#15: Escaping Dictatorship, Hacking the Internet, and Building a World-Class AI Platform
40:29 Share

talk about it openly now because the regime in Syria just collapsed so I have nothing about it. So I end up hacking the internet, shut it down for two weeks. So you know, this is how that you know my revenge you know. The internet just in your city or in? The whole country. Welcome to Divot, a community for people trying to make their mark on the world where each week I'm interviewing the best minds in business, tech, sports and entertainment to see how they made their mark.

Go to divot.org to see all the episodes, or you can watch them on YouTube, Apple Podcasts, or Spotify. Divot. This episode is brought to you by Salesforce. You're in Syria. You're very young, teenager. You're putting digital maps inside of probably high-end cars.

And, I mean, how was it received? Did the product take off? Was it only in Syria, or were you doing it in other places as well? Actually, it was only in a few specific cities. It was only in Damascus. We never have the capacity to digitize the whole country. So we actually took the three biggest cities, and we started with Damascus.

It was amazing. I made a lot of money. I felt like I was the richest guy on the planet at that time. For six months, until stuff started going south, basically. I got this amazing call from the government, and they demanded me to hand over the company. What was their reasoning for doing that?

I'm not sure how much you know about Syria, but when you make a lot of money there, when you're not part of a specific group, you will be highlighted. The official reason to me was that you're not a trusted person in the system, so you have to hand it over. So did you...? No. I told him, "Nope, definitely not."

Now this was my initial answer and I was not too smart. They put me in jail. Four or five days later I stopped counting. I was like, okay, I'm ready. Just take everything. I will sign anything. I'm not fighting back. And they offered me one dollar to hand over the whole company. One US dollar. One US dollar. And that was actually fine. I understand the system stuff challenging. What really actually...

made me a little bit upset. They didn't even give me the dollar. I'm just serious. They said like, if we're going to give you the dollars, it's a foreign currency and you're not allowed to have a foreign currency. So we can give you the dollar, but you go back to jail. And I was like, okay, not happening. So you agree to get out of jail for a dollar. The government refuses to hold up their end of the bargain of stealing your company for a dollar. They want it for nothing. So then...

Are you just ready to start your next company or what happens next? You know, in the old days when you'd be 18 years old, full of energy,

No long-term thinking whatsoever at least for me at that time. I decide just for revenge so Three weeks later I end up basically and I can't talk about it openly now because the regime in Syria just collapsed So I have nothing about it So I end up hacking the internet shut it down for two weeks So, you know, this how they you know my revenge, you know the internet just in your city or in the whole country. I

So you start a mapping software company, they come take it from you and put you in jail. How do they feel about you shutting down the internet in the entire country for several weeks? Man, you call me, you know, this is all boomer is now, but...

still give me some chills just thinking about it because, you know, shutting, it took me two weeks to shut it down. I have no, I feel no guilt about it because most people have access to internet was not actually the average people was, you need to be like very connected to the government at that time. This is like early 2000s. Yeah, early 2000. And, you know, three, four hours later, the security service, what you call mukhabarat, the shop at my front door,

And I said, "Cool, okay, I'm going to jail. Nobody's going to see me again. I'm done." Surprisingly, he was like, "Okay, we have a profile about you here. You're good with computer stuff, and the country under attack, and we need you to catch the guy." So yeah, for two weeks, I did not manage to catch myself.

Were you working at home or did you, were they have you in a facility somewhere working? No, they took me to a secret facility, you know, somewhere when you have access to direct to the server, to the routers, one bunch of computers. And I stayed there for quite a time. They check on me every two hours. It's just like, what's going on? Can you catch them?

And two weeks later, I just saw they actually brought experts from different countries at that time. Then I know it's a time to go. The internet comes back on. You realize you need to leave. And so, I mean, did you get on a plane? Did you drive across the border? This is even before because it's still everything down. I'm trying to fix it. Seeing a bunch of experts showing up. They kick me out because I'm not qualified, you know?

