#253 - How Vidio (Indonesia's #1 Streaming Platform) Built Great Engineering Culture — Now Supercharged by AI - Tommy Sullivan

[00:02:03] Introduction

Henry Suryawirawan: All right, so thank you so much for your time, Tommy. Today, we are going to talk a lot about, you know, building engineering culture at Vidio, also some of your stories. So yeah, welcome to the Tech Lead Journal.

Tommy Sullivan: Hopefully stories and jokes. We can mix ’em all up.

Henry Suryawirawan: Right.

Tommy Sullivan: Thank you for having me.

Henry Suryawirawan: Yeah. So for people who don’t know Tommy yet, Tommy is the CTO at Vidio. Vidio is actually one of the largest OTT or streaming platform in Southeast Asia. So it’s actually a pleasure to meet Tommy. So we had a connection through Mohan. Yeah. Mohan is my past episode, so always a pleasure to have, you know, people who are doing great things in engineering around this region.

Tommy Sullivan: Well, I’m glad that I knew about the Tech Lead Journal before Mohan introduced me to it. But I’m glad also I have a bit of a chance, an opportunity to rebuttal anything that he said without him being able to complain. But yeah, it’s, so thank you for having me.

[00:03:07] How Did Tommy Go From Silicon Valley to Jakarta?

Henry Suryawirawan: Yeah. So let’s start maybe a bit from your story. Like how did you end up here? I think if I’m not mistaken you have spent over than 10 years being in Indonesia. So someone who grew up in US. How did you end up here?

Tommy Sullivan: Well, I started out working for kind of a dev shop in San Francisco. And right after CMU, after college, then I was working at a dev shop. And there I had the opportunity to work on a project with a much better dev shop. And around that time, this is around 2009 and kind of in the middle of hyper scaling in, you know, SF area in terms of, you know, like Facebook and Zynga and all of those things were blowing up. And, when, and this is also like right past the 2008 financial crisis. And so I was just growing my career, starting my career. And I was working at a, again, I say dev shop somewhat pejoratively in the sense that, you know, a place is mostly focused on like getting something out, just getting it done, focused on the result, not necessarily how it happened. And I worked on a project with, also with some engineers from Pivotal. And at that time I saw the work that they were producing and how they were essentially like craftsmen in the truest sense of the word. And it was inspiring.

Around that same time, I also had the opportunity, I was interviewing with Pivotal and interviewing with Airbnb. And you know, I had the experience of working in that dev shop and seeing how a small group of engineers, like if you’re going to be a junior on a team, you could potentially be at the whim of whatever your team lead or the CTO wants to do. They could be a good idea. And actually I have no scars in that for my first company, you know, but I saw how they controlled the tech stack, the, you know, the decisions in that culture. And I was very, while I was you know, talking with Brian and Nathan and the early group at Airbnb, I was a bit cautious of joining a two, three person engineering team at that time, because I knew that no matter what, they could potentially, you know, steer down a route that, you know, who knows.

And so I decided to join Pivotal and thinking of long term, what was best for my career to become a craftsman in software development. And so I’m very happy with that decision. No, I certainly would be way richer being the third engineer at a Airbnb. But from Pivotal then we opened an office in Singapore. And Singapore is kind of like Asia 101 in terms of it’s a very comfortable market to get into. And, but in Singapore you also have the entire region’s kind of unlocked. It’s very close. And so there I was working with Pivotal, consulting with tech companies in the area. And Pivotal is interesting. You, they do engage for software. They, it is, they’re closed now, but they did engage for software development. But their SVP had a good quote once that was kind of like, if you’re engaging with Pivotal just to increase your like engineering manpower or you’re to add more devs to your team, you’re wasting your money. And because it’s like the goal of Pivotal was to go in there and to help improve the engineering culture of the teams. And so my job as a consultant was to help focus on engineering culture and principles. And so I was doing that throughout the region. And I started doing it with Emtek here with Mohan. Yeah.

Henry Suryawirawan: Right, I know a few people from Pivotal Labs. And in fact, I used to work at ThoughtWorks as well. I think we kind of like admire each other. So we were…

Tommy Sullivan: We didn’t touch on that!

Henry Suryawirawan: So we were kind of like into craftsmanship, building the right culture. And what you said is right, right? So they hire Pivotal Labs or ThoughtWorks, not just for hiring another engineer, but it’s to learn the best practices, inculcate the practices and all that, right? So I think, thanks for sharing that. Because I’m sure today we’ll talk a lot more about building great engineering culture.

[00:07:22] How Has Indonesia’s Tech Scene Evolved Over the Past Decade?

Henry Suryawirawan: But maybe let’s start from after the story, you know, you come to Indonesia. What did you see here in terms of tech scene back then? And then if you pull it out maybe 10 plus years now, like how has the tech scene changed?

Tommy Sullivan: It’s gone through quite an arc, right? When Mohan and I first came here it was just as Indonesia was starting to like tech boom all of that, you know. Sort of like Gojek and Tokopedia, Traveloka, you know, all of these things were starting, were at their seed phase, right? And so, and we saw those early teams there. And when Mo, and it was, also a lot of while that was happening, there were also a lot of traditional companies that were trying to digitize and become digital natives and stuff. You know, we had like Matahari Mall, I remember they saw Amazon and you know, back then there was a phrase of them having 500 million that they were investing in terms of just like. And they, so they were building a product to replace their, you know, pre, you know, non-online presence. And they were building three at the same time actually. Because they had so little trust that one would be successful and they had so much money. And that’s really hedging bets in terms of just, it shows such a problem with trust of they had a team working on a product and they had another team working on a competing product that just cause they were trying to hedge their bets. And I think that there was an issue with regards to trust and with regards to understanding software engineering.

And so like when Mohan and I came, first came here, certainly, I mean the tech, we had engineers, we had a team of about five people when we first joined. And several of them were on minimum wage, you know. And so the scene there then was engineering is just such a side function and the same as like an office boy or something in terms of just, you know, make the website work or whatever. And Emtek really invested in engineering with working with Pivotal and then pulling Mohan and myself in. And we saw, you know, like people here working on laptops like with or laptops connected to monitors, like with literally just down the office over here of like 12 inch monitors and stuff like that. And it can seem hilarious now, you know, like, huge, huge monitor. But, and I can understand why the company was in a state like that. It’s kind of like a vicious cycle, you know? Like you can hire engineers and then they want the fanciest tech and all of these things. But that’s just one piece of having like a proper engineering organization, having a proper culture, right?

But that cycle’s also two parts in the sense that it’s the company not necessarily having invested up until that point, but it’s also the engineers not investing in themselves to some extent. I want to, again, normalize that with respect to, you know, a lot. Socioeconomically, not everyone can afford, you know, a great investment in terms of the proper tech and the computers and all those. But I do think that one piece of, one sort of thought or thread or theme. One theme that I think, that I keep in my mind is the rifleman’s creed, which is, it needs to be taken with a bit of candor but is that, you know, like, this is my rifle, without it I am nothing. Without me, it is nothing. And I think that as engineers and in tech, our computers are our rifles, you know? And we need to have the most effective working tools at our disposal. And so the company not investing in them and engineers not investing in themselves for that, I think that that’s a perspective change in terms of these tools being so important and investing in oneself in the team.

So early on when Mohan and I came in, we kind of, that was the tech scene that we were familiar with. During that time it was, certainly we saw a lot of organizations that were more focused on LinkedIn post, and more focused on what is popular to say versus what they’re actually doing. And, you know, I remember early on we had a lot of poaching happening. You know, we’d hire engineers, you know, develop them, and then organizations poaching them. In some of the organizations there was a peer-to-peer lending situation for a while that kind of grew up. There were 300 peer-to-peer apps here, of which only like 200 were approved by OJK. And all those organizations were democratizing finance in such an altruistic goal, and yet they were charging 500% APR. And so in some of the, actually at one point, Mohan and I had staff that owed significant amount of money there. And we were building a team, the culture, building trust with our engineers. And even Mohan donated money to help, you know, pay that debt off with that staff, which I think is just a side effect. It is just a symptom of the trust that we’re building with the team and the level of engagement that we’re building with the team, right? And that has served us for years since then.

