themselves, and the person that should be worried about maybe losing the job is sadly not the engineer that's gonna rebuild the system, but the person in the reception. Not having to tick that box. And and it's fun, I told this story.

Goran Cvetanovski 55:32

I don't think so. I think it's like you still need uh a human person that will welcome you there, no? Yeah, maybe maybe not.

SPEAKER_07 55:39

Gladly, like if it takes one minute instead of three, and like hi Victor, here's the key. Yes, uh, welcome.

Goran Cvetanovski 55:46

So happy to have you.

SPEAKER_07 55:47

Yeah.

Goran Cvetanovski 55:48

That's it. If you need something, just let me know.

SPEAKER_07 55:51

No. And since it was an agent making that um making that um booking, yeah, the agent could also provide so much more information. Yeah, the agent would probably know that we were on a business trip and it could be nice to have separate beds. They would the agent would probably ask what flight are you coming with to make sure that the room is actually available at that time. Imagine if I landed and I got a text from the hotel. Your room is welcome to welcome to Amsterdam. Your room is ready, you can go straight to their port from their part.

SPEAKER_00 56:32

Yes.

Anders Arpteg 56:32

Do you think it's really important in the coming bar you know that we're building and to have a human standing there uh serving beers and definitely?

Goran Cvetanovski 56:40

Oh yeah, definitely how we're gonna get tip otherwise what we're talking about. No, but it's like I think that we are talking. I have two uh comments on this. One, I think that we are uh we are taking are we thinking short term or long term about the engineers being more requested? Because at some point of time we're gonna build all the products that are need to be built.

Anders Arpteg 57:04

No, I don't think so.

Goran Cvetanovski 57:06

But it's it's it's a theory, right? Um, probably you're right, maybe I'm wrong, maybe I'm not.

Anders Arpteg 57:11

Because you can adapt anything. I think you know, if SAMP you know instead had functionality in the software that said, I want to change this, so the receptionist could be the engineer.

Goran Cvetanovski 57:22

I don't think that SAP is a problem. I think it's the the the management of that uh company that is the problem because they need to probably um so hotels are right now connected to they must they they need to report to the government who is in the hotel, right? So that is like there are different systems that are connected to their system, and usually I don't think that it is even SAP, probably something that is built by hand by somebody internally or whatever it is, and then they're using it for the whole chain. Maybe it's based on on SAP. Um, so I yeah, I I'm not very sure that SAP is there to fail. But my point was I think that we are in a convenient stage where both for right now we are utilizing AI and everything, so we want to have more engineers because we want to have more productivity, we want to have more gains, more value, more everything. So bring more engineers in place. But at some point of time, uh there will need to be a rebalance. Like, okay, now we achieve what we want. What happens then? Do you continue or not? The second thing is the convenience you said, like about like um, for example, this is very interesting. You go to Dubai, right? It's completely different uh world somehow. And we are used to not uh the doors are opening themselves in Sweden, right? They're closing by themselves. You click on the button for the elevator, elevator goes up, and etc. Not in Dubai, you go there, there is a person sitting installed in front of the elevator. They the uh good evening, sir. Uh where are you going? Up, great, puts basically uh he presses the button for you, right? Uh, you're exiting, they're taking your bag, everything. So you this convenience. I don't know if you can automate. Maybe you can do it with a robot, but just somebody being there. I was now in uh I think that one two weeks before you went. Yeah, um, I was there for uh 10 days, and there was a guy who was uh working the entire night just clicking the button on the elevator.

SPEAKER_07 59:24

That can be automated, I know.

Goran Cvetanovski 59:25

No, but it's already automated. I can press the button, but just the the luxury of that happening, yeah, but this is the experience. Okay, it feels wrong, but that is what it is. After 10 days, you know the guys like hey Michael, how is the uh how was the night? Okay, serve up and down. Yeah, so would you replace it?

SPEAKER_07 59:50

Yeah, that would that I would like to replace, yes. Yes, probably nice experience for you, but I hope it's not if it's a good experience or not.

Goran Cvetanovski 59:59

I think that it's basically we will need to have a little bit more human touch in a future.

Anders Arpteg 1:00:03

So what about like a self-driving you know taxi? Uh and I mean do you actually completely automate that one and not an elevator?

Goran Cvetanovski 1:00:11

Uh no, elevators are already automated. I can press my own button. I'm just saying that the convenience makes it a luxury, and some of the things we will like to still have a person that is welcome you. So you will not replace the person.

Anders Arpteg 1:00:25

Situations where you prefer not to have a human there because you wanted privacy, right? Oh, yeah.

SPEAKER_00 1:00:30

Absolutely.

Anders Arpteg 1:00:30

Yeah.

SPEAKER_00 1:00:31

Absolute.

SPEAKER_01 1:00:41

It's time for AI News, brought to you by AIW Podcast.

Anders Arpteg 1:00:48

Yes, so we should usually have a small break. Not so small sometimes, but still a small break before we continue discussion here with Victor. Yeah. And yeah, speak about the recent AI news. Uh, Victor, do you have anything that you read about uh that happened in recent weeks that you'd like to bring up?

SPEAKER_07 1:01:07

Well, it's usually information overload when it comes to AI news. Ah, not not nothing top of mind. I know we spoke something just before I came, but uh you go ahead first if you have one.

Anders Arpteg 1:01:24

I might have a couple of them. Uh but before that, Goran, do you have something you want to or should I just go for it?

Goran Cvetanovski 1:01:30

Yeah, you can go for it. I think that I mean the most interesting is uh the for me at least, like the mask move in the line cursor. Yes. But you're gonna talk about it because it's a fan of the mood. So then let's say that Meta just released a new image model. So which you mean open AI? Open AI, yes. Um, sorry. Which uh I'm looking forward to explore a little bit more, especially now when I cancel the mid-journey um account. I kind of miss it, but it is what it is.