And I knew it was going to be just a matter of hours or, you know, to be something happening. So I took, you know, taxi, car, and I drove to Lebanese border. I went to Lebanon when actually I stayed there when I studied also my computer science. Yeah. And they sent me a message. If you're coming back, don't come back. If you come back, you're going to be in there. They did figure out it was you. Eventually, yes. And the message was clear. Don't talk about it. And don't come back. Yeah.

Yeah, that's a pretty good trade. Don't embarrass us and you can live your life somewhere else. Oh yeah, like the old regime in Syria, for sure. It's a great deal for me. Yeah. How did you end up in the United States?

Short story or long story? You shut the internet down in Lebanon? No, Lebanon is such an amazing, nice country. I spent a lot of time there. It was also challenging to me because we did not study anything English in Syria. So I show up in Lebanon. I want to study computer science. I know the ABC, and I can't count 10 in English. That's all that I know. And now I want to study computer science. And surprise, the first thing, guess what? Everything in English. That's supposed to be.

So I checked everything, what I want to do, looks like I have to do English courses for two years, and after two years I can study other four years of computer science, which is to talk about now like four to seven years if everything went perfect. So she's like, no, I'm not doing this. So I end up taking six months of course, just studying English for six months, you know.

Then I ended up basically, I said, "Let me put forth it. Let's go to college and I will take this book and we'll just figure it out and read it." The result was not a great first year. Actually, I failed almost in everything because I'm just a slow reader. I cannot go through everything. I ended up creating my first, actually one of my most popular open source software in LLB. I created an auto-summarization system just to take like a thousand pages, summarize it, and I only studied the summary.

So, a thousand pages, I only study like 100, 200. I graduate.

This is the end of Lebanon. They wanted to buy work for your companies and then move to United States when we started Rectus. How did you get into AI? When I published my first library in the college was actually one of those first early open source library in NLP processing. So we can actually take a lot of large amount of text and you can do full summarization. Was pretty simple system, very statistical model based implementation. I kept it open source.

And when actually I studied, I continued with my master in Jordan and in Dubai, quite a few companies reached out to me based on my open source repo, and they started asking me for enhancement, they started asking me for actually specific type of model. This is when I started actually building early version of what we call AI. Sorry, at that time we don't call it AI, we call it machine learning, you know, the cool name today called AI. But it was a lot of statistical model, we kept building a lot of those.

The real first, let's say, more like generative AI experience was started when we started the company. This happened in 2019 when I met my co-founder, May. We've been experimenting with a lot of statistical models, but then I was impressed a lot with the new encoder-decoder systems, the T5 models. This is when we started building our first generative AI models in 2020.

This is before OpenAI releases ChatGPT. Way before. This is when we start building our smaller models because we already built a bunch of statistical models at that time more related to translation use cases. The only thing was we thought that those statistical models can do really well. But then when you start actually building smaller models, the encoder-decoder type of system, we start figuring out the capabilities could be way more. We start with the text generation.

And then we find out actually if you scale those models, and this is basically 2021, 2022, when start being more like open knowledge, you know, more like open secret. If you scale the models, you'll get more capabilities, more emerging skills, and you can actually build more, let's say, feature on top of it. This is Palmyra, is that right? Yes, Palmyra. Yeah. So the original problem that your startup was trying to solve was around

Translation, is that right? Just language? So this is the stuff I've been always fascinating about is language. You know, I always been one of this difficult thing in my life, language, communication in general, you know, at least my wife tell me. But

Building those languages models was something I'm fascinating with. This one we started, the first use case how we can actually build implementation was what you call localization. So how we can take text and convert it to different type of text using some kind of machine and maintain the meaning. This is when we started. Anyone that's dealt with large scale localization or globalization projects knows that this is a

horrible problem to solve. It is such a pain. It is expensive. Sort of pre-AI. And I think with the sort of advent of AI, having spent hundreds of thousands of dollars on localization projects in my startup career, it was so obvious to me that localizing content was one of the first and most obvious use cases. So it's

It's really incredible that I think that you all, the problem that you were trying to solve at that time ends up being maybe one of the best problems to solve for the sort of first use cases to have in first revenue impacting use cases too, I think.

of all the things that you could have been trying to solve, that was the one you all happened to be solving. Yeah, totally. And I think also this is what's helped us and pushed us to go more with generative AI path.