Henry Suryawirawan: Yeah. So I believe back then right when all these tech scene hasn’t really started, right? Finding good engineers definitely is very difficult. I mean once after, you know, the Gojek, the Toko started, investment coming in, I mean, we can see a lot of engineers.

Tommy Sullivan: Softbank. Crazy money.

[00:13:12] What Happened to Indonesia’s Engineering Talent After the VC Bubble Burst?

Henry Suryawirawan: Yeah. Lot of engineers starting popping up, you know, the salary also increased very high. What do you see now because, you know, the economy situation, especially in this region may not be the best at the moment. And also, you know, with technological changes and all that, do you see people are moving off engineering or you know, like what is your sentiment here?

Tommy Sullivan: Indonesia went through a lot of brain drain, early on. So when we first came here, there was like good engineers in Indonesia who were leaving Indonesia because they could make more money in the Bay Area or whatever. And so that was something that certainly suffered. What is actually, I guess, a positive benefit of the crazy money that came in, the Softbank and all of that, is that a lot of those engineers started coming back because they realized the opportunity. And so we, those engineers have, you know, Indonesia has even surpassed some of the developing markets in the area. Vietnam, you know, I think of and Malaysia in the sense that because VC money and all this money came in and just saw the crazy opportunity and just kept flooding it to the point where it became a bit of a bubble in the sense that there was so much growth and opportunity here, that it led to, essentially organizations that were never going to be profitable and based on hype more than actual results. And so now it has kind of come down a bit and it is, I think that now, you know, we’re in a more mature phase of tech companies here that are focused on profitability, that are focused on economics.

[00:15:03] Why Is Indonesia One of the World’s Most Exciting Tech Markets?

Henry Suryawirawan: Right. So maybe for the audience here who are not familiar with Indonesia or this region, right? Tell us why the opportunity in Indonesia is such a big deal. Especially considering if you look at the global, you know, population and countries, right? Indonesia is always there. Like maybe the digital social media penetration, number of people. Definitely population is huge. What, if you can summarize, what do you think is the opportunity in Indonesia compared to other countries around the region?

Tommy Sullivan: I think that, you know, there’s that funny quote where Indonesia’s the country of tomorrow, even tomorrow. With regards to some, because a lot of times it constantly feels like it has so much opportunity and sometimes it’s not realized, you know? And that opportunity stems from vast natural resources. It comes from a relatively significant middle class. It comes from a huge population. People don’t realize fourth largest population in the world. And so VC, companies, engineers such as us, you know, see that, you know, if you can have a product or something that sticks here, it can grow so quickly. And Indonesia is actually very liberal with regards to business, I think. It can be, in terms of the employment side, it can be a little tricky in terms of, you know, hiring and firing people is much harder than in the United States or in Vietnam or something like that.

And certainly regulations can be tricky, but on the macro scale, the ability for tech penetration is significant. I remember for example, a concrete example of that is my father was involved in healthcare in the United States. He was the chief doctor for Medicare and Medicaid. And he was working on telemedicine in the United States. And in the United States, your doctor’s license is for one state. And so if you want someone in California to consult with someone in Massachusetts, that is a totally different, you need to be - granted your MD needs you, but you need to be relicensed in that other state. Whereas Indonesia, it is you, just the entire archipelago is. And so, the ability to, if you have a product or something that is working in Indonesia, I think that you can scale it very quickly.

[00:17:26] How Do You Build a World-Class Engineering Team When Starting From Scratch?

Henry Suryawirawan: Right. Yeah. Country of tomorrow, I hope tomorrow will come eventually. So thanks for the quote. So let’s start into talking about building engineering culture. You mentioned when you came here, you know, maybe not enough good engineers available. You in fact mentioned about five people. Some of them are like, you know, kind of like maybe juniors starting from a very low wage. Maybe my first question, a lot of engineering leaders would be frustrated given this situation. How do you actually overcome that? How do you actually approach this situation? Because you came from Pivotal Labs, I’m sure like, you know, top of the cream, you work with great engineers. Having to face this challenge, like what’s your approach here for leaders out there who are also maybe in similar situation, they don’t have a good tech team, but they are tasked to actually build a great culture. Maybe let’s start from there.

Tommy Sullivan: So there was a quote I can’t attribute, but it was, I think it was Rob, me or someone who was talking about building a team and having the opportunity to hire engineers who are familiar with your tech stuff or hire engineers who aren’t in any way familiar with tech stuff, but have that quality and that experience, right? And always leading towards the whatever, the experience and quality, right? And, you know, in terms of hiring engineers, sometimes people can be, okay, I’ve got a X, Y, Z, you know, tech stack, or I want to build, or I want to build some AI python stuff, you know, so I need to find AI python engineers, right? And that is very shortsighted and it’s better to find people who are hungry and who are smart and who are growing and learning and to then work with them in the tooling that they’re familiar with, right? So Mohan and I came in and we, it wasn’t, we knew we were limited in terms of finding the team. It was more grow the team, right?

And part of the extreme programming principles that we brought in from our Pivotal experience was, one fundamental thing was pair programming back then which is an incredible tool to onboard engineers and to onboard anyone in any process where you do the work together, right? I think that pair programming is certainly changing in this whole, you know, AI world. But in terms of two people sitting together, communicating, discussing the problem at hand, and whichever tools you use to solve that problem changes, right? But the goal is to find the hire for attitude and not just pure aptitude. There’s a, oh, I think I’m gonna ruin one of your, there’s a phrase that’s I’d rather have a hole in my team than an asshole in my team. And it’s, you know, hiring for that fit and letting the team grow, letting the team hire to build that culture that you want, right?

So what we did is we spent a lot of time early on defining the culture that we wanted to have here. And many of that is from extreme programming. Some of that is influenced, you know, from our own experience. A lot of it is around hiring people you wanna work with and that you that enjoy solving problems. And so we grew the team from about five people to at at one point, two hundred and at one point, four offices. And now we’re back down to just one office here in Jakarta. But it’s still a great team and, I think that the tenure, the average tenure of the engineers on my team, which is about seven years, is indicative of we’ve, we’ve succeeded in that regard.

Henry Suryawirawan: Right.

Tommy Sullivan: I think another aspect, I guess, just that I take a bit of pride in the team that we built is early on when we were here, we were engaging head hunters, right, to help find, to find engineers. And one of ’em was, geek camp, I mean, not geek camp, sorry, Geekhunter. And we were working with them to find engineers. And in their onboarding form of applying to use them as a head hunter, they had a, an input box that was with regards to which company would you like to kind of potentially poach from and what type of engineers. And I’m very proud that their number one input was our company at that time. So the results kind of showed there.

[00:22:01] What Are the Hidden Benefits of Pair Programming Beyond Code Quality?

Henry Suryawirawan: Right. I think it’s always great if you can build culture from the ground up, right? You shape the people that you hire with and all that. I wanna go back a little bit on the pair programming that you mentioned, right? Because I dunno, like these days it’s not well covered because of the AI and all that stuff, right? But I know that Pivotal Labs actually practice it like very religiously. In fact, I heard like eight hours, you know? The whole working day is actually to sit in pair and work together, right? So what maybe from your view, having done this so many years, what are some of the great benefits of pair programming, and also the one that you see succeed really well here in Vidio.

Tommy Sullivan: I mean, certainly onboarding. Onboarding, knowledge transfer. In terms of not having, you know, when you have two people working on the same project. The same project and having a same deep understanding of the code base, your bus count is reducing, right? I think that also there are subtle aspects of it that aren’t obvious from the outside in terms of also mentorship, right? When you’re working with someone more senior than you, mentorship isn’t only when you have a one-on-one with your manager, right? If you are working together, that is an aspect of mentorship. That’s a softer skill that people don’t, you know, see, intuitively from the outside. And so, I mean, pair programming is one of the many sort of practices within our the extreme programming that we were using. I think that now, with regards to the world of AI, I think that programming in terms of like writing, you know, the code is we’re becoming more abstract layer on top of it, right? We’re not writing. I’m curious, have you ever written assembly?

