AI Layer for Data Security | Rehan Jalil, CEO of Securiti

Rehan Jalil is the CEO of Securiti, a platform that enables the safe use of data and generative AI. They've raised $156M in funding from investors such as General Catalyst, Mayfield, and others. He was previously the CEO of Elastica, which was acquired for $280M by Bluecoat. Before that, he was the CEO of WiChorus, which was acquired by Tellabs for $180M.

Rehan's favorite books: Good to Great (Author: Jim Collins)

(00:00) Introduction 
(02:14) Founding Securiti and the Evolution of Data Privacy
(06:08) Why Data Security Needs a Unified Platform
(09:32) Scaling Challenges and Product Decisions
(13:17) The Role of AI in Data Security
(17:20) Navigating the Enterprise Sales Motion
(21:56) Go-to-Market Lessons from Elastica to Securiti
(25:43) Competing in a Crowded DSPM Market
(29:00) Shifting Buyer Personas and GenAI Adoption
(32:11) Rapid Fire Round

--------
Where to find Rehan Jalil: 

LinkedIn: https://www.linkedin.com/in/rehanjalil/

--------
Where to find Prateek Joshi: 

Newsletter: https://prateekjoshi.substack.com 
Website: https://prateekj.com 
LinkedIn: https://www.linkedin.com/in/prateek-joshi-infinite
X: https://x.com/prateekvjoshi 

Transcript

Prateek Joshi (00:01.665)
Rehan, thank you so much for joining me today.

Rehan Jalil (00:04.962)
Ravik, thank you so much for hosting. It's my pleasure.

Prateek Joshi (00:08.813)
You founded companies before with amazing exits. Let's go back to 2019 when you started security. What problem did you see that compelled you to launch the company?

Rehan Jalil (00:27.018)
I back at the time, just before that, myself and the team that actually helped start this thing, we were at Semantic, actually running the cloud security business and CASB and all. And I think one thing was becoming very clear was that while data is at the center of everything, everything that businesses do and...

And frankly, that is the center of most attacks. Of course, you have to penetrate many other boundaries to get to it. There was not one comprehensive platform that would provide you complete visibility on what data assets you have, which literally represents all your business, and provide one place to actually do all fundamental controls on it. There are parts and pieces of different products existed.

And we also had this thesis that two things, one was the scale of data is growing so fast that this requires innovation. And also, although this whole generative AI, the way it sees today did not exist, it was very clear even at that time that you can apply at that time what was modern AI technologies to crack the problem of better understanding the data, classifying it better and all. That really was the genesis and thesis that you need a platform.

provide more diverse controls in one system. And that's how it was literally how it was started.

Prateek Joshi (01:54.253)
Amazing. And for people who may not know, can you explain what security does and also, so what it does today and also what vision are you working?

Rehan Jalil (02:07.342)
Absolutely. Think of this way is that whether if you're trying to do your any AI project today, if you try to do that inside the enterprise, you can bring the best AI models from a variety of different places. And they are certainly magical, but they are trained on public data. When you're trying to bring that into the enterprise, you have to give it

permissions and apply to your proprietary data. And while you're doing that to the proprietary data, you have to make sure that it is safe to use. Which models can touch what data? When the data goes to these models, is it clean and sanitized? When models generate the answers, who can see the answers? What kind of answers are allowed? What kind of questions are allowed? The entire thing around

providing visibility on the data and providing full control on data is something that we provide. Of course, I have to say that AI use cases are now, but there's requirements on controls and security and privacy and governance data. They are not new. But what has brought it to the center of the attention is the advent of, know, of generative AI and its needs to use unstructured data.

Unstructured data was never used that way before. So at security, what we provide is what we call a data command center, which can provide you full visibility on what unstructured data, structured data you have, what is sensitive inside it, and allows you to actually, within this data command graph, you understand what models are there, what agents are there. It also gives you understanding of who has permission to do anything on this data.

and provides the capability to put controls on it and check if internal rule books of usage of the data, it is being honored or not being honored. And if not, then you can take remediation actions on top of it.