Because it was clearly, when people think about translation, they only think about it as basically, it's like you mirror the text. In reality, you're regenerating everything. You're actually regenerating the text in a different type of, sometimes same meaning, but different type of, let's say, words. Let's put it this way.

So that challenge is still, you know, it's really big and help us a lot to start discovering different type of technology. And this is what we, you know, I love just those encoder models today. Like really they generate everything more or less from scratch. And they give you this amazing meaning and creativity. Most startups and early stage companies are building on top of the existing technology.

LLMs that are out there you you decide to build your own and you have your own why why do that it seems like it's a it's a big decision uh and it's had a really positive impact on you it seems like or on the company and your customers but you know it might have been much easier just piggybacking off of what somebody else had done and running with that why didn't you do that yeah uh

Great questions. And I'm going to answer it from our point of view, how we think about it. Because I don't think there is a right way or wrong way using third-party model or build your own. But from our perspective, the way we've been actually thinking about it, we've been trying to say, look,

We work with, from day one, and this is from day one, we work with enterprise customers. Those enterprise customers, the only reason, the only reason they work with a small company, a small startup, because you can innovate fast and you can help them see that innovation and apply it. They don't work with us because we're safe. They don't work with us because we're secure. They work with us because they cannot sometimes move fast enough and we can actually build faster features for them to implement.

And in the early days was actually was what, GPT-2 and as we have the Finchee, none of this was stable. So we needed to have our own models. Later on in 2023 when stuff start escalating, we start seeing more and more models, it was clear there's actually nothing we can consider what's called enterprise ready models. Like something simple like roadmap, like how I can go to my enterprise customer and tell them, look, for my next features and to fix the next bugs,

I don't know when, I'm going to go to Twitter and I'm going to keep reloading the page because someone going to tweet something about new models and maybe Sam Adelman, he was hinting there's something coming internally. So this is the new features.

It cannot work. We need to actually provide something clear roadmap, which bugs you have, when you fix it, when you deploy it. You need to have something simple like backward compatibility. Can my customer build into my technology and deploy? And when I have a new version of new models, it just work today. And this is one thing like I will point it out. Every research lab in the market, they do not maintain backward compatibility. Every agent you built with those models, at minimum, you have to do full evaluation and adjust the prompt.

In reality, this is one thing we do. When you have using Balmera models, if you have the stuff working, it just works from there. And I can give you more stuff, like a long list, but I can do a lot of things to mention it here. Something like what's called Transparently.

Can you actually be transparent with the customer? When enterprise work with us today, one of the things we provide, what's called a model card, we'll tell them how we build those models. We tell them what type of data distribution, we tell them how we fine tune it, which trick we use. They need to feel safe to scale and build with you. Cannot be black box, cannot be just a trust me, cannot be, oh, it's a secret, cannot be access. No, those models have a lifespan in around six months before they get deprecated.

At least in those six months, just tell them how you build those models so you can rely on it. They can scale with it. But this is at least from our point of view was we have a demand from the customer to get something stable, and this is why we're building our own models. This is one thing that's pretty unique about Rider is that you are getting enterprises to buy this software, and it's all AI-driven, and as somebody that's

sold into enterprise and has tried to sell the AI versions. There's many companies that just refuse to buy any AI enabled software. And so like what,

I love what you're saying. It kind of makes sense why people are buying it, but tell us explicitly why do people choose you over enterprises specifically? Why are they comfortable working with you where they're not comfortable with, in some cases, any other AI solution? Yeah. An AI industry today will see stuff moving fast. And the reason we see it change like every nine to 12 months.

Three years ago, every company wanted AI because they want to play with the stuff and think it's cool. Last year, we'll see a lot of move around. Actually, let's catch up with it. It's important for us. Recently, what we're seeing, at least in the last nine months, is a big movement about focusing on solution.