Henry Suryawirawan: No.

Tommy Sullivan: Yeah. Have you written C?

Henry Suryawirawan: Yes.

Tommy Sullivan: Okay.

C++. plus

Henry Suryawirawan: C++. Not C directly, but C++.

Tommy Sullivan: Okay. And, you know, it’s like, I don’t think you’re ever gonna write C again, you know, or C++, you know. But we’re kind of like layers abstract on top, right? And so as that abstraction layer continues to rise, there’s still that huge, you know, the, as the activity of two people fleshing out, what do we want to get done? How do we want to get it done?

One aspect of pair programming that I think is very hard to translate now in terms of the AI thing is we used to do a lot of ping pong pairing, where it’s like I write a test and then you make it past, And you write a test and then I make it pass, which was quite enjoyable. And that’s why Pivotal when they IPO-ed, at some point they IPO-ed, they all had like ping pong paddles because it was, you know, part of how they did engineering, right? And that is a bit harder in terms of the, you know, with an AI agent and stuff. Unless you really lock it down.

But that said, you know, extreme programming is, there’s a quote from Martin Fowler that’s, if you’re doing extreme programming the same way you were doing it a year ago, you were not doing extreme programming, right? Because it’s just constantly about assessing what is working in your organization. If it’s working well, what does it mean to take it to 11, you know? And an aspect of that pair programming is just a component of two people thinking on the same problem. How do we take that to 11, right? And we’re adapting that in the new world.

[00:25:28] How Is AI Blurring the Lines Between Engineers and Product Managers?

Henry Suryawirawan: Right. Yeah, definitely the world has changed because of the AI, especially in programming, software development, right? Yeah. AI seems to be like a huge disruptor, I would say. A huge leverage as well for some people. So maybe if I can ask you now your pair programming practice, how has it changed with this AI? Which part that you think will get reduced? Maybe like, you know, the ping pong thing, the test writing. But what part that gets more leveraged?

Tommy Sullivan: Yeah, well, I do want to correct a little bit ‘cause you said in your pair programming practice which is, that’s not, our practice is a bit more on the extreme programming, which is many things, right? Which is a component of values and then principles and practices that practices, pair programming is a practice, right? And practices are defined by principles, and principles are sort of driven by values. And these lead into each other. And that pair programming aspect is certainly changing. But the other components of communication, of transparency, of fast feedback, I think that a lot of those are just as true as they were 10 years ago.

One thing that I think is changing a bit that, an experience that both Mohan and I had, is as we were working for Pivotal, one of our, the office we were in got sold to a Japanese company, a Digital Garage. And we had the opportunity to work with Eric Ries a bit. And Eric Ries was well-known for his book The Lean Startup. And I think that as writing code becomes more commoditized, many aspects of extreme programming are still valid but other aspects in terms of building a product are becoming more valued. And so I think that there’s a bit of a mesh that we’re seeing in terms of engineers becoming a bit closer to being able to take on a bit more of product and objective of what is the goal of this individual component or feature that we want to do.

And at the same time, also engineers, you know, people can, a solo person can launch a company now. You know, OpenClaw or whatever. And they need to have that product sense in terms of what they’re building, right? And The Lean Startup is a very good way to validate that, right? Because, Musk has some quote around, you know, one of the biggest problems engineers have is like over optimizing something, right? And optimize, and a common problem is optimizing something that shouldn’t exist, you know. And so sort of those lean startup principles are, you know, validating the product, getting it out there, getting feedback. And so us learning from that sort of Eric Ries investment slash organization, I think is also influe- is quite influential into where we’re going, right?

We’re also seeing that sort of product mentality. We’re also seeing here in Vidio as well, in terms of product managers and designers are starting to code a bit more, right? And because it’s opened that up for them.

[00:28:48] How Do You Justify XP Practices to a Results-Driven Business?

Henry Suryawirawan: So XP as a practice, right? I think we all know as a software craftsmanship, it brings a lot of value. We can see it produce great culture, you know, great set of engineers and great quality output, right? And not just that. I think when you see XP practitioners, they’re highly regarded as a software craftman, right? But I don’t see in the industry it is widely practiced, widely used. My question is like how did you sell XP back then? Like, you know that this will work at Vidio. And the other aspect is why is it not widely practiced, you know, around the industry?

Tommy Sullivan: So you asked about XP and how it is practiced, but not wholly practiced across the industry. And how we were able to, sort of, apply it to the team that we were building. And when you join a team, I think you get a runway of trust that that works for a certain period of time. And we were able to, through, you know, focusing on MVPs, through focusing on sort of like lean startup principles, we were able to get something out and to deliver results, which is what the business is focused on. And from that, we gained more trust and gained more leeway. But, you know, there’s that Ben Horowitz quote that we mentioned, which is how startups tend to be like result focused. And equally important is not necessarily what you get done, it’s how you get it done. And while Mohan and I in the leadership were very focused on delivering results, we were also incredibly from the team perspective, we were very focused on how we deliver those results. And in understanding that, that we were building a scalable platform.

And the word scalable so frequently in tech is with regards to the number of requests or anything like that, which for me is not what I’m referring to. It’s the scalable in terms of after you grow and develop this platform for a certain number of weeks, months, years, it isn’t falling over, right? And a component of XP definitely supports that which is in regards to the test-driven development, which is equally as important in XP as pair programming or even more so in the way that you are write, building a system. Sometimes I think about programming as like you can write a script and it’s kind of one way. But if you’re writing it through in a test-driven development manner, you’re writing code that creates that output, but it is also validating the output. So it has that feedback loop.

And feedback is a core principle of extreme programming. And feedback in terms of that feedback loop, in terms of the code that you’re writing. The feedback loop in terms of having a CI that’s running all of those tests. Feedback loop in terms of if you are a pair programming or working very closely with people. That feedback, that human feedback loop in terms of like maybe we don’t, you know. So I think that the XP there, we were able to apply the principles that work, show results, build on those, and to continue to grow.

I think a very tricky thing in engineering. Now, if you ask McKinsey, they’ll have a different opinion, but is to measure like ROI or measure productivity of engineers. And fundamentally, I think that it is damn near impossible to do that. I do think that the DORA metrics are relatively useful, but they aren’t, you know, an organization tries to like measure people at the individual level and that’s incredibly hard. And DORA metrics don’t do that. Instead, they’re looking at how is your engineering organization running efficiently? The deploy time, the mean to correct errors, you know, to roll back all that. I think those are still very valid, right? I think things that when someone writes a blog post that says, yes, we have figured out how to measure engineering productivity, I think that’s a very click-baity because many of the other non-DORA aspects of that I don’t believe so much.

One of the other aspects of that that are highlighted is in terms of like developer happiness, which is valuable, right? But one aspect of that McKinsey post is around, you know, how many engineers are looking for jobs or how many engineers are being approached. And I take great pride in when we started the organization here and engaged with a few head hunters is that essentially Vidio was, back then we were under a different name, but was one of the top listed companies of people wanting to work with the engineers and poach from us, right? Which for me is is potentially an indication of, you know, developer productivity and success there.

Henry Suryawirawan: Yeah. And not to mention also the culture, right? Because people want to work in a great culture company. So I think, yeah, you kind of like mentioned a little bit like why companies should invest in XP. And I like the quote, the Ben’s quote, like, you should not just focus on getting things done, right?

Tommy Sullivan: It’s you got them done.

Henry Suryawirawan: How you get them done. And I think it’s more applicable now in this AI era, because I’m sure you have heard a lot of people thinking that AI can, you know, produce results fast, you know, doing vibe coding and all that. But not necessarily the quality is always best. Especially if you wanna talk about scalability, if you want to evolve the systems that you built over weeks, years, you know, decades and all that, right? I think it will be tricky if you don’t have all these core practices in place.