SPEAKER_07 1:01:58

Maybe it can do slides now and we're completely outdated.

Anders Arpteg 1:02:01

Yes. Right.

Goran Cvetanovski 1:02:10

I mean I used to work with 11 labs, uh, just very 11 labs to create this uh you know announcements for events, and now I just went to to uh studio of uh Google and uh and you just uh do the voices. You have different voices, you have all the parameters, so things are changing very quickly. But go, I think the biggest one is Elon Musk, so go.

Anders Arpteg 1:02:35

Yeah, as usual. No, but uh not speaking about Terrafab this time, but uh now suddenly um XAI, one of the Elon Musk companies owned by SpaceX, uh, since SpaceX acquired XAI some uh weeks ago or months ago. But they are now exploring collaboration. We'll see if it stays at a collaboration or an acquisition uh of cursor. Now, also recently they've been speaking about like um uh a triple kind of collaboration here with Mistral and Cursor and XAI. It's just today, so but still, this is very interesting, and you can think a lot about you know why this is happening. So, okay, Elon and XAI. Uh of course they have Grok and they have a lot of different things, like uh they have the macro hard company and trying to automate all the cool stuff you can do with uh operating systems and office suites and uh automating the the workflows that you're having. But we haven't really seen a big improvement in Grok for some time, and we know that they fire a lot of people in XAI. It could be because of the acquisition from SpaceX, and they simply replaced some of the management people there, but it could also be, and I'm just speculating, that there is Elon Musk is very annoyed that they haven't seen the progress that they would like to in Grok. And now um then Cursor, of course, they have their own model as well, this whole composer, and um it works really well for coding. And uh cursors, of course, have all the scaffolding around the model to really make it work for coding in a way that very few others can. And they also have moved, of course, into this kind of agentic future where the latest uh cursor 3, I guess, this uh is having this kind of new agency kind of uh overview where you simply start to work with a set of agents in parallel and stuff. So, I mean, I I think potentially they're trying to leapfrog into the future and make XAI the frontier AI lab that he really wants it to be. But now that he's also speaking about you know Mistral, I was a bit I don't know, scared almost. First, I was scared about that. Oh, is he's going going to acquire Mistral if he acquires Cursor, which he could, he placed a bid, I think, on 60 billion dollars for it, uh, or a 10 billion dollar partnership. Um but if he's do that doing that in Mistral, then I would be afraid. But I don't think he Mistral would ever ever sell out to that, but potentially collaboration. Okay, so a lot of speculation here, but very interesting moves. Uh, and you can try to guess you know why XAI is doing this. I think for one, of course, they want to leapfrog very, very quickly into becoming a frontier AI lab, and then having Cursor and having Mistral on board, either in a partnership or an acquisition, would you know potentially make that happen? What do you think?

Goran Cvetanovski 1:05:37

Well, uh, if you want to go ahead, please go ahead. Well, I think uh there are two uh one question that you started in the beginning is like, is the fight to get like uh who is going to be the predominant model over everyone? Um is that the game? Because otherwise everybody can live in their ecosystems, right? So uh yes, XAI. Will need their own model so they can work in their own way and stuff like that. Or is that comp or is it a competition for a customer demand? I think that we saw a lot of things that didn't happen. We were expecting OpenAI to go commercial and do ads and everything. That didn't happen. They tried it, it didn't work. I think that uh things are changing every year. So now you have like uh uh Anthropic is right now on the top, right? Last year we were talking about Google, before that, we were talking about Open R AI. So you don't even know who is going to be the winner, is because it's uh the the model defines the winner of the moment at this point in time, at least currently. Yeah, so I don't know. I I yes, they grok is great, but I don't think that it did so much progress on it, is not delivering so much money. Um there are many things, but Mistral, for sure it's gonna be bought, but is it gonna be European or somebody else?

Anders Arpteg 1:06:56

I don't think France the French government would ever the assault system can buy them.

Goran Cvetanovski 1:07:01

The assault system the assault system is one of the biggest companies in the world, it's a French company, it's uh very, very known uh uh in France. They do all the engineering, they do a lot of defense systems and everything else. They also have their own Google, uh they also have their own cloud that they're utilizing that they're just now um making it more public. There is there is buyers in Europe if you want to buy them. Yeah, but I don't think Mistral would sell out though. No, I think that it has also a very good backup from uh from the French government. So yeah, yeah. So I think it's becoming a little bit more like a national treasure, and yes, yeah.

Anders Arpteg 1:07:38

So okay, but very interesting move, and uh it's just the the continued extreme like AI race, you know, between the frontier labs is just exploding, I think.

SPEAKER_07 1:07:50

It's exciting.

Anders Arpteg 1:07:52

I think one uh other potential interesting uh news is you know the mythos model from Anthropic, yeah. That is the next level from above Opus, so to speak, that was not released because of the potential dangers of cybersecurity and whatnot. Now, someone actually reverse engineered it. And they tried to understand you know how does mythos really work. And it's called Open Mythos, and they published a bit what they think you know mythos really does. And uh, I've been a long spokesperson for uh latency based reasoning and not continuing this kind of auto-regressive kind of token reasoning uh version that we're seeing in the current LLMs, and actually what they say in Open Mythos is that they do exactly that. So they actually instead of just doing a forward pass through all the layers, they have a much smaller model and then they like loop 16 times in the same layer in the latent space to do the reasoning in latent space without producing any tokens at all, and obviously that must be much more efficient than having to go through the whole model every time for every token to do the reasoning. So I think you know it's just one more sign that we're moving in. This is the Jan Likun kind of direction with Jeppa models, etc. And uh potentially this is actually what Mythos is doing. We know also it's it's potentially 5x more expensive, but perhaps not because of the number of parameters, it's just because it's looping so much internally in the model that makes the compute really expensive.