Prateek Joshi (04:12.909)
Amazing. Taking a step outside the company and looking at the data market as a whole, you've been in data for a long time and you've seen a couple of cycles. Today, what's the state of play in terms of data infrastructure? Meaning what layers exist, what layers are becoming more important, what layers will become redundant? And if somebody is coming in new to this, how would you explain?

the state of Flay in data today.

Rehan Jalil (04:45.356)
I think the data side, at least the way we were focused on, every layer is getting more more innovative. One was purely, if you think about the scale of data, what was needed. I will put the lens, because I know you very much focus on the AI side of things, so I'll put a bit of that lens, because there are so many other requirements, so I'll put that lens. When you are trying to apply AI,

on data, of course you can apply it on structured data, which traditionally all the businesses have done, that machines could actually process structured data meaningfully, applying machine learning, and meaningfully would be used for all kinds of business intelligence use cases. That's where you see big data bit houses and know your snowflakes and data breaks of the world. That's how it got all started.

What has changed and what needs to change more and more is actually the unstructured data. Because previously, if you look at the use of unstructured data, it was produced by mostly humans, right? And it was consumed by humans in the sense of intelligence layer. But with the LLMs, you can extract intelligence out of these unstructured data, which was not possible before as easily.

And now it's a mainstream. I mean, that's the main thing right now. And of course, you can create more unstructured data, not humans, but you can. And there's just lot of actually intelligence and wisdom in those things. That has changed as compared to before. So it has implications on certainly in terms of how this data needs to be brought in, how it needs to be cleaned. It's not traditional ETLs and all that you can apply on top of it. We very different and right requirements with there.

Now you actually have to understand our PDFs and spreadsheets and all that they need to be brought in, how it needs to cleaned up and all. And this data needs to go to different places. They may need to go to VectorDB, they may need to go to an agent, they may need to be combined a little bit differently, they need to be inspected very differently. That whole stack on unstructured or structured data has a whole stack established over decades. The stack on the unstructured data is in the formation.

Rehan Jalil (07:04.046)
And it will land somewhere of course is forming very fast. We will land somewhere where we say okay This is the way it's going to be moving forward as you know AI is is a shifting sign So it accordingly the requirements and frankly the innovation that's coming through on it is also It's pretty neat that's coming through on the underlying unstructured side of things

Prateek Joshi (07:24.813)
Let's go to the aspect of company building. And you've done this a couple of times. So the way you do it now is just different from a first time founder. So going to the early days of security, how did you get your early customers? And also part B is how do you guide a first time founder to get their early first five customers?

Rehan Jalil (07:49.646)
think in the enterprise side, everyone has their own journey. I'll tell you for mine and we can actually pick in general. It really ties to your value. What value you doing? You have to ask the customer, find the customer that is really tied to your value proposition. You ask a wrong person, you're going to get basically the wrong validation or wrong invalidation.

So that targeting is actually very, very important because you're trying to get opinions which is going to form your own opinions. if you don't get that right, you're going to maybe build something wrong or you're going to basically or miss something which you shouldn't be missing. So actually picking that thing is important. What is your right target, you know, target persona? And for us a little bit, because we were coming from, you know, certain background.

your personal relationships can actually really help. So in my own case, you could call up people. That's what would happen. You'd call up people within your closed circles or friends of friends, you can call and get the input from them. And then you can from there, it can convert to customers. But you can only do that for a handful of them and eventually you have to go beyond it. There you again, go to the places where the people you're trying to sell to, they congregate. They congregate in

know, web forums, they go to podcasts like these, they go to some events, and that's where you're going to find your, the right customers. Because if you target those events and all these things right, you can, you know, build some kind of rapport with Exhaust to actually convert them to prospects and customers. That's what I would say, use your relationships, but first of pick the right one, use your relationships, then go to places where these, your target person will come.

whatever that be.

Prateek Joshi (09:42.357)
And you rolled out your product to enterprises, big companies, and in those cases, things like safety, privacy, guardrails, matters a lot. So in your deployments, what guardrails or safeguards have mattered the most to your customers? Like what are they thinking about when it comes to data and AI?

