The Unconventional CTO: A Journey from Developer to Founder and CTO with Harold Thétiot

keywords

CTO, Sylaps, video conferencing, software development, AI, mentorship, technology trends, startup challenges, UX design, leadership

summary

In this episode of the Becoming CTO Secrets podcast, Philipp Deutscher interviews Harold Thétiot, the CTO and founder of Sylaps, a video conferencing tool. Harold shares his journey in the tech industry, his unique perspective on the role of a CTO, and the challenges he faces while managing multiple responsibilities. He discusses the evolution of the video conferencing market, the importance of UX design, and how he maintains motivation over the years. The conversation also delves into the impact of AI on software development and offers valuable advice for aspiring CTOs.

takeaways

• Harold sees himself as a software architect rather than a traditional CTO.
• The video conferencing market has evolved significantly over the past decade.
• Maintaining a focus on UX is crucial for product success.
• Delegating responsibilities is essential for effective management.
• AI's impact on software development is significant but not a replacement for developers.
• Mentorship plays a vital role in personal and professional growth.
• Quality assurance and testing are critical for product reliability.
• The importance of releasing products in a timely manner cannot be overstated.
• Understanding the market and user needs is key to success.
• Finding the right people to work with is more important than equity.

Sound Bites

• "I'm many things at once."
• "AI is stupid."
• "Shares is not everything."

Chapters

00:00
Introduction to Harold Titeo and Xylabs

01:42
Harold's Journey as a CTO and Founder

03:46
Inspiration Behind Xylabs and Early Development

05:02
Acquisition and Delegation of Responsibilities

07:00
Current Market Landscape for Video Collaboration

09:13
Challenges of Juggling Multiple Roles

10:13
Maintaining Motivation and Productivity

15:41
Differences in Experience Between Small and Large Teams

17:07
Innovation Strategies in a Small Company

18:46
Strategic Responsibilities and Business Aspects

20:56
Differences in CTO Roles Across Organizations

23:08
Navigating Code Ownership and Delegation

25:50
Staying Updated in Tech: The Role of YouTube

27:15
Collaborating with Freelancers and External Partners

28:24
The Future of Meetings: Quality Over Quantity

30:37
The Evolution of Software Development and AI

35:39
The Changing Role of Developers in an AI World

41:28
Advice for Future CTOs and Developers

🚀 Becoming CTO Secrets ist ein Podcast von Philipp Deutscher Consulting.

🎯 Interesse am CTO Coaching Programm? Mehr Infos findest du hier.

💬 Vernetze dich mit mir auf LinkedIn!

Transcript

Philipp Deutscher (00:00)
Welcome to the fourth episode of the Becoming CTO Secrets video podcast series. My name is Philipp Deutscher and today I'm talking to Harold Thétiot the CTO and founder of Sylaps, a video collaboration and meeting tool that I was also looking into a couple of years ago when we met the first time we kept writing and emailing and talking.

A couple of times since then and yeah, I'm very happy Harold to talk to you today because I'm having a lot of questions about your role of a CTO, how you interpret it, how you see the business and the market. So happy to have you.

Harold (00:40)
Hey Philipp, I'm so happy to have you too and see what you're doing now after we met a while ago and let me know if my accent is fine because sometimes my words, you know, in French you just ask me to repeat and I will, no worries, no shame in that.

Philipp Deutscher (00:53)
All good. I'm

having a German accent, I suppose. I think we're... As long as no native speaker is listening into it, we are good.

Harold (00:56)
Yeah, well, Germany is lovely, so that's fine.

All right, all right.

Philipp Deutscher (01:05)
All right, Harold, maybe you can start briefly introducing yourself in just a couple of sentences. So who are you, what are you doing and how do you see yourself in your role at Sylaps as a CTO, as a developer and CTO? Tell me.

Harold (01:21)
Yeah, my name is Harold, I'm 40 years old and I'm considering myself like software architect most likely in time of what I can do and the technology I can build things with.

Right now I'm the CTO of Sylaps, a company I created 10 years ago that is basically video conferencing for free, end-to-end encrypted and a couple of more features on top of that. We've been acquired in 2018 by one of our biggest customers and since then we still have a free service and we have a paid tier for medical application and that's what we do now.

And so yeah, it's a really small company. So that's why I'm CTO and founder too. So yeah, and we have some resource, but that's why I guess you asked me to come today. I'm many things at once because you don't need to burn costs in many humans if you can have control of everything sometime too. that's why.