SPEAKER_07 1:09:30

Interesting. Will they release it soon again?

Anders Arpteg 1:09:33

I don't think Anthropic ever will, but Open Mythos is released. You know, that's an open source version of someone at least guessing what mythos really is. Exciting. That's really cool.

Goran Cvetanovski 1:09:43

If I can add one more very short one, because I think it's good to tie it with what our discussion is. This was actually uh just uh released by um Computer Sweden. It says 75% of all the code in Google right now is AI generated, which is 50% more than last year. So everything is basically um AI generated by Gemini at this point of time.

SPEAKER_07 1:10:10

Coming back to our discussion, like surprisingly low. Uh yeah, 75%. 75%?

Anders Arpteg 1:10:17

Yeah, but anthropic is higher. I think they said above 80% in anthropic at least.

Goran Cvetanovski 1:10:22

And how much higher in Google in anthropic the last that you discuss all these things uh in the past uh six months? Sorry, what how much how much higher? New higher.

Anders Arpteg 1:10:31

Yeah, yeah. It's still a lot. Anthropic is really increasing. Yeah, they're not removing engineers, so it's kind of interesting. But also, Google had the big um next event or conference, right? So this is actually, I think, where this is coming from some. So that was one of the announcements that they they of course uh are using AI to generate the code, but still you have a human that tells the AI what to do.

Goran Cvetanovski 1:10:56

Yeah, but it's always when you announce something like that, you need to know that somebody is selling something. So you never know because I mean, what is the the pet beef of you saying like we internally use 75% is also like hey, this is stable. If we can use it, you can use it.

Anders Arpteg 1:11:12

So I still think it it is you should really interpret this properly. Mean saying that 70 or even if it's 90% of the code that's generated, it doesn't mean that you don't have engineers. Even they themselves say engineers are moving to work or moving towards orchestrating agents, they're not removed from the loop, so to speak. They they're just orchestrating agents rather than writing lines of code.

Goran Cvetanovski 1:11:37

Yeah, but what happens at some point of time where we have seen that this is stable now? We don't need to monitor anymore. What happens then?

Anders Arpteg 1:11:45

Well, you still at some point need to tell it where which direction to go.

Goran Cvetanovski 1:11:48

But at some point of time you have perfected, it's a Mona Lisa already, okay? You don't need to touch it anymore.

SPEAKER_07 1:11:53

You couldn't, yeah. I agree with you that they probably don't need to review the code, but it's still gonna be decisions that require business context, etc.

Anders Arpteg 1:12:06

Um you could have a point of singularity, which is the recursive self-improvement point where humans are completely cut out of the loop, so to speak. Yeah, yeah. But that I think will not probably be wanted to, or or we will take a little long time.

Goran Cvetanovski 1:12:20

Or we bet on it, which year is gonna happen so we can continue with the tradition.

Anders Arpteg 1:12:26

If you want to lose one more bet, then I'm open for it, Coran.

Goran Cvetanovski 1:12:28

I have never lost that. Stop spreading these rumors around.

Anders Arpteg 1:12:33

I actually lost the more bets than I bought, so yeah, usually right, Goran. Uh still, but I think you know, even them themselves said, you know, engineers is moving more to work towards orchestrating agents, and of course, that's what we're going to see in coming time as well. I think also they they they built they released something um in Google Next, uh, the Enterprise something. What was it? Enterprise agent platform or something. So of course, I mean it uh it's nothing really revolutionary. I would have expected something more really revolutionary to be announced. But at least they have this kind of enterprise agent platform thing where you can do the full life cycle of agents building it, running it, and everything uh in a more holistic way. So, yeah, of course, that's that's cool. Yeah, but nothing really revolutionary.

SPEAKER_00 1:13:25

It's been a quiet week.

Goran Cvetanovski 1:13:26

I mean a big week, but still a little bit quiet. Nothing, uh no Kardashians this time.

SPEAKER_07 1:13:32

I I came a it's not a big news, but I came across a case study from Stanford University on how token spent correlates with productivity. Um I would argue before I was always like, oh my god, you spend so much tokens, like wow. Um and really making an equal sign like a positive, yeah, yeah, like oh, a lot of tokens spent much meaning you're really, really productive. But yeah, they did see some correlation, but it was surprisingly weak after some time. Um just because you're spending more tokens doesn't mean that you're more productive. But with that said, there are a few individuals that are spending crazy amount of tokens that are extremely productive, uh, and there there is like a productivity gain when you're like, for example, spending up towards$2,000 a month in tokens, but then it's doesn't necessarily mean it's not a strong correlation.

Goran Cvetanovski 1:14:46

Exactly. There is a I I heard a story. Let's leave it like that. And the story is like um, we will not name companies and everything else, but companies right now are burning the entire total budget that they have dedicated to AI within two, three months because of this, right? Yeah, so now you're coming to a question is that is the company first, you need to have a company that actually knows what it's doing, and then they are pushing for this is the incentive, right? More you you use tokens, you're actually this is great. But what do you do in a traditional company that actually doesn't understand the value of the token or the um uh understand uh what is called, or you cannot measure the productivity per token. Um, and you ended up burning a lot of the budget that you have. So most of the companies at that point of time they will say, like, okay, we burned all the budget, there is no more recruitment, that's it. Right? And this is a true story, actually. This is what happened, right? So it's the two polarities between companies that are uh incentivizing uh token utilization because they can be more productive. Engineering, I can understand that, but then you have traditional companies. How is that?