Rehan Jalil (10:05.934)
That's a great question, actually. So what do people care about? In large enterprises, number one, people care about the data does not get lost. Nobody wants to be in the news. They don't want to say, well, I collect 100 million people's worth of sensitive information, and now it's sitting on some website because it got hacked. Exposure of data, number one priority. It cannot happen. So that's publicly exposed.

Even the second thing I'll put is that internally exposed, you can't have a 10,000 people company and you basically have mishmash of each other. This data can be seen somehow because of what variety of reasons, misconfigurations and wrong place it got placed, wrong place it got moved. So as company grows, the complexity of having the right permissions, who can see what kind of information.

It actually can be a severe issue for the company, in addition to, of course, regulatory issues and all. The AI side has brought a third thing. The AI side has brought a whole new requirement to understand to say, look, I don't want my data to be used by some model to get trained and then start generating answers. So usage of the data with the AI, people want to understand fully.

and they want to make sure they safeguard it, right? And there they want to make sure the data is not getting sent to wrong models and vice versa. People are not relying on wrong responses from these AI systems to actually make their business decisions on it. That's the third thing. The fourth thing, people do care about compliance and compliance is not one thing. You know, there are like dozens and dozens of different compliance depending on

which industry you're in, you would have to comply with different kind of regulation. So depending on that, they want to make sure that, it always almost is tied to use of data and how you storing data, how you protecting it, where is it going, where is you receiving from, generally compliance actually tied to is a fundamental construct of it. Of course, there are other things, infrastructure and access side of things, Then people, I'll tell you,

Rehan Jalil (12:27.448)
People right now also care about how much garbage data you have. Garbage really means they're collecting data from decades. And if data is sitting, and if it is like petabytes of data or hundreds of petabytes of data, and they're never going to touch it, they want to get rid of it. Not just because of cost, although it's a very high thing to which you're doing. More data you have sitting which is not to be used, it increases every other risk that I just mentioned. Because that data is still may have sensitive information. It may get stolen.

Hackers don't care it's older, but mean, it still actually could be, from that vantage point, it still could be very valuable information, even if it's not. And for AI projects, it's even more important because you don't want wrong data and garbage data to be fed to AI to be creating garbage outcomes. At the highest level, these are some of the examples of where people really care about solving from controlling the data.

Prateek Joshi (13:24.353)
When we deploy AI in the enterprise, there's a bit of tension between, on one hand, we want AI to reliably automate everything, like don't bother me, but on the other hand, we want to be able to track and audit every single action because sometimes things go wrong and I want to know exactly what went wrong. So how do you, from a product design perspective, how do you balance the need between

hey, automate everything, don't bother me, versus I need to know everything because I just, I need to track it.

Rehan Jalil (13:58.934)
It's a good question. I think you genuinely need both. Genuinely need both. So first of all, for many use cases, we're not there where it could be full self-drive. It's mostly copilot. And that is good. That's also very good. But let's say if you do go to copilot, full self-drive, you need a way to actually intercept and take over. And you need to understand

if something went wrong, to have observability to go back and see what happened. You need both. they're not at the, actually, if you do it right, they don't contradict with each other, if you do it right. Because observability, you can always put whether it's a co-pilot or whether it's a full self-drive. It doesn't matter. And you have to embed that at different points in your AI system, what data is getting picked.

who, what sources it came from, what entitlements it had, where it got loaded, what if you're creating, you know, embeddings or responses, what responses were created, who was asking a question, what rights did they have, what answers did they get.

to actually have this full visibility is almost very, very important, particularly in the agentic AI side of things where it will do things on its own. It will go to try everything depending on what tools it has become available to it. It's going to try everything. You need that full graph visibility on that, right? And that's what we provide.

full visibility in terms of what on the data and then of course, what data is being used. And at any given time, it's the best practice to have ways to provide feedback, human in the loop kind of feedback and ways to maybe fix it over time. But again, the system has to be built in a way that it's able to provide the visibility but also take the input to actually fix for the future outcomes. And they don't contract each other.

Prateek Joshi (15:55.604)
Right. Yeah, think that's well said. To build and ship your product, you've had to do many difficult things, like petabyte scale data, automated discovery, classification, multi-cloud deployments, on-prem. Looking back, what's the hardest technical nut you've had to crack to make this work, especially when it comes to big logos?