This is how we sell our software. We don't go to them and tell them, "I have another tools for you to figure out. I have another API, another SDK, another framework." They're overwhelmed. They see a million things, open source, closed source. Everyone comes to them, they give them just another tools to build and figure out. What really actually want, they want a solution.

Can you come to me, actually speak my language? Can you show me something clear? And this is our offering. When you have the full stack, I'm not relying on a third party models. I have the model, I have the lock system, I have the agent builder, and I have the agent itself. So when you go to customer like, hey, you want financial services? This is 10, 20, 100 use case ready to use from day one. So they can take it, implement it, customize it after. So definitely that

focus from the solution point of view that's what actually helping us today maybe you could just step back for one second and could you tell us what problem does the product actually solve right up today it's basically a platform for enterprise what you call it the end to end to end the platform helping custom we're helping our customers specifically in CBG reach and CBG and retail financial services healthcare building and implementing agent at scale

Now, I know I mentioned too many fancy words, AI, enterprise, agent scale. But in reality, in the world today, we have a solution that can increase productivity as internal tools. So any company that works with us today, the way you think about writers, you deploy it and you will have what you call spaces. So you have a space for your writers.

for your products, for your marketing, maybe for your claim review in the healthcare, maybe for your financial risk management in the healthcare. And each one of those space have their own customized agent. They can run customized monitor. But eventually the core value, it's increasing productivity, which is what the promise of AI, right? This is what we promising the whole market for the last five years. We're going to increase productivity. This is what we're trying to deliver on.

So I come to Rider, I give you access to my data, it's a trust, you're holding it, you're not training against it, you're not sharing it with anyone else, and then anyone on any of my teams can feel safe and have access to

improving the way that they work or different specific ways that they work because you're trained against all their documents and their internal processes and if I'm in HR, it's my HR stack. If I'm on the engineering team, it's the coding side marketing. It's all of our marketing materials.

And it's just kind of your training, the model kind of boxed in against its own data. Yeah, so definitely we do not share the data. Each customer have their own. This is one of, you know, I think we learned early that having something called like air gap deployment or having single tenant is very important when you have like not just what's called like we should have a physical isolation to those instance is very important for the customers.

At the same time, just get them started. On-prem? Are a lot of people having a physical place or is it all in the cloud? All in cloud. We do not work on a prem. It's all in the cloud. We still need GPUs. So all in the cloud. But we still provide some of the core most single talent so you can have your own instant that physically isolated and specific, let's say AWS account as example. So you can have more control around the network around it.

In the same time, your team have what's called a pre-built agent. From day one, if you are HR, you will go there, you'll find 50, 60, or maybe 100 agents. But also, we have the second part of the cycle, that not just the agent, it's the IT.

because your IT team also part of it so IT will can now have a way to monitor control can see which agent could be accessible by which team they can decide if this agent actually could be access which data so you have the IT team part of this basically being the HR for agent you know and we we have that your team having the their colleague which is the agent at this time

And do your agents work across any tool or is it? It could work across any tools if you put it this way. We have a framework actually, it's an open source framework.

we allow you to build any agent. We provide no code and we call it low code deployment, so you can build any kind of customization. And also we have an option you can use that one of our coding models, you can generate the code for you for your agent. So just connect it with any tools. So in reality, the way we think about it, we have a multiple level of agents. So we call it, for example, like L1, L2, L3, L4. So you have simple agent when you have a prompt,

prompt with some input, like summarization is a level one. We got some more complex agent, which is stuff required, like connect your Google Drive or connect your SharePoint, and you can do question answering. Then you have a more complex agent, the one can actually connect to data source, but can do action, like read my...

my email, but then send a notification or go update Salesforce record. And then you have what's called multi-agent. So we have that agent that's called multiple agent. Could be one of them symbol agent, could be one of them action agent. But, you know, I know it sounds a lot of agents here, but this is one, what we try to avoid when you talk to customer. We don't go to different, I have all this agent. Like this is the solution, this is the problem, this is the solution, and this is productivity. Are agents the killer enterprise?