SPEAKER_07 1:16:04

Yeah, it's gonna be super interesting to understand like where is this token spend gonna be put.

Anders Arpteg 1:16:13

You mentioned, Victor, a bit about the uh you know the recent uh advances in AI could potentially cause a big boom in data engineering and like business intelligence needs. Can you just elaborate a bit more what you meant with that?

SPEAKER_07 1:16:29

Yeah, well, um first of all, like recently we have seen an increase in pretty much all areas. Um but the areas you're mentioning now around data have definitely increased uh significantly, like triple digits. And you can see that proxify. Yes, absolutely. It's a it's a massive increase. I can imagine a lot of companies are thinking about like okay, what's gonna be our moat in the product, for example. If everyone can build this, what can we have that no one else has had?

Anders Arpteg 1:17:09

Right.

SPEAKER_07 1:17:10

And data is something that not all models know. So, for example, our Eros model, yes, you can build the infrastructure, but it's gonna be impossible to know the weights in the model, etc. It's like to know what engineer is the best one. If you don't that's not public knowledge, no. So I can imagine that's uh of interest among many country companies. Like, how can we how can we increase um how can we increase our position here for improve our position here? And usually they come to us asking for someone that can like build the models, but they rather need someone that needs to structure the data, so they end up with some data engineer.

Anders Arpteg 1:18:00

Oh, interesting. Um that makes a lot of sense, of course. Uh interesting to hear. Well, what are the top demands? Did you have like a top list of type of engineers that you're getting most requests from?

SPEAKER_07 1:18:11

Yeah, it's still a lot of like traditional software engineering roles, definitely.

Anders Arpteg 1:18:17

Like backend engineering, yes, absolutely.

SPEAKER_07 1:18:19

Yeah. Python is up a lot, I think, um, as an example. TypeScript, your favorite, is up. Also like PHP is really PHP. Yeah, yeah. It's certainly like on a um also increasing. And it's also surprisingly hard to find talented PHP engineers because many of the new ones don't want to do PHP. They prefer to do Node as an example.

Anders Arpteg 1:18:53

Oh, people still asking for COBOL engineers as well.

SPEAKER_07 1:18:57

Yeah. And legacy, like if you're a new engineer, actually you're gonna get better paid if you go with a legacy programming language. We can see in our Yeah, yeah. And uh that's probably the most extreme case. But I would never go that far.

Goran Cvetanovski 1:19:18

All the best things.

SPEAKER_07 1:19:19

But like uh uh the engineers working with.NET as an example, always quite mature companies asking for those. Right. Longer engagements, better paid, maybe not as fun projects in some cases, more legacy, but definitely better pay. Interesting.

Anders Arpteg 1:19:45

So that that type of you know competence then is is really high. When it comes to seniority versus junior, do you see uh increase in seniority demands as well?

SPEAKER_07 1:19:57

Or since we only have seniors or around seniors, it it's hard to say.

Goran Cvetanovski 1:20:06

But maybe there is a correlation. If your business is going greater because it's senior-based, that means more seniors are recruited.

SPEAKER_07 1:20:13

Yeah, fingers crossed. Yeah.

Goran Cvetanovski 1:20:14

Yeah.

SPEAKER_07 1:20:15

But what we do see an increase is that the clients are asking for like more like product taste and that they want them to have more uh a PM role, the engineers. So we can see how they were asking the engineers to go more towards business and products. Yeah, that's for certain. Yes. And it's also very common that they're asking for heavy users within within AI.

Anders Arpteg 1:20:46

Okay, so AI is built. Yes.

SPEAKER_07 1:20:49

Yes.

Anders Arpteg 1:20:50

These are super interesting stats.

Developer Experience And AI Slop

Goran Cvetanovski 1:20:52

You know, you should publish this or something, make a report like that, uh the annual Proxify report card, like uh yeah, it will make sense. I attended uh before uh I attended one of your breakfast workshops that are very cherished and they're becoming very popular now. You cannot even find places in the breakfast that they are doing in their office. Uh kudos to Paolo and the team there for making a great stuff. But at that point of time, you were presenting something about how to make um uh how to make uh developers more productive, and it didn't have anything to do with AI, it was more about the equipment and the hardware, software, and the way how they should work. Maybe um a good topic to talk about.

SPEAKER_07 1:21:38

Oh, gladly, gladly. Uh I'm surprised how much you can drive productivity in an engineering team. And if we look at many American clients, it's very common that you're even having a product and like a platform engineering team. I know you do it like Spotify. You you're like the only company that I'm know I'm aware of in Sweden that are so heavily focused on the developer experience. If you're having a team focusing on this dedicatedly, like not just like a DevOps doing a little bit, etc., I think you can have massive productivity gains. But even looking at the developer experience a little bit could dramatically in your productivity. And what I showed during this breakfast event was just a list of things that we have seen has a positive impact to productivity. And I will a few examples. Like having a policy, how you turn off notifications on your computer so you're not disturbed during your flow. Having a large display, or maybe two, that is much better than working from your laptop. Very controversial to say, but that that's what we what we have seen. And quick turnarounds on pull requests, reviews. There are a lot of things on that on that list. Yes. That even with like short-term gains.

Goran Cvetanovski 1:23:26

But that is interesting. We usually talk about AI and software all the time, but we're not looking at the the environment and uh the experience that you're giving them around the computer. No, so that was very interesting for me. It was a little bit refreshing.

SPEAKER_07 1:23:41

And what what we're adding now a lot of to that list is how you can work with the workflows. Um I'm certainly not an expert here, but like how can you work with the workflows with agents, etc. Having one reviewing the code you have written, having automated tests much, much more. How could you make sure that the that your agent is getting more feedback, a feedback loop?