Rehan Jalil (16:24.748)
Yeah, think it's a, there are many, but if you ask me one, the sheer scale at which the data exists right now and continues to grow, particularly if you want to tame your unstructured data, there are billions, sorry, hundreds of billions of files per customer. So if you have hundreds of billions of files, and if you...

For a MAR lens, we provide what we call a data command graph, where you have full visibility on a data object, which could be a file, could be a table, and everything that associated with it. We have to provide that visibility. And then based on that, we provide the way for people to write policies on it, what we call toxic combination policies, to find basically threats, find compliance issues, find fundamental access permission issues. So first to...

One is to actually build all these hundreds of different ingestion points connectors to all these systems, but that's one thing. But to bring that and have a comprehensive view that we call our data command graph and to scale it to per customer to hundreds of billions of these files and still stay performant is a tough nut to crack. It's not easier said than done.

And even when we started, we really did not, frankly, what the kind of challenges we were going to face with it. But eventually, I think you get past it. But at this point, it provides very high value to our customers and frankly to us, a very high level of differentiation.

Prateek Joshi (18:02.265)
Let's talk about the go-to-market motion of infra products. You've done this a couple of times. If you had to share your learnings about just enterprise sales, specifically selling a product like yours, what do you wish every first-time founder knows? Or maybe you would learn earlier in the cycle, this is how infra enterprise sales works.

Rehan Jalil (18:32.205)
Yeah, think infra and it's in for sales, particularly to our personal, that's like a chief information security officer and their teams. And of course, now more and more AI security, AI teams also getting mixed in it. It is very much like enterprise sales, where your value it. The very first thing is your value has to have a good fit with the pain point.

Like somebody must have needed it because there are so many other things to do. Regardless of how good your sales engine or your GTM methods are, if the product doesn't fit in, there is no magic. There's nobody can actually sustainably can sell this thing. So that's built a very awesome product, which actually is, that's fundamental, I'd say. Once you get past it, you have to make sure that you are engaging the customers yourself. Don't think somehow there's gonna be a PLG on it, very difficult.

Unless of course you open source route that you go through it in our case. It's not the case It's not that some partner will magically start selling day one eventually they go right It's not that some bigger size will day one you have to invest in them They will not be one actually start selling in in that so you actually you have to invest in direct relationship with customers

and then get to a point where you can make it repeatable and make it easy for others to actually play. And of course, you have to create a movement around it. The second thing is don't think just GTM. Think creating value, delivering. Think that you have to make this thing to work for your customers. For enterprise customers, one thing is that every customer has a nuanced need. For the same thing, there's a nuanced need.

It's not like you build once in the beginning and you just start selling hundreds of them. You will be nuanced. It's just the nature of it. If one bank needs a supplier, we sell to large, most very large organizations. Same thing, every customer deploys a little differently. In the beginning, you'll be learning and trying to basically get to a point that you can then have everyone just pick one of the options that you have. In the beginning, be open to it.

Rehan Jalil (20:42.894)
And it requires very detailed attention to the customer needs and seeing what others will need and all, picking the things. That's the GTM here. Eventually you will want to make sure your partners, eventually make sure your professional services partners. But internally, it's not just selling, it's delivery. Customer happiness.

Prateek Joshi (21:02.861)
It's amazing. And you made a very good point about nuance. And I wanna dig into that because in enterprise, more often than not, I think what you mentioned, people think that, I build a thing and I'm gonna sell the exact same thing a thousand times and it'll all be nice in the end. But in practice, every customer has slightly different needs. the services component of enterprise software, how do you deal with that?

How should a new founder just think about, this is the product and this is my services offering, you have to do it anyway. So how do you think about that?