Philipp Deutscher (02:25)
True. And I can also recall from a conversation we had before when I asked you to come to the podcast, because this podcast is for engineers who want to become CTO one day so that they learn all the aspects of how it is to be a CTO or to become a CTO. You just told me, Philipp, but I'm not a usual CTO. I want to say, okay, what is a usual CTO? I mean, you're a CTO as well, the same as a CTO of an enterprise.

So there are so many aspects of how to interpret the role of a CTO. So that's why I also wanted to hear yours, because as you just said, you are wearing different hats and roles. You see yourself more as an architect than a CTO. And that's the reason why I also invited you, because this kind of aspect of being a CTO is super interesting. I just wanted to ask you, that's something that was going on in my head for quite a lot of time. What inspired you in the beginning to...

to develop Sylaps to build the product and the company.

Harold (03:26)
Yeah.

10 years ago, it was all about Skype. So to invite someone to talk, you had to send him a Skype invitation and he had to accept it. And then you could talk basically. And I was in South Korea at the time and with my wife, was not working basically sometimes for some weird reason. And at the time, to...

The exact WebRTC was just coming out in term of API in Chrome.

So being a WebRTC lover, I had definitely tried it. I love UDP also, it's special, I love UDP for some reason. So yeah, I tried it and I had a prototype pretty quickly. And over time I created a free service and being able to monetize it in the time I knew there was B2B. So more like white lab all and so on. Not monetizing the free version, because at the time I knew it was so difficult to make money from free user.

Yeah.

Philipp Deutscher (04:29)
interesting. And then the company at some point was acquired, right? So you were building it and then at some point you had some investors knocking at the door or not only investors, buyers, right?

Harold (04:41)
Yeah, initially in terms of staffing, because I didn't want to wear all the hats to make sure I was doing the right decision. I even elected a CEO, Hervé Pievache. There was a buzz of mine in the past to be in charge of that, make sure I was doing the right thing in terms of accounting, billing, Most important things you need to delegate as founder or CTO is billing and accounting.

It's not your signature on it, it's theirs. So it's good. If they do a mistake, they own it and you pay them for that. And CEO is interesting because if you're doing product, you may want to have someone above you saying, no, it's time to launch, it's time to monetize that part before continuously building feature and burning money basically. And someone with experience is always the...

more interesting than your first startup. So I given share for that. And I also given shares for UI design and like currently, since I have some character, it was given to a woman called Meg Verlaine and she got shares. So at the beginning, I burned 15 % for a CEO and 5 % for someone in UI, in good design. And I kept the rest as a founder in terms of shares. So the two role I wanted to delegate was CEO and billing and invoicing, accounting behind it. And UI

because even if I know UI, I'm gonna be picky, it's gonna take too much time for me to achieve some UI where an expert in UI will, with some feedback, achieve it quicker.

Philipp Deutscher (06:16)
Okay, interesting. I mean you're still very much into the whole meeting and collaboration space, right? And market. How do you see it today? I mean, is the market already shared across Google Meet, Teams, Zoom? And then there is, I mean, maybe even Around. Who else is in the market who is relevant? And Sylapps, of course, right?

Harold (06:25)
Thank you.

It's a really good question.

And I've seen 10 years of video conference product. We met together when you were working somewhere else in the video conferencing market, a big legacy player you were working in. What I can say is I'm still here. Many products have disappeared, right? And I'm breakeven.

So that's what is important in the end. And I have a user base that continuously grow because we have to provision architectures.

every month more. yeah, I it's good. even like the fact that I'm not too big because if I had successfully like, don't know, like WhatsApp or to Facebook, I would have lost completely control over it. I would have to delegate every part of the business. I know I can optimize much better. So in the end, I'm glad at the size I am, the number of users we have and that we are breakeven. I'm glad that's the most important thing.

Philipp Deutscher (07:37)
still enjoying operating as a one-man show basically influencing most of the aspects right

Harold (07:41)
Yes, know

I have to have other projects on the site, like for example, background removal. We had that for a while. I'm always like two or three years in advance in terms of technology that we have in the product. We don't necessarily have it land in the product yet because we do B2B where we sell those features already. But what I have to say is to stay interested.

I have to work on more interesting features. That's why with Sylaps with video conference, as with the product we're using right now, where you're recording, you're playing to the clouds, there's going to be transcription, blah, blah, blah. There is always features you can build around, but keeping the core experience perfect, I think, is the key anyway. Like you're a cook in a restaurant.

you can do, if you go to restaurant and they do burger and pizza, you prefer to go to a restaurant and they do burger and pizza or only pizza and only burger? So I think doing one thing and doing great is better than doing everything.

Philipp Deutscher (08:46)
Okay, and still as I said, so you're the one man show still. How...