Anders Arpteg 1:24:10

How should there is a risk of AI slop, right? Meaning that you could have people that simply ask very casually, this prompt fix that, and then suddenly doesn't even review the code and just push it to production. Oh yeah, right? So I mean, how do you really fix for that? Do you have any thoughts about you know how to avoid AI slop and these kind of problems that can occur when it's so easy to have AI generating code?

SPEAKER_07 1:24:43

I know I don't have a good answer for that, sadly. But it's definitely gonna be an issue. And I'm uh I don't think it's gonna be a case where the receptionist is tweaking the system that's gonna be used by the entire hotel. Um as an example.

Anders Arpteg 1:25:04

Well, if you simply tell a system, let's say you are a receptionist, and you're standing there and you can see if I just remove this stupid question about smoking or not smoking, yeah.

SPEAKER_07 1:25:19

Yeah, that that's gonna be a great case. Yeah, however, there are also gonna be a lot of other cases that are not as good. And giving the freedom for like imagine a case where anyone could request a product change. If we took every product change that came into our request box, our we would have so many features.

Goran Cvetanovski 1:25:44

Yeah, yeah. But it's also like the one of the whole points with having systems is actually to have a process. And if everybody changes the process, then you have anarchy, you don't have any process. Then what do you build upon?

Anders Arpteg 1:25:55

I guess we need an AI to prioritize in the request. So you can send the request, but um it needs to be evaluated and prioritized, yeah. Which is actually what the L is doing in Tesla companies, by the way. So anyone can push ideas and then they have an AI that is evaluating and prioritizing the calls.

SPEAKER_00 1:26:14

Yeah, okay, that makes sense. I mean, yes.

Future Skills Everyone As Engineer

Anders Arpteg 1:26:19

Cool. Um and if we were to to move a bit more, the time is flying by here. So so if you move a bit more to the philosophical domain, I I know you you're not a favorite uh of that, uh, but but still, if we try to think about engineers as a profession in general, and we know at least it's moving from writing lines of code to some kind of more abstract level of orchestrating agents or telling agents what to do. How how what kind of skills do you see people you know should have? If you do you have kids, by the way?

SPEAKER_07 1:26:58

I do. May I ask how old they are? Uh they are four and two.

Anders Arpteg 1:27:04

Okay, still very young.

SPEAKER_07 1:27:05

Yeah.

Anders Arpteg 1:27:05

But still, at some point you need to guide them into what kind of education they should have. What would be your advice for them to educate themselves?

SPEAKER_07 1:27:16

Yeah, that that it's a scary question, for sure. Because they're never gonna be smarter than an AI model.

Anders Arpteg 1:27:25

Yeah.

SPEAKER_07 1:27:29

I um it's gonna be very, very it's hard to give career advice for that.

Anders Arpteg 1:27:37

You will have to make that decision soon, right?

SPEAKER_07 1:27:39

Yeah.

Goran Cvetanovski 1:27:40

They will not listen. No.

Anders Arpteg 1:27:42

I tried that they don't listen if that's better. But okay, so if we were to choose, I mean we we can see a number of paths here. Either you choose to be like a super specialist, like I know everything there is about a car engine, or I know everything there is to know about how. Hotel receptionist skills. Or you can say that humans should rather move into a more generalist kind of understanding, meaning you instead like educate yourself in the STEM fields, in the science and engineering and technology fields, and you become more of a generalist, understanding how the world of the physics works, etc. Yeah. Do you have any thoughts there? I mean we can see both being interesting here. We just spoke about you know senior people being interested. I mean, you hire mainly senior people, meaning they probably have a lot of you know special skills, yeah. Like a T-shaped kind of uh expertise, right? Where they have at least some kind of depth.

SPEAKER_07 1:28:48

And you should just skip the junior step and go straight to senior. That's gonna be my advice.

Goran Cvetanovski 1:28:54

And what will the juniors then do? Yeah, yeah. They do startups, I guess.

SPEAKER_07 1:29:00

Yeah. I mean the startup area, like um as that's still gonna be an area, hopefully. I I'm so surprised as societies become richer and richer, like when the GDP uh goes up by like two percent each year. Like, what are you gonna do with the money? And you're so surprised what kind of ideas that pops up, and that's just gonna continue.

SPEAKER_00 1:29:26

Yeah.

SPEAKER_07 1:29:27

I was like maybe pressing the elevator button.

Goran Cvetanovski 1:29:31

Yes, for go right. Because I think more we get richer, more convenience we are searching for, and more luxury in life, you need to fill that void of you being rich. Um but I don't know. I I'm not very confident that everything will change so much that people will not be able to. I mean, I think engineering in the future will be very important. We're not talking about engineering like uh software, we're talking about hardware, we're talking about chemical engineering, process engineering. Uh, and even now, I think that we are in pain of such uh uh skill set and a workforce in every country, uh especially now when everything is getting actually again de-globalized and et cetera. I think industries will go up. So we will need to look more for STEM people. We will need to have also a lot of human uh touch to everything that we do. I think that when everything gets automated, people will search for that human touch in something. Um, so I'm not very uh doomsday oriented. I think it's I'm more pro-topian that things will happen gradually and we will actually evolve with them as we go. But right now I don't see the the big thing is an important thing. Especially you said it. If everybody starts being a coder because they have a lovable or whatnot, wipe coping or Claudin, etc., then yeah, then everybody just acquired a new skill.

Anders Arpteg 1:30:59

So I think everyone will be an engineer actually in the future, and I think it's a really good thing because everyone can build something in the future. That that would be awesome. Then there will still be people that that do work with you know assembly language and machine code because they need to at some point, and there will be these kind of specialists, but the ability that AI gives to allow anyone to become an engineer, I think, will be a really beautiful future, enabling anyone without knowing how what TypeScript is or COBOL is or whatnot, to still build stuff. Yes.