Tommy Sullivan: I do think that it’s very, we need to stay away from grandstanding though, that, you know, AI produces like lower quality code and stuff. I think that, what I’m happy we’ve done here is set some boundaries in terms of people being able to vibe code things. And so as I mentioned, we have, you know, some designers vibe coding products. We have engineers certainly using AI assisted code in terms of our development. And I don’t wanna limit and prevent and say that, you know, like, oh, all this vibe coded stuff is gonna have massive security issues and all those things, which it will. But what we’ve done here is create a little bit of a sandbox where we allow designers and people who typically aren’t writing code to produce products that can be used by our organization internally. And so we just have those behind an IAP and they’re essentially walled off from the outside internet. I don’t need to worry about, you know, huge security issues there. If we were to take one of those and then give it to our users in production, I would be definitely more worried, right? But in terms of internal tools that that can make our team more efficient and stuff, that’s an area where if it works, it works, you know?

Henry Suryawirawan: Yeah. I think this goes back to the lean startup principles as well. When you see a problem, maybe using your product approach, build something, validate that, whether you know there are people using it, find it beneficial and all that. Obviously product productionizing, something is a different level, right? Especially if used by, you know, external customers. So I think thanks for adding that.

[00:36:11] What Has Worked and What Has Failed When Integrating AI at Vidio?

Henry Suryawirawan: Speaking about AI tools, what have Vidio done in their engineering team? You know, what kind of tools have you implemented? Maybe we can share some success stories, failure stories, that would be great.

Tommy Sullivan: I guess, let me start with a failure, and it was potentially me, you know, like not listening to the sort of lean startup principles a bit, which is, you know, AI and is, and LLMs are great at producing code and SQL particularly, right? And so early on we created a sort of text to SQL inside chatbot, right? Where anyone in our organization could chat and it would query, it would query our data lake and get them results and all of that. And that was essentially, I have heard of success of, you know, democratizing data and giving people access there. But it didn’t work in our organization because, you know, once it’s wrong, once or twice, then there’s no trust there. And it was fundamentally easier for someone just to poke a data analyst and tell them. And it’s also because of accountability, right? If someone’s just talking with a, you know, LLM and then getting some data results, it’s a lot easier to say, well, I got the information from X, Y, you know, this person and so they’re accountable. So I think that that was us failing by building a product and tool that actually our organization didn’t need because they already had data analysts that could answer those questions.

On the success side, I think that we have had gains in terms of using ad video are our first usage of AI is jokingly around anak intern, yeah. Which means like a child intern for the non-Indonesian speakers. And using, so a lot of our video streams don’t have SCTE markers, right? So typical terrestrial streams or that are live stream video that’s coming along, sometimes it’ll have digital markers in terms of where an advertisement starts, the break starts and ends. And initially we would potentially have humans, anak intern AI, clicking and then doing TVC replacement there in areas where it wasn’t digitally transcoded for that replacement. And then starting to use models, right, that are trained in terms of here’s an ad break, here’s not an ad break.

And so I think that the, that AI in Vidio, and we’ve gone much farther than that, right? In terms of using AI models to detect sentiment within videos, to identify characters, identify, you know, for metadata labeling and all of that. I think that it’s most important to start from seeing a pathway of this is a flow of information or this is a, you know, this is a process that we have, and then using AI to then automate it and make it faster and better, right? That’s where we’ve had more success.

Henry Suryawirawan: Right. How about in development? Are you full into, you know, having everyone having license to, you know, whatever AI coding assistant tools? Or are you using as like a new, like when you do pair programming, is AI now is like third person, like maybe in software development? What have you done?

Tommy Sullivan: In the development space? I think that it’s gone through a few sort of step changes, I’d say, right? Like initially with the like GitHub Copilot and stuff, I was never really impressed and the team was never really impressed, right, with those tools. Because that was typically kind of like one way of like you have our code base, you potentially can load a portion of it in your context, you can help answer questions in that regard. Obviously, the model’s been trained in terms of how to write a for loop and all of those things. And it’s very useful there. But in terms of working with a larger codebase, it wasn’t as impactful.

I think there was a very serious step change when we, I learned about Cline bot which I believe was a tool that was created from some ex-Claude engineers, I believe they worked at Anthropic and then started Cline bot, which was… what it was was the sort of Copilot on all of that LLM for coding. But it was also connected with a bunch of pre-prompt layers where it was, I’m gonna ask you about this code and what you’re gonna do is you’re gonna check this memory bank for this information about our system. You’re gonna give me these recommendations and all that. We’re gonna talk about it, work about it. And then you’re going to update that memory bank, and then you’re gonna update our coding styles. And that is creating a feedback loop. And that feedback, for me, that was a step change in terms of it’s not just a one way street to Copilot, it’s a feedback loop.

And so then we started using LLMs and AI much more significantly because then it’s also improving your documentation, which is also improving the recommendations and so forth, right? And then so that feedback loop is valuable. And now I think we’ve kind of gone with the newer models that are just so incredible in terms of being able to run much longer and have much more of a context window. I don’t know where we are.

I do know that we, you know, we definitely enjoy using Claude Code more for the engineers. I think that it’s a lot of success there has also been in terms of like here is a pattern that I have identified within our system, be it of the 20 different, you know, apps that we have or services that we have. You know, here’s a pattern that we want to, that we like, we use. And we are inconsistent in terms of the application of that pattern and help me apply it to this other place. Help me apply to that other place. And that is an area where we’ve had success as well. In those kind of, those kind of, those refactorings or those sort of like pattern translations.

Henry Suryawirawan: Speaking about these patterns, do you actually encode them as like, I dunno, prompts, you know, or maybe Skills these days. Claude has support for skills. Or some kind like AGENT.MD or whatever. Is there any such practice where you, yeah, you know, encode these patterns?

Tommy Sullivan: Yeah, I mean, we certainly use like we, the, our DevOps team is actually quite good in terms of they’ve created a layer for an MCP to help document, you know, the various aspects of the infrastructure that we’re running and what services are best at certain things, right? And so that MCP has been useful. Also, a great MCP with regards to, you know, our data lake in terms of the thousands of tables that we have and what you can find where, right? So those MCPs have been useful. They don’t cure cancer, but in terms of the, you know, like a bigger factoring in that regard, that would be something that would be, you know, someone on a branch. And defining it, writing it into some UML potentially and defining the problem within some sort of memory bank files, whatever, however those be named. And selectively applying those to pieces of the code. But that would be grounded with, you know, a test-driven development, pushing, steering some of these LLMs towards. It’s very easy to talk to them. And then they write the code and then massive, huge chunks of code and then later they write massive huge chunk of tests. And it’s kind of the opposite, putting the cart before the horse. And so employing them and working with them from the engineering perspective, certainly applying our principles is still valuable.

[00:44:19] Is AI an Amplifier or a Threat to Software Engineers?

Henry Suryawirawan: Right. Speaking about AI, I’m sure many people in software development are quite scared. I myself also fear one day, you know, I’ll be obsolete. Because this model tends to improve significantly, you know, week by week, month by month. And in fact, we can see every time Anthropic release a new thing, the stock market even could drop, you know, the SaaS, the security products and all that. For you, as a tech practitioner for many, many years, and in fact you are also someone who is trained in, you know, great engineering practice and all that, what’s your view about the impact of AI? Would you think now that with this AI you don’t need so many engineers, you know, you can cut number of engineers? Or do you think it’s more like an amplifier that you can even increase number of engineers or something like that?

Tommy Sullivan: Yeah. I mean, my conversation with the business is all of, look at how much more we can get done. Look at all of these opportunities that we can pursue we wouldn’t otherwise be able to, right? So it’s more on the amplifier side of things. Vidio is in or we try a lot of things and some of them fail, some of them don’t. What’s important is that we don’t invest too much in any individual features or products. We put it out there with minimum viable product and we see the result and then we double down, right? If it’s working, so certainly it has been an amplifying aspect, right, of what we are able to get done and what we’re able to experiment with, right?