app for AI? Is it the killer use case? Because there were sort of these early iterations over the last few years and then agents have kind of like started to catch on and I don't know who started talking about it first. I feel like Mark Benioff tried to like, you know, say, hey, I invented this. But now it's sort of this ubiquitous term for getting work done with the help of AI. Is

Is this just something that's here this year and will be something else next year or the year after? Do you think there's something sticky about this concept of agents? So far what we've seen and we see it is very sticky because I think agents now start basically...

encapsulate this concept of like could be anything could be background agent something just run in the background you don't see it at all you have no you know and do some work or could be more conversational agent like with a chat interface we'll see a lot of people today they prefer the chat interface but we also seen big movement to go and more was like I trust the agent they're doing their job can

Can't you just go around it in the back end? Can you just report back to me when you need some kind of help or guidance? So I think agent is going to be just basically this concept of there's something getting the work done and whatever the interface looks like, it's just the details.

How does the, we see this with a lot of new technologies that the business model evolves and we hear a lot about business models evolving around AI. Can you talk to us about where the business model started and what it's evolved to today? How you all, you know, how you charge customers or, you know, and is that what you came up with in the early days or has it changed? I think it changed, I don't know. But I think one of the things, it's feel like you,

You need to keep working on it, especially in AI. Markets move fast and you have to adopt for it. In the early days, we have a little bit more complex pricing. We have stuff around the seat and then we have something around token, and then we have consumption on top of it, and then we have ABI call. Today, we start to try to simplify it because what we start figuring out, token is cheap.

And customer cannot measure token. You cannot just go to them and tell them, "Oh, you need a million tokens." What does this even mean? I'm an engineer, man. I wrote myself a lot of tokenizers. I don't even know how I count the tokens. Like, what's a million tokens? I don't know what this even means.

So we're trying to get more about now seat and a platform fee. I'm going to give you the platform with a bunch of agent. If you know more agent, more customization, more security features, access to more users will increase the fee. I truly believe the stuff going to get simplified even further and get different because by end of day, it just needs stuff. You can link it to the business value. Like your customer need to look at it and need to figure out the simple math.

how much I'm paying you, what the benefit get from there. And this is what actually you've been seeing. I discussing it. Bless you. There's something actually been talking about it with a few friend the other day. We're seeing a lot of CFOs in the deals today. It's something very unique.

Usually you see CEOs, CIOs, CTOs, you see people from marketing business, but we literally actually recently, especially the last three months, we'll see a lot of CFOs literally coming and look at us like, hey, if I put you this, what the saving looks like? What a productivity to my team? How this will affect my actually my hiring plan or increasing plan?

So it's been very interesting and definitely that go back to your question, like is going to change how we think about the pricing. This is a natural thing that all startups need to embrace because

you have these ideas of how you're going to charge and maybe you've seen someone else charge that way, but actually charging in a unique way, even though it's kind of hard to explain in the beginning, it can be a really important differentiator. And it kind of shows that you're, as you listen to customers, you're like, you're being more sincere because you're not doing this old, you know, way of doing things. You're kind of being flexible and you're listening and you're tweaking and you're saying, you know, we could charge like that, but actually we charge this way because it's actually better for you.

And, you know, it's kind of confusing at first, but pretty soon I found just talking to customers, they really resonate with it. If it makes sense, if it's thoughtful, and if it actually like is better for them. And you can end up when they go in for these, you know, you're competing against somebody and it's like, well, you know,

this is how everybody's doing it but there's just one cool startup that's like doing it in a totally different way but actually like i like it uh and now that and it's just like it's almost feeling like you know it's apples to oranges versus oh i can't you know as a small company i can't compete with this 10 20 year old company that has all these features you don't need to like you you can compete on different things definitely making the stuff more straightforward getting the stuff actually uh

the way you can just measure it will make a big difference.