Harold (08:52)
No, no, I have

resource in the US that can develop for me. It's more on the front end in terms of backend DevOps. Yeah, I do definitely everything those days. But the size we have like between two and five people are full time on Sylaps It depends on the time of the month.

I do everything, but it's more like having a small company unless you have millions of users and you do millions in revenue. You don't need to have 10 people because you just burn money. So that's why I think I have the size I have right now. I could have grown faster, but would I have still enjoyed it now if I was bigger? I'm not sure. I've been through acquisition in the past, like from Daso system at Netvibes from a company with 10k employees.

I don't think growing too fast and having a ton of people, you get frustrated too, think, because you're never going to achieve everything you want.

Philipp Deutscher (09:42)
That's true.

What were the biggest challenges that you faced juggling those multiple roles within the company at the same time?

Harold (09:52)
Billing and accounting is really delegate that. The rest is UX. So make sure you don't do the UX by yourself. At the beginning we had webmaster, you had one role and you are doing everything, Now frontend became more prominent.

But the aspect of front-end that is important is UX. The technology used to do front-end is pure developer choice. You use React, Vite Angular. We don't care. That is not interesting, by the way. The more interesting is how the UX you do it, how much money you burn to achieve this UX iteratively. So you need to release this UX. It should not stay in limbo for years. So you can put it in user hands and tune it. And for that, think mock-ups is always underrated. People start to code right away.

And I think more like, and Figma is more prominent now, but 10 years ago, everybody was not necessarily doing wireframe and mockups, even if they knew it was the right way to do it. Mockup is more preeminent, I think, those days, because of good products that came back. So iterating on product UX with people that know UX is one part I would delegate to, in terms of doing everything, and you should not. DevOps is good for security, to have other people looking at what you do.

So even if you do dev app yourself at some point, you know it's a cat and mouse game. Always invest in third party to audit. Make sure you are on the right direction. Even get some little certifications that mean nothing depending on the budget you can allocate in it. They're gonna poke in holes you know are open. You don't want to close them there. Close them. They're gonna be happy. You can say proudly to your future customer, yeah we have that badge of audit from them. Blah blah blah.

Philipp Deutscher (11:11)
Hm, true.

So if I understand you correctly, it's like you can do a lot of tasks, not simultaneously, but being responsible for different areas in the product. And as soon as you identify areas there where you see like, okay, maybe I'm not the best fit for doing it, but, or maybe I need an expert supporting me, then you're pulling them in, right? As you said, for accounting, for billing, for front end, for UI, UX and so on.

Harold (11:57)
Sylaps was born like that too because quickly after South Korea when I came back to the US, someone called Antoine Pitrou, also is a founder of Sylaps and it was all about the UX for him. So that's also why the UX was always a slice of Sylaps that I knew. All I wanted was a terminal basically but UX you need... The pillar of Sylaps was easy to use, secure and no account. So it was all about UX anyway.

Philipp Deutscher (12:22)
How do you maintain motivation and productivity when you are shouldering so many responsibilities more or less alone? Over such a long period of time? You just said it, like more than 10 years now and it's growing, you are saying it's break even. How are you doing it?

Harold (12:42)
You cannot really replace having a conversation. Like we have right now, exchanging this information the way I do it, orally through video, you see my emotion. There is no other way in the world to exchange information as fast. You can write, but it's biased, you will read it. If I do uppercase all the way, you will think I scream, but I'm not.

So what I'm proud with Sylaps is the millions and millions of hours we gave people talking to each other. You cannot really put a price on it. What did they do with it? Did they talk about the loved one that was missing? Did they exchange knowledge? I know already what they do with it. But you cannot really put price on it. So what is motivating me is the number of users and the millions of hours they use.

to talk to each other and there is no price on it. If everyone was talking more to each other, it would be much better than interpreting text or things like that, I think.

Philipp Deutscher (13:41)
Yeah, and that's also that, I mean, that was one thing I realized a couple of years ago when it was the first time we met, that you were super passionate about the industry, like the video collaboration and video meeting space, and that you're actually the most profound person I have talked to who has knowledge about all the nitty-gritty details in video call engineering, let's call it that way.

about all the different frameworks and technologies. So that was quite impressive. So I can hear all the time when we're talking, your passion about this industry and your product. So that's amazing.

Harold (14:21)
Think 2001,

Stanley Kubrick wanted to predict the future with 2001. What he did is a father calling his daughter for birthday from the moon. There is no value you can put in exchanging information human to human. There is no price on it. That's why I believe.

Philipp Deutscher (14:41)
Amazing.