I think that it is scary in the sense that, you know, both we have spent considerable time learning the syntax of various languages and stuff, And that is something that you, I’m sure many people I’m sure are having existential crisis in terms of, you know, all of that is no longer valuable. I think that I started writing code because I did like to get things done and I like to see the result, right? And it’s just a tool in that process. And I too have feel a little bit disheartened in terms of, you know, the amount of time that I’ve spent coding that those, that syntax is no longer important. But I need to also remember that, wow, I can have an idea now and get it out so much more easily, right? So I think that we should I, you know, talking with the team, we were talking about pair programming and stuff and we’re doing less programming and more, I hate to say prompting but, you know, leveraging the tools to get things done.

[00:46:59] How Does Vidio Use Team Rotation and Shared Ownership to Retain Engineers?

Henry Suryawirawan: Right. So maybe speaking about culture, Vidio culture, what are other things that you think are unique to Vidio and that you’re proud of? Maybe other practices that you have, apart from, you know, those things that you have mentioned, like the XP practices, you know, the AI usage, the experiment thing that you have. Any other culture that you want to highlight here?

Tommy Sullivan: Your question was about unique aspects of Vidio’s engineering culture. I think one thing that has been valuable in terms of building Vidio’s engineering culture is we did start with that pair programming, we did start with small teams. And when you have pairing and when you have a good distribution of knowledge within the team where you don’t have silos of, you know, oh, this guy knows that thing and he’s the only person that knows it. When you build that resiliency in terms of multiple people knowing what’s going on in one place, then it’s easier to move people around between teams. Now certainly we have experts who focus on, we do have some engineers who are focused on just video transcoding, right, and that is not necessarily applicable to other parts of the business. But when you have engineers who are not necessarily wed to a particular component of the system because they’ve built it with others and they’re pairing, what you can do is you can move engineers between teams more easily. And so that has been a valuable tool for me as an engineering leader in the sense that I can see someone who potentially is getting a bit bored on the current team with the particular problem that they’re solving. That doesn’t mean that they’re not necessarily putting in the work, but I can see that it’s, they’re not learning as much new within that domain. And so then I can move them to another team and when we create a culture where moving between teams is quite normal, then it’s easier to maintain someone’s interest.

And when, you know, I had an engineer who early on, was just exceedingly good in a particular area, and I didn’t necessarily have a career growth roadmap of, oh, now these engineers are gonna report into you. Now this is gonna happen. And so instead, the goal, the way that I dealt with that was by to try and move him to another team that was a bit more, in this example, he was, it was a bit more ML-focused and, you know, the trending things, so he could at least get some knowledge there, right? I remember my dad used to always tell me, you need to work at a company, you need to either earn, yearn, or learn, right? And I’m trying to pay my engineers the best I can. But in terms of yearning, yearning in terms of just like loving the company, fortunately we do have a product that is quite fun in cases, in terms of we do have some entertainment. But the one thing that I have the most control over is learning, right? And so being able to make sure that someone in their career is constantly learning from others. And it’s not always gonna be forever on one team, you know. You need to keep people mentally challenged, right? So that’s one aspect of our culture of moving people around that I think has been valuable.

To that extent, to double click on it further is we have team leads, right? And we use the phrase as someone being an anchor, right? And I think it’s quite unique that here the team leads are not always the most senior person on the team. And because if it’s the most senior person on the team, then well age is linear. And so it doesn’t really rotate. So what we do is a lot of times we have team leads who are, you know, experienced, but it could be someone who’s hungry and learning in that space. And so then you have a more senior person on the team who can back lead them as the team lead. And so that’s another opportunity instead of switching someone from one team to another team. But it’s to potentially rotate the team lead such that they can play a different role within that team. And then that’s mentally and intellectually challenging for them, while it’s still, you know, bettering our business, right?

Henry Suryawirawan: Speaking about learning. I think thanks for sharing all these great things, right? Because we tend to forget that you have to build a culture where the people in your team is learning, right?

Tommy Sullivan: Yeah.

[00:51:16] How Do You Protect Code Quality Culture in the Age of AI?

Henry Suryawirawan: Touching a bit back to AI about learning, right? Because a lot of people now also have the fear, we are not learning as much as possible if all we do is just prompting, asking questions to AI and this model gets smarter. But we don’t necessarily get smarter because we don’t have critical thinking. We don’t, in fact worse if we don’t validate the output that AI is doing, how do you prevent this in your team? How do you prevent like, for example, AI slop that the team is producing because, you know, it’s so easy now to produce anything within just a few prompts. You can, you know, like generate something. Like how do you prevent people to stop learning and producing crap output?

Tommy Sullivan: Well, I think that’s something that individually people need to be responsible for, right? I, it’s very hard for me to, if you know, there, I’ve seen. I’ve heard of some interviews and so forth where people, where the interviews have anti-AI tools to make sure that, you know, you can’t use AI in the interview and stuff like that. And to me that sounds a little bit like banning calculators, right? And so I think it’s important to use the best tools that you can at any time, right? And so I’m not gonna… That said with regards to like brain plasticity and all of these things, you know, if someone is offloading their thinking to an AI tool and not thinking themselves, then it’s really hard for me to inspect, you know, whether or not they’re offloading that thinking, right? Critical thinking is important, not just in so, you know, in so many aspects of life, right?

A funny aspect in terms of producing slop, dunno if we’ll make it in the cut, is, you know, we have built a culture of people wanting high quality and high quality results. And in the minute, you know, you allow some slop, be it, you know, human-generated or AI-generated that you allow some slop to come through, then, you know, the ball rolls downhill, right? People, you know, new engineers come on and then they see this slop and then they reproduce that slop and it keeps getting worse, right? And I remember once a person on our team in us having built the culture of constantly trying to raise the bar and to, or at least maintain it, right? Someone was so passionate on our team with regards to the work that someone else produced. They were looking at reviewing someone’s work and called it haram. Because they said that this is not of the quality that you are being paid to produce. And so it’s essentially you being lazy and not putting in your full responsibility as a Muslim. And so I think that, that was just indicative of the, now both of those engineers are still in our team. And but I think that it was indicative of the passion that someone had for defending a code base and not allowing just garbage to come through.

[00:54:16] What Metrics Actually Matter for Engineering Quality?

Henry Suryawirawan: Yeah. Yeah, culture is definitely very important, right? When you have these people setting the bar that everyone should meet, right? You know, it gets replicated very easily. Same thing happens if you start, you know, producing worse quality, right? Your culture will also degrade.

Tommy Sullivan: There’s that, what is, what’s the phrase, and correct me, it’s you can have speed, quality, cheap. You can have it fast, quality, cheap, and… but then another aspect of that is also is the scope, right? And so on in terms of the, there’s the fast is one that I will compromise on, right, in terms of like, this may take a while. There is the cheap while we’re within our budgets, you know, however they are. But the quality is not one that I will sacrifice on. But the one that engineering leads need to be most careful of is with regards to the scope. And actually that’s a direct quote from the extreme programming. And so is we need to be very careful. And that is the lever that we need to play with.

Henry Suryawirawan: Speaking about setting the quality standard, right? Sometimes it’s very hard to define, but is there any metrics that you guys always look at to ensure that, okay, we can see it dropping, that people should, you know, get it back up? Is there any metrics or some kind of…

Tommy Sullivan: Yeah, metrics are always, you know, I think that some of those, you know, like, certainly, you know, early on people can take, what is it? Goodhart’s Law which is that any, when anytime any metric becomes a measure, it ceases to be a metric. And I have heard of teams concerned with regards to lines of code that are covered in test and those things. And I think that the metrics in terms of quality that we are most focused on is we do have blameless postmortems, right? And whenever something goes wrong and you’re having a postmortem, ideally it is not one or two things that failed. It’s like six or seven things that failed, right? When a plane crashes, it’s not one failure. And so I think that it’s in that review of what are all of the causes that we go through as a team, blamelessly, you know, and that is where we look at, you know, potentially an aspect of the system was built. You know, rarely it could be true, but rarely are we framing it with someone, you know, did a bad job or someone, you know, dropped the quality bar. Because it’s always under the understanding of, you know, quality as a function of time that was spent on it, right? And so but we look at them from that, those postmortems, from that angle. And it’s there where we decide.