Even something like go back to the token, when everyone pricing the token last year, and we can even including us, you know, when we start actually just making our pricing a bit different and simplify it, we did not get pushback at all. It was actually very positive feedback. Everyone's like, oh, finally someone will tell me token. Oh, finally someone is going to ask me to do exercise to measure my token, you know.

we started getting this and it was very plus, very big advantage. What do you think the future of AI looks like in relegated industries like finance or healthcare? How is it going to be different? I think regulation in AI is still early. We still try to figure out what regulation looks like. I think a big part of this is we cannot comprehend

the capability of those systems. We build the models, but those models, they look different every six months from the capability side. So you have a regulation, you need to understand what you're really dealing with. And since we're still having those actually, like the new generation of model every six to seven months, sometimes three months,

It's significantly different and more powerful, regulation is still early. What we're seeing today, at least with those industry like in financial services and healthcare, they try to take the safe part. Can I have at least some kind of control around connectivity? Can you make sure my data not accessible by everyone? Can we make sure we have some kind of guardrails for the model itself? Can we make sure we have isolation?

which is, this is really like fair request and we should have it. I don't think it's enough, but definitely I don't think we can regulate the system yet. Like we still have to wait to get more,

you know, let's say, stabilize. There's a lot of innovation and we still have a lot to build. It feels like every day you refresh X and it's like some new thing and you're probably in Slack and you get a message from somebody. It's like, did you see this? And it's like, you know, you kind of hold your breath, I would guess sometimes. How do you balance like

Staying ahead and iterating on what else is going on while also staying true to the path that you've been on and are projecting to be on.

I wish if I have perfect answer for this, but today the way we do it, because it's like everything moving fast. You cannot comprehend everything. You cannot analyze everything. You cannot be ahead in everything. But what I can do, I can focus.

This is when I started representing Reiter, we say we're focused on financial services, CBG and healthcare. Even in those industries, we're very focused on specific use cases. Because now, what I can do, I can stay ahead in those specific industries and those specific use cases. Now, does that mean if someone else is coming from a different industry, they want to try and build different use cases? I would say no. Definitely, I can't do it and build it.

But from my roadmap, from the focus, from the implementation we have, we'll try to stay ahead in those specific and go as deep as possible in the implementation, in the solution to stay ahead. Because it's almost no one ahead today, like no one ahead in everything. There is no company literally have the best in everything. And even if we have the best in everything, how long will it stand? For two or three days, maybe a week maximum? Yeah.

How do you decide who to listen to and whose things to prioritize and who's not?

The cool thing with AI today, we'll see a lot of similarity in the feedback. In the early days, it was harder. When you have basically three to four to five customers and all them enterprise, this is when you start getting more conflict feedback sometime. I will not call it conflict. I'll say you get a lot of feedback and a lot of demand, but you don't have resources. But today, when you go on your resources, when we, us internally, we use AI to write most of our features.

What we're seeing from all our enterprise, especially focusing on those specific industries, we'll see similarity. We'll see a lot of specific feedback and features we have to build. So it's not that challenging to get easier, especially when you acquire more and more customers, but definitely is way more challenging when you have like a few customers. How do you think about

open source versus proprietary inside of AI? Are there clear lines of where you think it should be open source and where not? And for your own company, what areas that you need to own versus... The short answer I think should be open source. This is the short answer. Now the question, what of open source? Because usually when you say open source, people will jump to like, oh, is it a free Apache 2 MIT open source?

Not necessarily. We have a bunch of stuff in MIT license, Apache 2 license. We have stuff also open source just for our customer. But the idea behind it will go to the promise of AI. When you work in enterprise, and I'm sorry, I'm saying a lot of the word enterprise a lot today. But when you work with these big companies, you promise them to increase the productivity 1,000%, 2,000%.

They need to trust you. They cannot be trust me because just, you know, look at me. I'm nice. It's just going to happen. They want to rely on your system. They want to see how you build it. So you have to be flexible. You have to open source it. To open source it just to show them how they build it, to feel confident they can build and scale with you, you know, what if right now today decide to change direction? What do you think? You tell our one of like our, you know, huge cost, like the fortune, you

200 we have today and our customers, oh, good luck, rebuild everything. No, they need to have a vendor lock and a plan. They need to figure out implementation. So the short answer, it must be open source. It needs to be open source. The question is, what is open source? Who can access it? And when you look at some of these big companies, everyone, every large company in tech is going after this in some way. How does that feel? Because

You know, you've raised hundreds of millions of dollars, which probably for most of us feels like a lot. In the AI space, it's not the most, right? I mean, you're still big, but, like, there's mega big fish. And then there's all the public companies that are mega, mega big. And...