Rehan Jalil (21:42.04)
I think, so this is varying degrees of services, right? Some people actually have to point that I have this thing and I have a forward-to-report engineer. We're not that company. But that's also great companies in that category. For our company, it is important to say, we'll do slight tweaking and then help you deploy, but we're not like rewriting code for you, right? We're not kind of really re-instrumenting everything. So depending on where in the spectrum you are as a company,

you have to think about what kind of people are needed. If you are completely doing forward deployed model, you need software engineers that are going and writing code, the PMs, or sitting in the field. a little bit on our side of things, will be where you need very smart people.

who can understand customers' need, and they can actually configure the platform and then communicate internally wherever the nuanced tweaks are needed, options are needed. Somebody uses different kind of credentials to be, somebody uses different kind of walls. So they can bring that back and still collaborate very constructively internally. And then there'll be, of course, very much on PLG side of things that you put one thing and then just that's the same thing that gets used.

where you probably don't have this kind of service. But I would say you have to think through where you stand and pick the right talent. Your talent needs are also very different across the board.

Prateek Joshi (23:07.307)
Now, let's talk about team building, which is a huge, huge part of startup or founder life. many times you get it wrong. So when it comes to team culture, especially in your case, what maybe, what cultural principle of yours has scaled the best from year one to now versus maybe something that worked really well in the early days, but now you had to change it because it just

doesn't scale. So what are these two examples in your case?

Rehan Jalil (23:41.806)
I think basics, I would say foundational basics.

is particularly in a technology company, team is everything. Like there is just nothing else, I would say. And team chemistry is everything. Also, not just the caliber, but actually how the teams work together. That's literally everything. And within that, think the most important thing is that the competence in the areas that you're working on, the bar has to be really high.

self governed by the team that bar is really high because you're at the tip of the spear. If you're really a good company, if you want to be a good company, you will be at the tip of the spear in your domain. And if your team is actually not at the tip of the spear, there's just no other way because you're nothing but more than the composition of your team. Right? So, and then it boils on to within the team, the formation, like who's doing what and they have to, again, have to have a high caliber but then collaboration.

within the team, different companies actually subscribe to different cultures and you know some are very it's okay to have a lot of conflict and some okay to have a lot of discussions and collaboration without having conflict right I think we probably more on this side of things and that I think works really well because you're constantly learning you know because market is evolved first of all your understanding of the need is evolving

Market is itself evolving. The AI game in the market evolved. You're dealing with different customers, which means now you're picking small snippets of different things across the board. So you may have built something in the beginning. It is constantly has to evolve. And I think the teams, the good teams understand it. Good teams actually understand that that's the journey they're on. And they're able to actually collaborate very effectively.

Rehan Jalil (25:43.394)
to actually come up with whatever adjustments are needed and whatever advancements are needed on the product. Other important thing is we all know in a technology company, it's not like a factory you deploy and it starts producing the widgets. You basically continuously have to evolve and become more and more actually advanced capabilities. So while the mainstream roadmap is going on,

which with your own vision that you're probably putting through that not every customer is trying to tell you, you also actually looking at more tactical needs. So you have to, how do you balance? It's only possible if your team is of course very competent, but also very collaborative through and through through with this new signal that is coming.

Prateek Joshi (26:29.293)
And if you look at Rehan the founder back in 2015 to today in the last 10 years, what's been the biggest change that you've noticed in yourself as a founder?

Rehan Jalil (26:46.054)
I think, look, the basics, or at least for me, the basics have not changed. Basics remain the same. You're trying to pick for, you know, solve for the biggest problems, which are real pain points. And the passion, if you don't have passion for that, I mean, that hasn't changed. So also, the way you actually...

deal with customers, the David Deeb teams, particularly in your own teams, how you structure the teams and all that has not changed. I think the basics have not changed at all, I would say. Of course, you learn through the lens of your number of mistakes that you make over time. Number of mistakes in all areas, whether it is people, whether it is picking the technology, whether it's picking the areas, whether it is execution in GTM. In all areas, you can say you evolve over time.

But you never there you can say I've arrived. You can just never say because you're always learning along the way. So I would say in all dimensions, you'll see you evolving and learning and you feel more and more a bit because you can look back and say, I made 50 other mistakes in this area. I'm not going to repeat those. I think that's the bit of an advantage if you're doing it multiple times. But it doesn't mean you you're going to encounter issues and

and you're not going to see things that you're going to still make new mistakes along the way. That's the evolution I would say in fundamentally all areas as compared to 2015 versus 2025. All areas a little bit more mature I say.