Now, certainly if you’re having, I saw you had a great interview with Your Code is a Crime Scene. And that’s an area where you can, you know, step back and look at the metrics and see whether something’s being changed too frequently and potentially. So there are potential quality metrics and could, that could be looked at in terms of design of code or design of systems, right? But it’s nuanced in everyone’s contextual, right? Certainly like downtime and uptime of systems, those DORA metrics in terms of, you know, the time to recover and all of those. But those are more an aspect of the team having produced a system that has quality, right? Not necessarily like is that line of code high quality.

Henry Suryawirawan: Yeah. I like that you touch on, you know, blameless postmortem, right? Coming back to psychological safety within the team. And also the systems, right? It’s not like one component, you know, versus the whole thing working together.

[00:58:07] How Will AI-Generated Content Reshape the Streaming Industry?

Henry Suryawirawan: So maybe I wanna switch gear a little bit now, you know, speaking about, you know, content, you know, what Vidio is doing. I must say that I’m also a Vidio user. It is a very nice app. Comparable to, you know, all other streaming apps out there, you know, the Netflix and all that. The quality is quite good. But speaking about the content industry itself, we all know that there’s a major competition about this. And also the AI now can produce great, seemingly great content as well. What do you think is the changes in this, you know, content landscape? What do you see happening within maybe the next one year or so?

Tommy Sullivan: Well, I mean, I’m, it’s been for the past, you know, 15 years, well not 15, maybe past 10 years, where we’ve seen auto generated content on YouTube, right? And, it’s very easy to have in the publishing space. It’s very easy to have auto-generated articles. And I think in the news space, those auto generations have been around for a long period of time. I think that it’s where the content space is going, is you can look back at sort of the publishing industry.

And so like the Wall Street Journal maybe 50 years ago, the Wall Street Journal was probably 80% advertising, in terms of the revenue that they were receiving was from advertising, and 20% subscriptions. Back then the physical newspapers that people were buying. And now it’s shifted to advertising is much smaller and much more on the subscription side. And given that news articles and the articles being written in the Wall Street Journal can be auto-generated very easily and very quickly. And then the question is, why are people still subscribing to them? And I think that that’s an aspect of trust and of quality. And you know, when you’re reading an article from the Wall Street Journal, you don’t need to worry that this something, this - now granted they can be factually incorrect or something. But you know, you have such trust that they’re have high quality journalism, right? And in the Vidio space, in V-I-D-I-O space, or in the content space, I think that, you know, YouTube has some incredibly high quality content but it also has a sea of auto-generated kind of garbage, right? And so in the streaming space, I think that it’s gonna be going through a similar shift, right? In terms of the advertising component is going to, is shrinking and shifting more towards the subscriptions of, you know, for high quality content that has been curated.

Now granted if AI can produce high quality content that people like, then that’s great! But that’s not gonna be a one-shot, produce a 10 series episodes that’s good enough for Netflix, right? I was just In India working with JioHotstar or talking with them about potential collaborations. And they don’t really use AI for dubbing at all. It, dubbing, language is a huge component in India And they still feel like the quality isn’t there. And so, I think that, producing content that is of the highest quality possible, cause people’s times are limited. And so, yeah, not eroding the trust from your consumers by creating that slot for them, right?

Vidio does have, recently we just launched a series a totally, sort of AI-generated series, but I don’t think that it necessarily… There’s a lot of fear in the industry in terms of destroying jobs or in that regard in terms of artists and content producers. But it was a very long running series in Indonesia that was five, six years, 500 Somat episodes on TV. And I remember going to that production house about 10 years ago, and it was probably like eight, nine people, you know, guys, girls working in, you know, in Blender, creating some animations and all of that. And now it’s eight, nine people, that are working with, you know, Veo 3, Seedance, various models, still crafting those together to tell a story. And I think that the only difference in, the size of the team is about the same. The cost of the production is pretty close. I think that we’re getting better in terms of reducing cost of production there. But the output is so much better. And so it’s about leveraging the tools to create something that consumers want.

Henry Suryawirawan: Right. So if I’m not mistaken, this series is called Keluarga Pak Somat, right? It’s like Pak Somat’s Family, in the english term. But this one is speaking about animated series, right? What about live movies? Because like if we see Seedance, so many people have prompted, you know, a lot of, you know, seemingly cool, kind of like live action thing, right? Do you think it will also kinda like disrupt the industry? You know, like all these people, maybe the stuntman, or you know, the background, whatever that is, right? What was your view on this?

Tommy Sullivan: You know, I guess before I get into the live action, one thing that I would flag this as somewhat similar I guess is, so Vidio, a huge component of our consumers are looking at live sports, right? And so I could just as easily, you know, create an AI copy of, you know, Liga 1 or EPL and have little players run around the field to kick balls. And, but at the same time, I think that you know that that wouldn’t mean anything, right? So I think that, I don’t know how to describe that of, well, AI wouldn’t work there, right? But in terms of the live action and in terms of the artists or storytelling that you’re telling, I think it really just depends upon the story that you’re telling, and I think that it is a tool that is useful in the process. I know we’ve, one of our recent original series, you know, had quite a few zombies and all of those things. And currently right now, it was easiest for them to film it with humans. Well, zombies. But we, the team, now that they’ve gotten more involved in the Keluarga Pak Somat and gotten better at AI generation stuff, is they took a few parts of it that weren’t even the human aspects but they were just, for example, like blowing up a bridge or some FX, right? And I don’t think anyone can complain that, oh, the FX was replaced by, because it’s all just what is the tool to make that happen, right? And so that’s a tool that makes that storytelling cheaper, easier, more accessible, right? And so, I think that in the content space, what’s gonna drive success is going to be telling good stories that are meaningful to people that they wanna follow. And artists will be involved there. I think more… I don’t know about more, but certainly equally important is the storyteller. Yeah, that’ll have value for a long time.

Henry Suryawirawan: Right. I mean, speaking about all these effects, right? Visual effects, I mean, people have used a lot of tools. I mean, in some movies, in fact, even like those cheap visual effects is there, but people still watch, right? I think what you mentioned, maybe it’s a kind of like true, right? People don’t want to see crap movies, even though with good visual effects, right? So I think storytelling, maybe also the acting, you know, the expression and all that, all the props. The cinematic thing might still…

Tommy Sullivan: I saw an amazing example of use of AI recently. It was around for advertising of sports artists, act - sorry, of athletes. And who are busy many, many months out of the year and in them lending their faces and their likeness to a completely different person. But then using that for commercials and TV of that regard, which is interesting place to be in.

Henry Suryawirawan: Yeah. I’m sure people have seen, you know, all these scammers now can also replicate people’s face, people’s expression easily. We’ll see some more use cases I guess in advertising, content generation and all that.

So speaking about Vidio, I’m sure you must be proud that is one of the biggest streaming platform, OTT platform in Southeast Asia.

Tommy Sullivan: Yeah.

[01:06:51] What Does It Take to Stream at 4 Terabytes per Second?

Henry Suryawirawan: Maybe to grow to that stage seems to be a lot of hard work, I assume. What are some of the memorable challenges? Maybe technical challenges or maybe, I don’t know, like industry, content challenges that you remember that you can share with us?

Tommy Sullivan: We certainly have hit scale at a few different points. I think that, you know, scale in, so in, the OTT and the over-the-top space, you have sort of AVOD, advertising base, and then you have SVOD. And Vidio overall has kinda shifted a bit more towards SVOD. It’s a lot easier to make money when someone’s paying for it. But when we had very large livestreams that were AVOD based and anyone can just, you know, subscribe. They can just log in and then watch those, then we have had some very interesting scale situations. And I think that one of a memorable one was like the Asian AFF, which is Asian football something games. And you know, we’ve hit around the four terabytes per second in terms of bandwidth. We’ve had, about like 2.2 million concurrence. Now those aren’t Indian Hotstar scales, but, you know, in Indonesia, it was a significant portion of people that are watching.