SPEAKER_07 1:31:31

I I love that with Lego, like it's so easy to enter Lego, even a four-year-old can. Yeah. But there are some people that are really good at Lego.

Goran Cvetanovski 1:31:42

Yeah, yes, exactly. Exactly.

Anders Arpteg 1:31:46

I mean, it's like uh I wanted to move into music. I tried even to take singing lessons, you know, and Goran, you're an expert singer, but I tried and failed, but with AI I can.

Goran Cvetanovski 1:31:55

So uh no, but we have been arguing, and there's a lot of discussion about like uh well, you know, you're making uh songs in Suno, that is not real music. Yeah, right. Um uh is it coding if you're uh coding with AI? Yeah, it's engineering it absolutely. Yeah, if people are if people are building uh a product with lovable, but they don't know how to code uh is that and if that product is still making money, is that actually bad or not? No, I think that this technology is giving the people that were helpless in uh establishing their dreams. So, for example, in music, you can be a good lyricist and you know approximately how the song should look like, but you're not you're very good in uh uh in in um arranging and production, and then you have a tool to do that, and suddenly your ideas can be in life and you can show it to Andersh and say, like, hey, listen to the song that I did. This is actually dedicated to you, Anders, uh, my best buddy, and whatever it is. It's such a beautiful thing, and I think that uh in this future of everybody becoming an engineer, it's a it's a it's a beautiful thing because we will see product that we have never seen before. Oh, yeah, because the the ideas was reserved to the ones that were capable in deploying. Definitely and now everybody can deploy, so it's such a good thing. Yeah, um, Victor, is there any topic that you would like to uh put uh that we have missed uh during the one hour 44 hours or no one and a half an hour?

SPEAKER_07 1:33:22

I think we cover a lot. I'm very content.

Anders Arpteg 1:33:26

Yeah, one more if we if we go a bit um doomsday issue as well, and we spoke about the recursive self-improvement, meaning there may be a point where you know AI is just keeping improving itself. And uh, if we take an example of a Tesla car or whatnot, and and uh today, you know, you must have to tell improve that kind of water path or improve that and fix that, but at some point the Tesla car may be actually improving itself without a human being involved, uh involved at all.

SPEAKER_07 1:33:59

Absolutely, and I think it's quite near term where you give a very broad description to an AI like improve the profitability of our business.

Anders Arpteg 1:34:15

Yeah, right, a single prompt. Yeah, and then it that's a whole company, right?

SPEAKER_07 1:34:19

Yeah, pretty much. And I can I can imagine that's not that far away.

Anders Arpteg 1:34:27

Yeah.

SPEAKER_07 1:34:28

Where it's then like pushing commits and deploying marketing campaigns and uh recruitment and everything, maybe hiring a human to do something, scary stuff.

Goran Cvetanovski 1:34:42

Yes. Well, let me put some questions here because uh I I would like us to come at least to a two point uh two um two hours, and we are almost there. And uh we'll get something. So I'm interested, for example, like uh for where do you see the biggest demand for your services? Is this more from traditional companies or is it for uh um uh more digital um born companies? Is it more from mid-sized small companies or large enterprises? How do you see this?

SPEAKER_07 1:35:08

I would say for all of them. There everyone needs engineers, every company needs to make this journey now being AI first. For some companies it's more urgent than others, but it feels like most people in management teams, etc., understands now that very soon we're gonna have to adopt to a new reality, and it's gonna be mostly engineers that are making this transition. True, and to us it's it's everything from industry companies in a German suburb to a one-man show to Fortune 500 companies, it's any industry, but of course, like if you're tech heavy, it might be a little bit more urgent. So that's what that's where we're seeing it the most right now.

Goran Cvetanovski 1:36:05

Where do you see for example? Like, is it easier to let's say that you need to sell to Anders who is technical or me who is not technical at all, right? To whom it is a little bit difficult, uh is more difficult to sell.

SPEAKER_07 1:36:19

I prefer the the professional buyer for sure. All right. We we never sell to, for example, an HR organization unless the person we're talking to is very tech savvy. Right.

Goran Cvetanovski 1:36:31

We and why is that?

Who Buys Engineering Talent Today

SPEAKER_07 1:36:36

Yes, we are cheaper than, for example, uh a person working in Stockholm, for example, but as a remote option, we're priced priced higher, I would say. We are an expensive remote option. But the people we're having are really, really good. If you're not a professional buyer, you're not gonna see the value. The value. Like you need to understand when you're talking to this person why it's worth paying more. It's still a lot lower, a lot lower cost than hiring someone, for example, in Stockholm, but um for going remote, we're slightly at a higher price point.

Goran Cvetanovski 1:37:23

All right. And it's more like uh which type of a titles do you usually buy? Is it gonna be a CTO? Is it going to be like head of AI engineering? Yeah, etc. So I'm trying to basically paint uh the because uh all my questions are related correlated. My point is like, okay, uh like you have, I mean, for me, this is very um what is called valuable just to get your insight, like, okay, who is hiring at this point of time? Is it startups, is it uh, is it uh uh mid-sized companies, yeah, digital born and uh and industries, is it the AI and et cetera? Because this is actually gives you also a moment of time. Yeah, so if you tell me like oh manufacturing is right now killing it, right? Yeah, yeah, they're really like left, right, and center hiring people, you can see that there is a pulse. Okay, so why are they doing that? Manufacturing usually usually utilizes machine data to innovate and etc. Yeah, right. So it's a it's uh it's provides you a macro perspective where we are at this point of time, and that is where I'm trying to get some kind of a uh your own intuition. What does it tell you where we are right now in a uh moment of time, both in the maturity of the buyers, uh, both in the is it novice, mature or experts that are hiring? Is it more like one industry towards the other? Is it more traditional? Titles like CIO's chief information officer, is more for AI because this gives you a lot of things.