Harold (14:41)
That's why

AI, I'm joking about it. AI is stupid. We all see the AI bubble right now. The nitty gritty that it's not thinking. It is stupid. And yeah, it no stuff. But like an auto-complator. It is just an auto-complator on Transformer from 2014. Yeah, it's nothing. Yeah.

Philipp Deutscher (15:04)
Yeah, you're right. You're right. And in what ways

do you feel like your experience differs from that of a developer in, for example, in a larger team? So because with all the things that you are doing and all the things that you're doing over such a long time and so on.

Harold (15:14)
Yeah.

I also worked in a larger team, so I know what it is. I worked at Netflix in a team of 50 for Open Connect, that is Netflix CDN. Content Delivery Network for streaming, basically. So I know what it is at scale, like 135 countries.

with that number of Netflix subscribers. So I know what is a priority when you want to grow, when you want to spend the money. Now, not to be bothered when you are really big. That I think helped a lot. And also I work with people in the team and mentors that I take inspiration from taking the decision or not and delegating the decision to a mentor.

Like even you, Philipp, at some point if I needed like a vice president or someone, I will definitely turn to you based on your experience, the trust relationship we have. I would have come to you for this kind of position because I cannot do everything, for example.

Philipp Deutscher (16:20)
Thanks a lot. And still, mean, you are working a lot of time, you're working independently from others, right? How does this time, I mean, that's how I see it at least. And I could also see like this is difficult for, that could be difficult over time for problem solving or innovation strategies. So how do you approach this? Because, I mean, how do you stay sharp in also trying out new ideas and problem solving that you're not always falling into the same trap and how you're doing things?

Harold (16:46)
Yeah.

So I think that's a part of it. I wanted to ensure to not delete my work, but to show you that I'm I'm trying to work smart. Sylaps is a lot of open source.

And like we have the iOS WebRTC implementation that is used by many people, even some big names. And we have EasyRTC or OpenEasyRTC that is a Node.js WebSocket implementation that is used by Sylaps And because I'm small, instead I maintain some big open source package used by Sylaps that is used by others so they help me. So I have three developer.

That's how I do it. But you have to have some package you maintain with a good community. If you go there, will see all tickets are, all issues are tagged. We have a workflow and we have also a good community. People help between each other, so I don't have to reply. So when I go, they know it's for a big deal for something. And I call for help. And we have donation in open source too. And sometimes people, donate to me, but I redirect the donation to a better contributor.

that did contribute and was not rewarded or can contribute to the bounty but I cannot. So I want to mention a big part of my work in terms of package is open source for Sylaps with some big packages in the community of video conferencing basically. That helps a lot. Otherwise with this contribution I could not do it.

Philipp Deutscher (18:15)
Have you found a way how to work or how to switch working from hands on development and then working more on the strategic level or are you not doing the strategic thing at all?

Harold (18:25)
No,

strategic. I have a vice president called Craig Nossobat that is currently in charge of that. And we ensure that we have long term revenue growth and recurring revenue in B2B. That's our strategy. My current executive is more from the medical world, so there are more long term B2B growth.

On the B2C I do almost what I want because it's free, they want me to monetize obviously, that's not where I'm gonna go today.

it's like a test base because I have lot of paid users nobody is aware about. The free version is I release first in production for free and per quarter if it's stable I release for B2B and it's ultra stable because I have millions of users that going to tell me something is wrong anyway. So it's funny, it's not the first product. I have free users that I sell in B2B like Netvibes that was acquired by Daso was the same. And if you can have a SaaS version that you monetize but free user it's like your QA

It's really good, I think. With a set of features.

Philipp Deutscher (19:33)
And

so are you often coming in touch with the business aspects of the role of a CTO or is this mostly outsourced to this VP?

Harold (19:39)
Definitely.

No, nothing is outsourced. The VP is basically reviewing the proposal, working on the legal, making sure it's a fair deal. There is no sub-licensing that we will...

be not aware of, for example, like reselling what we're or IP that is safe and so on. That's what the business is managing. Me, I'm always the first person on the sales call with the prospect anyway. Yeah, because I don't want to sell to everyone. We refuse customer. If they don't meet my standard, I do not sell them.

Philipp Deutscher (20:17)
Are there any, I mean, coming back to what I just said in the beginning, like that you felt that you are not a usual or typical CTO. What would you consider like the primary differences between your responsibilities and those of another CTO, maybe in a bigger organization?