So hitting those scales, we have had to solve technical challenges. And the technical challenges aren’t necessarily the technical challenges of how do we build more scalable infrastructure. It’s many of them are how do we deal with our users’ devices and the internet, that access that they have, right? So we’ve gone pretty far in terms of trying to use the most advanced codecs that we can to reduce the stream to make the highest quality, you know, stream that people can watch. We also have at times, looked at all of our consumers and the network that they’re on and give people different streams depending upon the ISP they’re on, because we know that that ISP is gonna be hitting sort of bandwidth constraints, right? So I’m quite proud of our ability to scale to some of those large events and the engineering effort we’ve put in to make the experience better for the user that isn’t necessarily on do the servers work well and stuff, but it’s more meeting that user at their, the technical maturity that they have on their side.

[01:09:26] How Do You Keep a Streaming Platform Stable During Massive Live Events?

Henry Suryawirawan: Yeah. I haven’t worked in the, you know, like streaming companies, especially the live stream companies. From my perspective, it must be very stressful whenever there’s a live event. It could be live sports, entertainment, whatever that is with higher number of users. Is it stressful for you? Like how do you actually ensure a large scale event go seemingly smooth, you know, less issues. Or if there’s any issue during the show, how do you actually ensure that you can recover really fast?

Tommy Sullivan: Well, I’d be lying if I were to say that it hasn’t been stressful me. have been some stressful times. But one thing that we do benefit quite a bit from, as I mentioned, that sort of AVOD, SVOD space, right, is Vidio does have a lot, we have a lot of free-to-air TV that people can watch. So kind of like YouTube Live, I believe in the States and, so forth, is people can subscribe to Vidio and basically watch the equivalent of a bunch of cable TV channels that are 24 hour running, right? And so from that, we, you know, we have that primetime boom, and we have that standard sort of, the daily sort of rise and fall of people going to bed and waking up and wanting to check the news and/or watch kids stuff in the morning. They turn it on for their kid and then they’re at work and then, you know, maybe during lunch break they’re watching. And so Vidio typically, on an average day, has about an 11x dynamic range of people, you know, the prime time people watching and then then going down.

So that’s where I feel like it’s actually been quite a bit of our benefit in the sense that when we have had big premium sports things that people wanna watch, I’m already used to this dynamic range here, right? So then, and to see your system as it scales from that 11x, then it’s a bit easier to understand, you know, how it’s gonna scale on the other side of things. And that’s because on that 11x you’re obviously gonna be optimizing in terms of not over capacity and running things very efficiently, right? So from that you can extrapolate a bit more in terms of how to run the bigger events at scale.

We have hit some hard limits there. Yeah, yeah, we’ve hit some, I remember once we had, we needed to like Redis. We were running some Redis and it was certain starting to hit sort of operations per second limits. And we had a big incident where we already sort of went through the playbooks of okay, you know, if we’re running into capacity limits, how we would scale up, you know, spawn another instance and run there. And those capacity limits weren’t tested at peak load. I was spinning up another Redis. And that was definitely an issue that we’ve run into.

Henry Suryawirawan: Right. Like one thing that I would imagine, because all these, like not all, I mean like some of these streaming platforms are now going into the live streaming thing. I remember like Netflix is doing maybe the first big live stream they had like Mike Tyson against, you know, Jake Fury, Tyler Fury, I forgot his name.

Tommy Sullivan: Yeah. Jake Paul.

Henry Suryawirawan: Jake Paul, sorry, yeah. It’s like the number keeps increasing, you know, as the, you know, rounds go by, right? How do you actually plan for such event that you didn’t even anticipate before?

Tommy Sullivan: Yeah. I think that, you know, I certainly watched Netflix from the tech side of things as that live stream, that was, Jake Paul versus Tyson was their largest single day signups ever. And I think they had around 60 million concurrent. And that’s from a podcast that the CTO of Netflix gave. And so they had those 60 million concurrent, and we have never hit that scale. So that’s fine. But what, but for, you know, Vidio scale, I think that that dynamic range, that load testing that we do, that we get to do naturally from that rise and fall, has benefited us quite a bit. Well, it, I mean, it’s mainly the fact that we have that dynamic load that comes every, all the time. And I mean, we will have, I remember before Vidio actually, when I was working at Emtek, we had, when it was just the publishing platforms, some celebrity, some art, you know, popular celebrity died, and we saw more traffic in that one day than people wanting to read the articles and all of that. So you can’t plan those deaths. But you can optimize a system at, you know, its peak in those loads and, yeah, optimize around that.

[01:14:12] How Did Vidio Survive When Other OTT Platforms Failed?

Henry Suryawirawan: Where do you see the industry is moving towards? Like maybe if you, because you have seen in the last few years and maybe a little bit of glimpse of what is coming, you know, with, you know, maybe using AI more. Where do you see the industry is going?

Tommy Sullivan: Actually, maybe I can circle on that dynamic one a bit. Oh, okay. So one benefit that Vidio gets from being, we are mostly now SVOD but we still have users who can come in and just watch sort of like TV on Vidio. And so we have that AVOD aspect, which is very lightweight in terms of people can just sign in, sign up, and watch without even subscribing, right? And so we get that, it’s easier. We get that free consumer that, and there’s obviously then a lot more scale there, right? Which is why Hotstar has massive scale. They do have tons of subscribers, but they also have a, the majority of the revenue is coming from advertising base. And I think a big differentiation between us and Netflix in that regard is to Netflix, if you are to play a single video on Netflix, you need to log in, subscribe, right? And so when they have that walled garden of a credit card or whatever payment process then that really prevents them from having to optimize for that advertising base, that is the, just so much higher throughput, right?

And I think that in the OTT space, I’ve seen sort of, I’ve seen organizations rise and fall. And the, I remember I was on the stage in Bali once at an event talking at a advertising event. It was called APOS Tech and it’s quite a fancy event. And I was on stage with, Hooq, and Viu, and Iflix, and then Vidio. And half of those companies are dead now. And it was Hooq which was a joint collaboration between Singtel and other partners in the area, I think maybe even Sony. And then you had Iflix which was invested and expanded and SVOD throughout the area. And I think it’s very difficult for those companies that start out SVOD and start out where every user is gonna be paying this much. And so as long as the user is gonna be paying let’s say a dollar a month, and so that as long as our costs are, let’s say, you know, 50 cents, you know per user or whatever, then, you know, they can optimize on. But what that does is that means that the business, and they’re gonna try and grow quickly, is they’re gonna use vendors and they’re gonna use a lot of third-party services and they wanna keep quality high. So they’re gonna use a lot of third-party services to get the service, the highest quality out there as soon as possible.

But then what happens if you were to scale into to start adding advertising as a mechanism for people to consume your service. Then you went from a dollar a month down to 2 cents. And now as you scale massively, there’s a funny phrase, it’s what we lose on margin, we make up in volume. And it’s funny cause it doesn’t make any sense and it just means that you are going to be massively hemorrhaging money. And of those on the stage that Hooq, Iflix, Viu, and Vidio, Viu and Vidio both started out on the advertising space and then we’ve moved into the subscription model. And the Hooq and Iflix and even Netflix has introduced the advertising space. And I think Netflix will, well Hooq and Iflix have died and I think Netflix will be fine. Obviously, they’re a pioneer in the space and they’ve such an incumbent. But in this market and in, you know, certain parts of the world, you know, the ability to subscribe and pay is reduced. And that’s why Netflix has added that advertising space to expand their, their business. So I think that having that high efficiency and being very critical of every dollar that you’re spending on a third-party is incredibly important to ensure the success of your company.

[01:18:15] Why Does Vendor Independence Matter for Both CDNs and AI?

Tommy Sullivan: Now, I think we’re gonna also see the same thing in the whole AI space, right? You have, and granted Claude or Anthropic and OpenAI, I think that both of those are the Netflix incumbents, right? So I think that, I mean, it’s interesting to see that, you know, but even… Who’s the founder of OpenAI?

Henry Suryawirawan: Sam Altman.