SPEAKER_07 1:38:47

And I can imagine that there are a lot of insights you can get from the data we're getting in. Um but not that I've seen for next time. Uh let me let me crunch some data for you.

Goran Cvetanovski 1:39:00

Yeah, it will be very interesting to see actually. I I I have some data as well I can share for data innovation summit because we did some uh we did some um maturity assessment uh exciting, which is uh shows that it's approximately 60 to 65 percent of the organizations, but this is mid-sized and large enterprises and private sector. They are in the um in the mid-reach, um implementing there is like around like 10 to 15 percent, uh very high advanced companies that are talking about agent workflows and etc. I will present this on Data Innovation Summit, and there is like 15 to 20 percent still talking only about BI and business analytics.

SPEAKER_00 1:39:42

Um which we should combine the data.

Goran Cvetanovski 1:39:45

I think it will be some interesting stuff that we can do there.

SPEAKER_07 1:39:48

But I'm so fascinated like how the most obscured companies need so much engineers.

SPEAKER_00 1:39:55

Um what do you mean by obscured?

SPEAKER_07 1:39:57

No, but like when you're thinking about companies needing software engineers, you're mostly thinking about oh, they have this website or they have this tech product. Yes, but there are so many companies that you have a feeling are so distant from tech, but they have massive RD teams working with software engineering. Um just like we have one client that is having a team that is only focusing on how can we ensure that this screw is tightened enough, and there's a massive tech team surrounding that question. One are drilling holes in um drilling holes in I don't know what you call like the not the sand mine? No, um, you know, like surrounding Amsterdam, like so the water doesn't get in. The banks um like they are making tests aside like how how much water does it require to break this wall. Yes, as an example, like yeah, yeah, they have also have big tech teams. Yes.

Goran Cvetanovski 1:41:15

Um for us uh finance and manufacturing are since 2015 until now, they are the biggest actually um buyers. Finance first, then uh then uh manufacturing. And finance has the most, I think that in my in my opinion, the is the most advanced uh industry when it comes to AI and analytics, because they were very early in regulation, they needed to have strict rules about data and everything else. So I think that they're right now leapfrogging very fast. Of course, the use cases are not the same as different companies where uh I'm talking about traditional companies, I'm not talking about digital-born companies like Spotify and etc. Because they're digital-born, yeah. The whole point with them, right? I mean, Spotify had the first engineering team, what, 2000 and from the beginning, the old engineers. The founder is an engineer. The founder is engineered, so it's 2006-7, they were when they started. So, but for traditional companies it's very uh difficult. But manufacturing, I think that uh the second actually um the second um insight that uh the data shows is that actually in manufacturing, you have manufacturing is always about digital uh twins. How do we make all processes all to be as digital as possible? How can we have direct representation of what is happening in the drill, which is in the middle of the Mexican Gulf or somewhere else? Uh and it's not exciting data, it's just a linear data like pressure, temperature, and stuff like that. But it's like uh teams of 50 people, just like how they can uh increase the pressure of the drill without breaking it, uh making like uh the factory uh I don't know, furnace to to drive two, three more years because it costs millions to represent. And usually we uh we are misrepresenting those. So uh this year there has been a lot of focus on supply chain and operations agents in manufacturing, which is very interesting. Yeah, um, and most of these uh German companies and Swedish companies. I was with uh Sumil Gupta Kowald yesterday. We were talking about this. It's like uh it's really, really agentec is profiliating itself into supply chain and operation quite a lot. So finance and manufacturing side.

Anders Arpteg 1:43:32

Well, isn't it rather clear that you know engineering will grow much more in the future since everyone can become an engineer in the future without knowing how to program in Python TypeScript and whatnot?

Goran Cvetanovski 1:43:42

I was leaving negative there, so I cannot uh back you up on something. So otherwise I will contradict myself. You need to help him out here.

AGI Timelines And An Optimistic Future

Anders Arpteg 1:43:50

Anyway, what's next for us, Proxify? What are you going to expand? Do you have any plans to expand in different areas of job roles or in countries, markets, and uh or something else?

SPEAKER_07 1:44:03

We are now adding uh PMs, so product managers. That's like the only role within a classic software engineering RD department that we haven't filled yet. So the missing piece. Recently, we have also stepped into like enterprise tech products like SAP, Salesforce, etc.

Anders Arpteg 1:44:27

Surprisingly high demand with a low Salesforce engineers or what?

SPEAKER_07 1:44:32

Absolutely. There are so many configurations that need to be done, and doing these configurations is really expensive through consultancy firms, etc. We have seen a massive spike from companies wanting to have consultants doing these small tweaks themselves.

Anders Arpteg 1:44:53

So no signs of the um SaaS uh apocalypse kind of uh helping no absolutely not no.

Henrik Göthberg 1:45:02

Interesting.

SPEAKER_07 1:45:03

But I mean I would I would love to see that in some sense. I'd much rather have every company building their own SaaS products. No. So that's gonna require a lot of engineers for sure.

Anders Arpteg 1:45:18

Okay, uh Victor, we're usually um you know trying to end um this podcast on a very philosophical topic. And um go ahead. And then we we can imagine, you know, if AGI were to happen, uh do you believe AGI will happen, by the way?

SPEAKER_07 1:45:33

Or yeah, at some point. Yeah, it will.

Anders Arpteg 1:45:37

Do you have some estimate for you know when that could be? Or is one year, five years, fifty years?