Harold (20:35)
even more delegation. When you are a CTO, think the first thing you have to have a bigger crew, so you delegate more decisions, basically, I guess. Hiring is most likely something you do. mean, in big corporations, you always have a turnaround. So you have to make sure that your manager are hiring right now for positions they need under themselves, I think.

and you generally looking at the raw cost of functioning, I do too. Like we were at OpenConnect, we were looking at the functioning cost of CDN for Netflix. And that's always the target. think as CTO, you're driving a big boat, a cruise, and you have your product guy doing body on the deck, know, with features for user pool and so on. But the real question is, who must crash the boat to drive on the sea, right? And the CTO is the most knowledgeable person, I believe, in the company. What is the raw cost of functioning?

So I'm not sure this is answering all your questions by doing that.

Philipp Deutscher (21:33)
do you think your role, when you compare your role with the role of other CTOs, does it require a different skill set?

Harold (21:42)
Yes, maybe a CTO I'm currently interacting with when I integrate my product in themselves is they are less hanging on technology. They are not developer obviously. They may know the good guidelines and the most used framework or of course they have foundation, but and they're mostly aware of some stuff, but they are not aware of scale sometimes or nitty gritty development good practice. Yeah. They like they enforce this.

Philipp Deutscher (22:09)
Could you

see yourself being a CTO in a larger organization or is this something that you never want to do?

Harold (22:16)
Give me an army! I always say give

me an army. Yeah, I always say give me an army. Give me an army. Give me an army.

Philipp Deutscher (22:23)
But

it was also required that you're adapting your interpretation of the role, right? That you would need to delegate more, then maybe you need to take off your hands from the code, at least to some degree, and delegate more responsibilities to others. I mean, after such a long time in this situation where you're controlling literally everything of the product,

Harold (22:30)
Yes.

Yes.

Philipp Deutscher (22:47)
How would that make you feel? mean, could you realistically see yourself growing in that direction or would you rather stay in where you are right now?

Harold (22:52)
This thing is sick.

So I've

been a developer for 30 years almost, so it's not the first project I have to work and let go or delegate more part. And I felt something I want to tell you. I've contributed a lot of projects and some developers, they are like territory. don't they touch my code. It's not the same. So they even commit like bullshit after you commit after that. Yeah, I coworker like that. So I'm not, because I'm a big open source person, I don't feel any ownership of the code or product because for me,

so much collaborative effort anyway. So it's more putting it the right hands than letting the baby go is more the real question. Like I want to put it in the right hands, basically make sure it, and it keep working because it's just growing, doing nothing almost, but we keep always improving support, adding features, updating iOS app, Android app, know, stuff like that. yeah, if I had to let go, it's just a matter of the right people. Yeah.

Philipp Deutscher (23:52)
And I would assume that you would focus more on code reviews instead of writing the code by yourself.

Harold (24:00)
If I had more time, I would write games. I like to program games. I would program some other stuff. I like to program on my free time. So many different stuff. Whatever, I like to program anyway. I think that's the big difference between CTO, know, the most likely. I love code. I have to tell you. I do it on my free time. I love open source too. You contribute to the same ecosystem. I can code many other things.

Philipp Deutscher (24:25)
I'm also

coding on my free time still, and even though on the business side, I'm not coding anymore for, I don't know, more than a decade. So yeah, that's difficult to stay in shape, so to say, and to still being able that in a discussion, in a technology discussion or a workshop that no one is able to bullshit you, right? Even though you were not, that you were not touching the code.

Harold (24:49)
Exactly. I'm so proud you're saying that.

I'm so proud you're saying that. Just not being bullshit, but it's an AI, it can do that. But it's an Indian in India looking at the camera when you check out for AWS, for Amazon shopping.

It's not new, what saying is not fake, it's been released, I don't know if you read about it. In the end, yeah, you bullshit investor. Where you can actually do it properly, no need to bullshit. You do recognition, yeah, there is a low level of recognition, but can be trained, it takes time. Instead of selling dreams that are not truth, I think.

Philipp Deutscher (25:29)
Very interesting. How do you stay up to date and updated with the latest trends, not only in the video space, but also in what's going on in technology without the support of a larger team, right? YouTube?

Harold (25:42)
Yeah, YouTube basically. YouTube,

there is really good YouTuber these days. I'm not going to mention them by name, there is ton. Pramagen is one of ex-Netflix employees, Pramagen, I think. I'm going to find the name after. There are a few YouTubers that are really, you know, they are opinionated like me, right? But they are opinionated because they have tried the technology.

Philipp Deutscher (25:54)
Come again? I didn't get that.

Okay.

Harold (26:08)
They know how it scales, how much server, CPU, RAM you need to put behind how many users. That's always the question. Yeah, I can use fancy language, but how many users is it going to cost me? So it's like, Does it scale? Yeah.