Prateek Joshi (28:27.691)
Now, going even higher and looking at a 10-year timeline, a macro, if generative AI adoption, if it's like a baseball game, nine innings inside the enterprise, where are we today? Like what innings? Is it very early? Is it kind of enter the second? Where are we today in terms of adoption?

Rehan Jalil (28:34.188)
Yeah.

Rehan Jalil (28:53.452)
I do think clearly it's very early, right? It's only been a couple of years, very early. mean, yeah, things have been going on, but the fundamental shift on amount of compute and the GPUs that were made available, and of course the kind of techniques that got used, was still very early in the mix. If you think about it, how many things are...

are fully automated. How many things are just on their own doing things? You know, as much as we say coding is going to have, yes, it's assisting, but it's not really, you know, everything is getting coded automatically. Your cars are still driving, a little bit of cars are driving by itself, but not all cars are driving by itself. How have you applied this on human health? Understanding human body, you know, you can look at all these areas which has so much potential.

for AI to be revolutionizing. If you look at just through the journey of that, through the lens of that journey, we're very early, right? Very, very early. We can say, look, AI teacher will be awesome. Where is it? Right? We're very early on it. It's doing magic. We're saying, is going to revolutionize medicine and personalized medications? Well, where is it? Not there. Promise? No question about it, right? So keeping that in mind,

It is voil- voil-ing everyone for the right reason. It is magical. But it is just at the inception.

Prateek Joshi (30:27.731)
And 10 years out, there are many different big directions you can take. Security can take like security cloud, data governance layer, AI safety company, and massive categories. So when you look at your company 10 years out, what's the big overarching theme or direction that you envision owning?

Rehan Jalil (30:52.556)
I think, look, the reality is that AI is totally magical, but it can do a lot of damage. So on one side, we want to absolutely leverage the AI for everything, but we need fundamental guardrails because it is for the consumption of the humans and the benefit of the humans.

And also there are certain other humans which is like cyber criminals and all they're going to use it against against basically the same people so The layer that is needed to make sure that you have guardrails you have security you have monitoring and observability visibility That is not going away at all. That need is frankly is going to be more and more Why because we're going to be more independent on automated agents and whatever is going to come next from it

We as a company have an amazing opportunity to actually serve that purpose to the industry, be that reliable partner to say, well, I'm going to turn on AI. Can you be sure that my data is going to be secure? I will have monitoring going on. It's not going to spill to wrong people. It's not going to generate and start sending sensitive information here and there. We want to be that reliable partner.

for world's largest organizations. as you know, data inside the enterprise is very different than the internet data, right? It's very, very different. It's just not the same at all. People don't realize it actually. And tracking that is something that inspires us.

Prateek Joshi (32:30.317)
I have one final question before we go to the rapid fire round. And it's a two part question. What AI advancements are the most exciting to you today? And part B is what moonshot technology is likeliest to make its way into securities roadmap?

Rehan Jalil (32:53.486)
I think for me, think the most important interesting things are where it helps human beings directly, maybe the most fundamental level. So things that they're getting applied at for the humans directly, like you have alpha four three, which actually has full 3D structures and understanding of proteins, DNAs and RNAs and all, Which would decode something at the...

full molecular level, which is the machinery of the life to basically demystify and understand life as it exists. That's probably the most, I would say, inspiring. And frankly, there is definitely an opportunity to do it. And also, simply you have Open Cell Initiative, which is AI model for the cells and all. And there's a list of things. To me, I think that's like you crack that.

is like very, who's going to be, you know, directly helping with medicine and all that to achieve for human care. And of course, there many other inspiring things, the mathematical models, the physics models to understand the universe, the weather system and all. think those are super inspiring along the way, right? I think for us, I think this is again a moonshot. This is what you talked about 10 years out. we were there, look, Asia is going to be real. Asia is going to be there eventually. On that path,

what the safety layer looks like, could we be participating somehow in it? Again, this is you talk about moonshots. What would that look like at that point in time? I can't say at this point in time that anybody's there, including us, but eventually if people get there and that's the path we're on and we're relevant, well, that would you want to be the layer that actually can be sitting right there.

Prateek Joshi (34:43.981)
Amazing. Actually that is a very good point. I think the statement of Asia is going to be there. It's too big and yes, it's very interesting. With that, we are at the rapid fire round. I'll ask a series of questions and would love to hear your answers in 15 seconds or less. You ready? Question number one. What's your favorite book?