SPEAKER_07 1:45:43

With today's definition, um but how how would you define AGI?

Anders Arpteg 1:45:52

Let's take uh some I prefer the some ultimate definition, meaning AGI will happen when we can have an AI system that is on par with an average human coworker, meaning you can take a receptionist or you can take a customer support human or you can take an human engineer that is on average level and actually replace him properly with an AI end-to-end. And I I would say we are far from that yet, but yeah. Because yeah. But when that happens, um when we can actually have AI replacing an average level human position that we have with AGI.

SPEAKER_07 1:46:35

Yeah, an an average person, yeah. Um six years maybe. Would be my guess.

Anders Arpteg 1:46:47

Interesting.

SPEAKER_07 1:46:49

What do you think?

Anders Arpteg 1:46:50

Well, uh I keep my same year uh time and time again. Uh similar to Ray Kirchwell. He's he's been saying 2029 for a long time, and I think uh I I stick to that as well. All right. Sounds good. Sounds good. Four years or three years. We'll see. Um anyway, if that were to happen, uh we can imagine a lot of things happening. Either we can take one extreme, which is the dystopian future, where you know the matrix and the terminator happens and we will have machines trying to kill us all. Or we can take the other extreme, which is a utopian future where AI actually do uh help us in medicine and cure cancer, and uh it helps us with the climate crisis, and it helps us with the energy needs, and and basically all the goods and services we can think of basically becomes free. So we have a world of abundance in some way, so we don't need to pay for anything, basically. We we have everything everything we we need in terms of clothing or food or or whatever whatever you would need to survive and live a happy life, or happy is a strong word, but at least live. Yeah, um that's very negative. Just breed, it's nothing like that. But still, we can imagine this kind of world of abundance, um, and and that could potentially be a good future. Um, and then we can see the full spectrum between the two: the dystopian versus the utopian version. Where do you have any thoughts you know where we potentially will end up?

SPEAKER_07 1:48:26

Usually in the business when we talk long term, it's like three years away. Uh so I don't spend this much time thinking about these questions, to be honest. Probably I should.

Anders Arpteg 1:48:38

Um because at that time, you know, your kids will live in this world.

SPEAKER_07 1:48:42

Yeah, they will, they will. So let's focus on these years so they don't have to work, maybe. No. Yes.

Anders Arpteg 1:48:50

I mean, we work, you know, 40 hours plus a week these days, and perhaps you know, in 20 years or 10 years, we will have people looking back and saying, Do you know 10 years back we actually worked 40 hours a week? And I think that's crazy more or less, right?

Goran Cvetanovski 1:49:06

I don't see that happening. I think that you will work double. Double.

Anders Arpteg 1:49:11

Oh yeah.

Goran Cvetanovski 1:49:13

Within the same 40 weeks, the KPIs are just increasing. So you're gonna do that.

Anders Arpteg 1:49:18

Yeah.

Goran Cvetanovski 1:49:20

I don't know. It doesn't look very positive.

SPEAKER_07 1:49:23

But but uh yeah, I I can imagine that you in general work less and less and less. I don't think that trend is gonna reverse. I think you we're gonna continue to work less.

Anders Arpteg 1:49:32

Yeah, I agree.

SPEAKER_02 1:49:34

Um so what are we gonna do with all this time that we have left?

SPEAKER_07 1:49:40

I mean, I think that's not gonna be an issue. People will figure out what they're gonna do. Um, have walks, yeah. Netflix, scroll, infinite scroll is always there for you.

Anders Arpteg 1:49:54

Yeah, the scrolling never stops. No. So the only thought. Do you think we will end up on more of the utopian or the I'm very positive.

SPEAKER_07 1:50:05

As a person, I think it's going to be optimistic. I think it's going to be great. Yeah. How do you feel?

Anders Arpteg 1:50:13

I think so as well. I think it's more like an 80-20 split and uh probably 80% positive. So I'm certainly rooting for the positive side.

SPEAKER_07 1:50:23

Fingers crossed. But with that said, I think there are going to be a lot of challenges as a society.

Henrik Göthberg 1:50:30

Yes.

SPEAKER_07 1:50:31

And I also see a big risk, for example, that some countries or some areas are truly falling behind. Like today, we are so in Sweden take it for granted to be in the leading position in the world. And I think that might not be the reality. That might change. It has changed many times before, might change in the future too.

Goran Cvetanovski 1:51:01

True. Whatever we are discussing today, it's a Western problem. It's not the whole world problem. There are still 2.5 billion people that do not have access to water.

SPEAKER_07 1:51:10

No, yeah, true. That, yeah.

Goran Cvetanovski 1:51:12

So uh I mean we're talking about AI and people are thinking about how we're gonna get the next uh meal.

SPEAKER_07 1:51:18

So and I was also referring to like how today, if we ranked the the wealth in the world, Sweden is definitely in the top.

SPEAKER_00 1:51:29

Oh, yeah, absolutely.

SPEAKER_07 1:51:30

But the position at the top has changed many times throughout the history. And true. I don't think we can take for granted that Sweden, for example, is gonna be in the top. Um I I think it will, but we can't take it for granted.

Goran Cvetanovski 1:51:46

We can and that with uh with a great thing. Uh how did uh Sweden come to the top by becoming an engineering company, the engineering country, right? In the 60s, 70s, and uh 80s.

Anders Arpteg 1:51:58

So with the help of Proxify to engineer Sweden and many parts of the world, uh I thank you so much, Viktor uh Janheimer, for coming here, speaking about the amazing work that Proxify is doing. So please keep up the good work. Thank you. And thank you so much for coming here.

SPEAKER_07 1:52:15

Thank you very much for hosting me.