Philipp Deutscher (26:20)
Any external partners that you're involving from time to time in the company? Or is anything either like, okay, maybe...

Harold (26:27)
Always,

there is always good partners that follow us for years. Actually, there are contracts on their own that we call once in a while. They've been, yeah, for DevOps to be to Yeah, freelancer, definitely. And UI, work. Yeah, yeah. I would say even Fiverr if you can find someone, but you do the real work with drawings.

Philipp Deutscher (26:39)
So freelancers for DevOps part? Okay.

Harold (26:54)
You know, it's not just Fiverr quick and dirty, you find someone on Fiverr that is good and you collaborate with him over a long period. Fiverr I tried, it's okay if you find the right person I think, for UX and drawing and marketing, something like that. want to mention. Freelance is definitely a good way to delegate for budget. If you get the budget too, you can just spend it and so on.

Philipp Deutscher (27:17)
And if you could make a wish now, for PsyLabs, what would it be? Like 10 times more users and meetings or what else?

Harold (27:27)
this one is funny. They always ask for many people in the meeting, but it's useless. I mean, I don't see why you would have 100 people right now in video. Let's say we are, right?

But even Twitter, they understood that with spaces. Because like that, you have a moderator and you select who is speaking. So when you want 100 people in a meeting, you don't want 100 small videos. You want 100 people and a moderator like you do now that can select who is speaking, right? That, yeah, we have that in B2B already. If we can release it for free, it's not a good use of servers.

But the technology is moving too. So that's why it's interesting because if we use QUIC, that is landing soon in the browser, we don't even need a server anymore. We can send almost RTMP from the browser with an intermediary, right? So then, yes, we will put it for free. It will break the market, for example, where people do a lot of money, where it's AI on the browser to remove background, and they make you pay for that using your resource.

There is no value in that. So same with streaming, the more we can do locally, we don't make user pay, basically.

Philipp Deutscher (28:37)
But you're absolutely right. I think that was a progress or that the company in this industry needed to make over the years to understand, well, we don't need like 200 people in a meeting. Yes, we need 200 people in a meeting, but they're not all active participants in a meeting. And it's never intended that they're becoming active participants. So maybe you have just a couple of broadcasters and the rest is more or less silently listening or they can become

Like they can become collaborators if they're raising their hands, so to say. Right. So you don't need to have 200 videos in one stream. That's I think there was when Corona came and suddenly everyone wanted to switch to videos and then there were the town hall meetings. So everyone wanted everyone to turn on the camera. I understand why this was like a requirement a couple of years ago, but it turned out that yeah, as you said, no one is using it.

Harold (29:35)
Yeah, and I like this format you're doing now of interview couple of youtuber or podcaster are doing it They are big on Twitter and I really like the vision of spaces on Twitter. I think it's fixed Met a niche that was needed, you know of space with a moderator most likely public you can join the space sometime and raise your hand talk on the back office with the moderator for so you can ask your question I really

Philipp Deutscher (30:00)
I mean, actually it's

the concept of the app Clubhouse that they implemented by themselves and as Twitter has the much bigger or X has a much bigger user base, it was becoming very successful on their platform. But on no other platform in the same way, like Spaces is only really working on X, if I'm not mistaken.

Harold (30:16)
Yes, exactly.

Yeah,

yeah. So I think the good evolution also is the using of voice. You can listen again, information, emotion, because even if you don't have video, you have my emotion in my voice. It's better than reading, more easy. And the mass of information I can transfer to you in these 20 minutes compared to writing, impossible.

Philipp Deutscher (30:44)
But let's talk a bit about your vision of the future of software development. So how do you envision the evolution of software development in the upcoming years?

Harold (30:52)
So in terms of what, I mean the big thing now is

AI, right? And it's funny because like last week, again, I'm reading, so I'm not sure I'm right, but you can quote sources and you will find it. Apparently the error ratio now in GitHub generated source code in terms of bugs is like about 45%. I think that I heard. And so they're realizing, okay, AI is not really making you more productive necessarily, right?

That's my current understanding of the recent article that has been released, And I'm not surprised because it's just trained on bad code that is open source and it's feeding on itself. So yeah, guess what? You put bad code with bad code. You just end up with bad code. That's why most of the developers are lazy. Nobody's writing tests. Nobody's linting.

Nobody, mean, most of the people are not spending the time in source code quality in the public area. They will make autocompletor based on the source code good basically for me. Yeah, it's good for complex source code. You want to refactor, you give it a code base. Yeah, keep this standard, just refactor it, make two function instead of three, blah. Yeah, it's gonna work.