Rehan Jalil (35:11.714)
I think my fundamental books like Emotional Intelligence are my favorite because they talk about core principles not tactics and techniques for business. It's by Daniel Bohlman.

Prateek Joshi (35:21.537)
Which historical figure do you admire the most and why?

Rehan Jalil (35:27.566)
I think many. I mean, you can talk about many prophets, will say, depending on what your religion is. I think they've inspired the whole humanity. So for me, is, it'll be, you know, accordingly, that I'll respect. But then your leaders, you know, like, you know, Mahatma Gandhi and others, you know, really very inspirational leaders.

Prateek Joshi (35:54.027)
What has been an important but overlooked AI trend in the last 12 months?

Rehan Jalil (36:00.366)
I think, I will say, biggest is not always the best. That's the trend right now. It wasn't like you throw so many GPUs and then you figure like, no, no, no, you can actually quantize, can distill, can prune and you can still fit into a small, know, GPUs like H20s and they still do magic and wow the world. I mean, that's really what human innovation has, has no boundaries, I would say.

Prateek Joshi (36:07.789)
Yeah.

Prateek Joshi (36:27.819)
What's the one thing about enterprise data that most people don't get?

Rehan Jalil (36:33.782)
Enterprise data is not internet data. It is sitting in silos. Every file has its permissions. You can't just use it. It's so different. It is sitting in thousands and thousands of silos, and it has control requirements on it. It's not internet data. You can crawl and come back to it. So most enterprise data, if you don't know it, it's not usable. You need technology to actually make use of it. You will not understand that.

Prateek Joshi (37:05.093)
What separates great AI products from the merely good ones?

Rehan Jalil (37:11.502)
think fundamentally if a product draws you back again and again and is so useful to you that you can't live without it, I mean those products are the ones, we know which one those are, if you go back to them again and again, they do magic again and again. Some are flashing the pan, you look good and then they just go away. But some you just basically have to create value again and again, those are magical.

Prateek Joshi (37:38.113)
What have you changed your mind on recently?

Rehan Jalil (37:43.747)
I think initially it didn't look, I mean very early, it like AGI was like, yeah, real or not. I think AGI is gonna be real. The massive amount of compute that is coming through all the effort on, even for small models that can do magic, I think it's really gonna be real. Even if you put regulations on it.

So much competitive environment exists in different countries and within a country that people will get there. And that requires that you have safety really, really somehow thought through. And hopefully, safety is thought through.

Prateek Joshi (38:22.413)
What's your wildest AI prediction for the next 12 months?

Rehan Jalil (38:28.878)
I do think the work that is going on in multi-omic and cellular and pharmacological data sets, I think they will crack the code for the human cells. And there will be the drug and cell interactions that happen. It will be cracked, I think. And there will be very personalized medication that will come through that path.

I'm saying it because I hope it happens because it's very beneficial to humanity, but I do think it's, we're probably getting there.

Prateek Joshi (39:07.713)
Final question, what's your number one advice to founders who are starting out today?

Rehan Jalil (39:16.482)
I think you need all your mental and emotional capacity and your physical capacity when you're building a company. And to do that, even through the thickest of things when you're trying to find the market and build a team and dealing with customers and all, just stay calm. And that's the only way to bring your full...

mental capacity, emotional capacity into it. And frankly, that will be one advice. If you asked one thing, that's what you'll advise me.

Prateek Joshi (39:51.277)
Amazing. Rehan, that's brilliant advice because many times, it's a long journey and it's like you have to maintain your mental, emotional, physical fitness. If you get burnt out quickly, you just die. I think maintaining being calm because startups is mostly ups and downs, mostly downs with the spikes of happiness. So you just try to bear through it and if you can't be calm, then it's hard.

Rehan Jalil (40:13.518)
Thank

Prateek Joshi (40:19.022)
Again, this is a brilliant discussion. Thank you so much for joining me today and sharing all your insights.

Rehan Jalil (40:24.864)
Absolutely, I really enjoyed it. Thank you so much for inviting me here.