Like, it's just, it feels daunting. I mean, even my own little pond feels daunting. But like, you're swimming against all these huge fish. Like, how do you feel about that? Does it scare you? Does it get exciting? Does it, like, both? Is it, I mean, do you sleep? Like, are you, is it...

When you decide to start a company, you have to forget about few things. One of them, sleep, for sure. So it's not regardless which industry.

But it's definitely to me is exciting. I just I love the challenge. You know, I love when I go to market with a huge competition. I will get very worried if I'm going to market with no competition. I'll be like, oh, there's something wrong here. But going to big market, massive competition, big mega company, let's call me like Microsoft and Google and everyone trying to take a bite of it.

we're on something here, you know? And I have the advantage. We have the money, you know? We have the market and we have the technology and this is what, this is the thing that makes me super excited every day. It's a unique time in history, like, we not just have the company and the time, also we have the

demand at the technology. Like how many companies started when the technology was not ready? What imagine if we don't have like the A100 or the H100 GPUs like where we'll be today? Imagine if we have everything but don't have the transformer. Imagine we have all the technology but there is no demand and the customer is still thinking about you know statistical model and and we're excited time and it's just like everything happening right now all what you need just to stay focused work hard move fast. It's a clear recipe.

Is it too late to start an AI company if I am not already swimming in the startup world? Never. Never late. There's always a huge industry. We're rewriting the whole software industry. The traditional software industry exists and there's still a huge gap. There's a lot of things. If you want to think about it, you see a lot of AI startups, but when you start to actually zoom out,

You don't find it that much. You find actually the numbers is not that big. You find a lot of, let's say, legacy company, they rebrand to be AI startup. But AI native startup just starting today, I don't think we have enough. I think we need more. What is the difference? Somebody is starting AI native. When you say that, what does that mean to you?

The way I would think about it is a company from day one build their solution by top of AI infrastructure to provide specific agent tech or AI solutions. Today what we're seeing in the market, we'll see a lot of, for example, CRMs. They brand themselves as agents. We'll see stuff like search or workspace search. I say we're agents.

In reality, agents can do all those tasks end-to-end and they can build it. You don't need any of those extra tools. You can just power it by a simple agent or a strong LLM.

I know I'd not explain it well, but like it's simply we have a strong AI system. Did you, can you build clear solution top of it? You know, you co-found this company. It's just you and a couple of people. Now it's this bigger company. How has your role changed and how has your sort of style of work changed? Do you prefer it now? Did you prefer it in the early days? What's, what's different?

I have a lot of mixed feelings, depends on the day and depends on the time of the day. You know, in the early morning... What do you mean? Yeah, in the early morning I miss, you know, the old days when I can't write code and just can't approve my own PR. Yeah, that's it, yeah. And then, you know, remember I don't have enough money to hire engineers, so like, I love it now, we have money to hire engineers, but I spend most of my time in recruiting with the customer, in meeting, reviews, you know, so...

Not even reviews. They don't even let me review now. They let me, like, you know. No, I said interviews. Interviews, yes. Interviews. I thought you talked about BR reviews. No. They're not letting you do reviews. Not even touch it. Either one of my engineers will kick me out or we have a bunch of AI who will kick me out. So either one of them. Yeah.

But again, just still to me, just it feel lucky. We feel blessed and, you know, and excited to just keep working hard and fast. It's just a unique time. And, you know, we it's one of those moments in the human history that we can literally make make difference. Let's give a big round of applause for Wasim. Thank you so much for being here. On next week's episode.

And my fundamental belief is AI is going to team up with humans to change the way we work, live and play.