But the trend of using AI, I think, reached a cap right now. So we'll see how it goes. I used it for some snippets. It was useful sometimes. But it's not much better than a good intern and a Google, basically, I felt. So.

Philipp Deutscher (32:24)
Have you ever

heard about like, how is it called, Cursor? Cursor is derived from Visual Studio and they were integrating a lot of large language models. So you can use, you can have the AI as part of your developer environment.

Harold (32:29)
This one doesn't ring at all.

I think I...

Yes. And it's read your code base. Yeah. Running locally, I believe. Yeah, I'll try to. I like it because it runs locally on your code and it knows your code base, your way of thinking. So the results are definitely more pertinent. And guess what? It runs locally so you don't pay anyone. So it's good. Of course, maybe there is a better model on an NVIDIA GPU in the fucking cloud. But whatever.

Philipp Deutscher (32:50)
And you can chat with the AI and it's, I mean, I have tried it out. I was,

Exactly.

Harold (33:20)
Now I think I'll roll this

Philipp Deutscher (33:21)
good thing

is what they also realized is not every LLM is best for every challenge in software engineering. So, and so they have integrated all, literally all of them, so that you can also figure out which LLM is the best for this kind of problem that you are facing. So that's great.

Harold (33:34)
It's obviously.

So I do

understand Transformer, like for real, like, works. So it just makes sense. You have a valley with the good practice, more you feed it the good practice, less valley where the learning curve will be is going. Simple as that. You cannot have an AGI with Transformer and LLM. You cannot have reasoning. It is not reasoning. It is an autocompletor.

I'm really mad at the word reasoning used by some people. It's not the word. There is no neuron creating you information. Yeah.

Philipp Deutscher (34:22)
But for me, for someone like me, for someone like me,

product like Cursor is excellent because I mean, I'm not spending as much time in software development and coding as like an employed software engineer, right? Who does this like from the morning until the evening. So for me, it's very easy to use Cursor to set up new projects and product ideas to set this up very quickly.

in different languages and it's able to create everything around my product. So it takes a lot of engineering work off my shoulders. I do not need to spend time in setting up new environments and all that stuff. Sorry.

Harold (35:05)
It's good for prototyping.

I think for you as an executive, it's good. Craig, my executive, and we talked about exactly what you discuss now. It's good for you as an executive to prototype something. It is doable.

Philipp Deutscher (35:18)
Exactly.

Harold (35:20)
It is viable. You know which technology you try to couple that is the way to go and then you can set your product or development. Okay, this is what's to be the core. Now do a product with it. That I believe is going to help you, the executive, to get less bullshit from developers. No, it cannot be done. We need 15 people to do it. And that's it.

Philipp Deutscher (35:42)
But how might this, like software like Cursor, how might this change the role of a software developer as such tools and also platforms are advancing?

Harold (35:54)
Less Stack Overflow, I it's going to... And we already read some article over that, it's taking some people from the time we spent as the developer on Stack Overflow. So I think it's going to reduce that, that's the first question I'm sure I will give. And after maybe the role, we'll have to change to more, as anyone reading AI, checking it's not hallucinating. Because it does, like package, not just package, that doesn't exist. Even with the Claude or...

GPT-5.0, you still have package that does not exist. I guarantee you. Or they are private and maybe they got crawled by the LLM, I don't know. But so I think more time will be spent by the developer as making sure it's good code. There is no case that was removed from the previous prompt. Sometimes you prompt it for more constraint on the code and you just say, give me this function. It forgets that you had a case in the previous case you had to keep and it did not.

So the developer will spend more time interacting with AI instead of his brain basically.

Philipp Deutscher (36:59)
So what skills do you believe will be essential? Is it like the cross-checking of the code that the AI is going to produce or is it more like orchestrating what several AIs are doing? What skills do you believe will be essential for the future?

Harold (37:10)
Orchestrating will come from product.

Orchestrating will come from product that you mentioned, the one that is inside the IDE. It's reintegrated with Google UX to ask the good AI or learn on his own source code. That's going to be integrated by people doing the app, not the developer, I think.

we see how it's getting even more and more accessible. Even Apple is integrating in an app and you have all the AI. So I don't think people will fine tune LLM by themselves except geek like us maybe. It's getting more and more accessible to train it on your own data and run it locally. So that is not a skill the developer will have to learn maybe to take.

a bit but not spend too much time on it. It's gonna rely on good prompting and selecting the good LLM maybe, benchmarking, yeah, yeah, the code this one is given me is not the high standard I want, or it's fucking stupid at some stuff. Yeah, the testing should be what they spend time because if you do good testing and AI changes source code and your test is good, the test went through and you're good. So I do a lot of testing like.