Tommy Sullivan: Sam Altman, you know, talking about his sort of, him firing back at Anthropic of, oh, we have more customers in Texas than Anthropic has customers, right? Which is it does show that scale is incredibly important and because they’re more advertising based and they’re gonna start adding, you know, more ads. And I think that Anthropic being subscription-only based again, you know, I think that right now what’s important in the AI is the quality and being able to describe. But I think that when we start talking about companies that are built on AI and are bringing AI to a very niche product market and stuff, I think that we will see some of them, you know, there are some jokes around, you know, like tell me you have an AI company and then show me your OpenAI bill.

You know, is it depends on the level at which I think that’s, it can be great to jumpstart your company and to solve the problem in that space. But I think that you’re not a tech company unless you’re thinking about every time where you’re employing AI there. And okay, maybe that could be not even an individual model. Maybe that could just, that piece of the system could be some rule set. You know what I mean? That costs, you know, literally, you know, nothing compared to tokens, right? I think that people need to, what’s the NPM module that’s like using AI to check whether something’s is odd, you know. I dunno. But you know the extreme cases where using AI in areas where it is inefficient. And so there will be so many places where companies are able to bring huge LLMs and into certain domains and evoke efficiency and help improve people’s lives, but they won’t survive forever if they’re beholden entirely to a third party.

Henry Suryawirawan: Yeah. Certainly when people refer to maybe AI bubble, you know, we see a lot of AI products out there, right? But if all they do is just like wrapping on top of, you know, whatever AI model that they use, probably there’s a bit of trouble maybe in the future if this model changes in terms of capability, the price always changes, yeah. So definitely a good advice for people who are building AI products these days.

Tommy Sullivan: Technically, one thing that, you know, one area where Vidio is beholden to is, I mean, 50% of our costs are network delivery, right? So I mean it, that is a huge aspect. And initially we used one CDN and optimized on that, and I think that it was painful to go multi-CDN, but it unlocked the ability where I wasn’t beholden to a particular vendor and I was able to say, well, I can just move this over tomorrow. And so I think maybe in the AI space, we need to think about that, is if you do use, you know, AI and you’re not running or building your own LLMs and things like that, is maybe being vendor agnostic certainly unlocks the skill of being able to negotiate between vendors.

[01:21:44] What Should Engineers Understand About the Agentic AI Trend?

Henry Suryawirawan: I think you have mentioned a couple of potential trends coming. Like for example, I like the discussion about the subscription versus advertising, creating more AI-generated content or AI being used in generating content in software development as well. Is there any other trends that you pick up that probably you can share for us here?

Tommy Sullivan: Given the current date and time right now which is, what? We’re mid-February in 2026. I think that it is impossible to avoid the recent trend of the OpenClaw and the whole agentic AI.

Henry Suryawirawan: Yeah. That’s the tie that you have there. Yeah.

Tommy Sullivan: It’s a lobster tie that I’ve had for a very long period of time, but it is identical to a lot of OpenClaw logos.

Henry Suryawirawan: Right.

Tommy Sullivan: And on my desk, you can see this Mac mini and I only have a stack of one, unlike the people who are trending on X who have, you know, 15 Mac minis all of that. So, I mean, it’s a lot of fun to play with, and it’s interestingly an exciting area to be in with regards to that agentic, letting, you know, some agent run along and burn some tokens until. But I think that the trend needs to, you know, be focused. People need to be cognizant of the Gartner hype cycle. And it’s a great time to get experience and to play with all the things that are developing in AI. And I think that it is foolish to ignore them and to not, you know, experience and see where their breaking points and their limits are, right?

So I have certainly, and when you play with these things and when you technically have an understanding of how they operate, right? And, you know, nothing is magic in tech. And I think that once you get a deep enough understanding of, okay, this agentic. The whole OpenClaw thing, right, is essentially, I mentioned earlier how, you know, we had the Copilot and then we had Cline that was feeding back into itself and getting better, right? Getting better based on markdown files, right? The limit is not sentience, but markdown files, right? That’s essentially all this OpenClaw is, right? Is markdown files and potentially vector, you know, embeddings and search and stuff like that. But it’s not sentience. And at least understanding how it, the underlying components of it, can help you leverage that for what it’s good at. It’s good at, you know, I have my own OpenClaw and the first things that my friends want to do is, you know, some, book some time of Tommy needs to be here at this time with me or whatever and to remind me. And it can be very useful for getting different perspectives on certain things. But if you’re gonna prompt it to go online and make money or something, and then I think that you wouldn’t have a good enough understanding of this is not magic that you’re talking to, it’s a bunch of markdown files.

Henry Suryawirawan: I like it that you pick up this OpenClaw. Some people know about Clawdbot, Moltbot, Moltbook and all that. But this OpenClaw is definitely one trend that is happening in the tech world. And in fact, the founder was just hired by OpenAI. Maybe we will see new evolution of this OpenClaw. And yeah, definitely interesting for everyone to just understand how these evolve. The AI pace is definitely very fast. But at the same time, I think we have to, I mean, for people who are in tech, we need to understand where it’s going, the limits, although it seems like the limits keep increasing day by day.

Tommy Sullivan: They do. But if you have a deep understanding of it then, you know, if it is, if AI is magic to you then it makes sense to potentially ask, you know, like go make money online or go cure cancer or something. But like when you understand that it’s just predicting the next word in a very oversimplified manner, then you can have a better expectation of what is possible, what isn’t possible, even if this word prediction gets better by, you know, several years, right? What is still possible.

Henry Suryawirawan: Yeah. Speaking of that, I remember one quote that I have with my past guests. Like AI is smart until it’s dumb. Like one day, you’ll find it dumb. It’s like, oh, it’s not magic after all, right?

[01:26:17] Tech Lead Wisdom

Henry Suryawirawan: So Tommy, I think we have covered a lot of things. Unfortunately, we have to wrap up pretty soon. In my podcast, I have a tradition to ask all my guests, you know, this question called the three technical leadership wisdom. Think of it just like advice that you want to give. Maybe for people, engineers, tech practitioners out there. Maybe you can leave your version today. That will be great.

Tommy Sullivan: Yeah. So I’m a huge fan of Malcolm Gladwell. And I’ve read all of his books. I love his books, right? And in one of his, I think it’s The Tipping Point where he talks about how practitioners who become excellent in a particular area, you know, the 10,000 hours, right, and how Steve Jobs and Bill Gates, you know, at that time when they were, they happened to be really at the right place at the right time in terms of having access early on to technology. And then obviously they took it upon themselves to play with it and learn and love and, you know, just absolutely pour time into and then get those 10,000 hours. I think that mastery of, you know, any skill takes time and I think that as people who embrace, you know, our new LLM overlords. But people who have successes and failures and leverage these tools and then see what works and then build on that. I think that that is as true as it ever was. And so these tools are something that, you know, shouldn’t necessarily be feared, but people need to be respectful of the Gartner hype cycle of not getting too overzealous. And keep focusing on the outcome of is it helping you get done what do you want to do, and double click on that. And double click on solving real people problems as opposed to optimizing something that, you know, or creating something interesting.

Henry Suryawirawan: Any other or that’s it?

Tommy Sullivan: I think those are the, I think that’s the only one I have.

Henry Suryawirawan: Thank you so much for sharing that. I believe mastery will, the definition of mastery will change definitely with AI, right? Because there are so many things that AI, and in fact, for example, if I want to get into a new thing, new domain, you know, new understanding about something, it’s very easy to pick up from AI. But definitely mastery here doesn’t mean that I know more things now because of AI, right? Mastery is something that you have done, experiment, fail I think it’s a big part of that mastery as well. So I think definitely the future is gonna be different.

So yeah, Tommy, thank you so much for your time today. I think I learn a lot about, you know, your practices, building culture, which I think is quite admirable. You know, Vidio is I think one of the big tech product within this region. Wish you all the success out there. And yeah, thanks for sharing everything that you have today.

Tommy Sullivan: Thank you for having me. It’s been very a learning experience for me as well, chatting with you. Thank you.

– End –