People say I'm crazy, but that's why I'm super confident. When I do a test, it goes through full CI, and it comes from Netflix. By the way, they have like an incredible CI with A-B testing. So if you see feature on Netflix that are missing, no, no, no, no, we did all the feature. We just did A-B testing and you watch less because of this feature, so we removed it. But that's, that I think is, if you are good testing, good pipeline, and you use AIs, source code generated, blah, blah, it's gonna ease too.

Because you're going to have trust in your pipeline to deploy, the tests are not broken.

Philipp Deutscher (38:47)
So you're an advocate of 80 % code coverage or even more? 90?

Harold (38:53)
90 90

yeah 90 yeah

Philipp Deutscher (38:58)
Wow. I mean, nowadays I'm talking more to people who say like, hey, you can also even have like 20 % code coverage if you're really testing like the critical parts of your software.

Harold (39:11)
What it's going to give you is simple. Trust when you deploy. If you have 80 % test and it pass, right? You are VC and they say, can I push the button? You yes. If they hey, we got a release, big release, we have only 20 % coverage Can I push a button? Yes, but if it break, are Friday, you stay here, right? That's it.

Philipp Deutscher (39:30)
Yeah, true.

That's true.

Harold (39:32)
And you know you need

to make your user churn. Churn is something. These days you have alternative products for every product. So it's all about execution. So if your release is breaking, customer dissatisfaction, they leave. They go elsewhere. So that's why quality, I think, for release with testing is all about that. think too.

Philipp Deutscher (39:53)
True. And how do you anticipate AI and all the automation impacting software development? Will developers be replaced by AI at some point in time?

Harold (40:02)
I don't think so, I'm a full stack and I do low level, so me, I'm not gonna be replaced.

But it's more like all the juniors are recently came in the industry after the COVID also they say, I'm going to have a work from home. And they wanted to become developers. They purchase training from any people that said you're training, you see what mean? And now they're seeing AI and I think a lot of companies are hiring less because they use AI in their pipeline too. Same in the game industry, it's big for assets and so on.

So I think it's going to be tough for people with less experience because the AI can do what they do as a lesser price. But you still need someone to use the AI. So that's maybe a new job that's coming. I know it's a caricature right now, but being able to prompt correctly something is getting interesting too, I have to say.

Philipp Deutscher (40:58)
Interesting

thought. I mean, that would mean that the hurdle for becoming a good developers is getting higher. Right. Because as a junior, you still can be replaced by an AI. So why would I hire a junior? Right. So you need to learn more and more to not come into the market as a software developer and

Harold (41:07)
Yes.

Great.

Philipp Deutscher (41:22)
and you could be easily replaced by an AI again, right? So you need to have more knowledge and more experience and expertise already.

Harold (41:25)
I'm sure I'm feverish in price right now.

I'm sure on Fiverr the price went down since we have more AI, so this Fiverr or Contractor friends can get more some contract. Because people pay their subscription to AI to Claude or ChatGPT and they give it a shot and they're actually learning. think it's funny, we also forget that by interacting with AI the person is learning something too. So they're getting more confident about their knowledge, like more knowledgeable. But sometimes it can be biased too.

Philipp Deutscher (41:57)
true.

Harold, we are coming to an end of our conversation. I have a last question for you. mean, what advice do you have for the future CTOs or for the ambitious developers who want to become CTO one day? How do they prepare for the future? How do they set up themselves for success?

Harold (42:03)
Of course.

Have good

mentors. Mentor is so important. I had good mentors in my life. People that trusted me and I learned from them. would be good. Whatever their position. can be a CEO, can be a mentor for you as a developer. Even if it's not in your technical area. Release good quality, but nothing can be perfect. Perfection does not exist. So you have to release at some point.

That would be my second advice. Because people, they try to perfect the product and it's too late to market or someone has already released. And instead about being the first, like Apple always proven, it's about having a good compromise between quality of what you value proposition and you should release at some point. And the last advice is, shares is not everything.

Maybe at the beginning of the company, you can give 5 % to the right person that will help you. The difficulty is finding this person, not giving shares, I think. That's it.

Philipp Deutscher (43:19)
yeah,

true. Wise words at the end, Harold Thanks a lot for the time. It was really a pleasure talking to you. Always a pleasure, I have to say. So yeah, looking forward to our next conversation, Harold. Take care.

Harold (43:30)
You're true to me, Philippe. I'm so glad.

Looking forward to the talk and to see more of what you're doing.

Philipp Deutscher (43:38)
we will do.

Thanks a lot. Bye bye.

Harold (43:41)
Bye,