EP. 52 Financial Clarity in an AI-First World with Kunal Agarwal

00:12 - Kunal's Career Journey and Unravel

Matt Pacheco
Hello everyone and welcome to Cloud Currents, a podcast that discusses the cloud computing, cybersecurity, emerging tech trends and how they affect businesses of all sizes across all industries. I'm your host, Matt Pacheco and I lead the content marketing team here at TierPoint, where I help businesses understand cloud and security trends to help them better make decisions about their IT strategy. Joining us today is Kunal Agarwal, co founder and CEO of Unravel Data. Kunal started his company at Duke University alongside a computer science professor with the mission of making large scale data environments simple, automated and efficient. Today, Unravel, his company helps some of the world's largest enterprises cut cloud waste, ensure pipeline reliability, and prove value of their AI investments. So today we'll be exploring topics like the hidden costs of AI infrastructure, why cost optimization is not the same as cost cutting. Dig a little deeper into that for our listeners, and how organizations can build and find that financial clarity they need in order to keep innovating. So thank you, Kunal, for joining us today. We're excited to chat. Likewise, Matt.

Kunal Agarwal
Thank you so much for having me.

Matt Pacheco
So I'd like to start the podcast with a little bit about yourself. So let's understand where you started. Can you talk about where you started in career and how you ended up where you are now at Unravel Data?

Kunal Agarwal
I was absolutely. So I was very fortunate to start in the what's called big data journey, which has now become the AI journey almost at the start of, you know, the wave. We first started off as a optimization and governance product for the MapReduce in Hadoop World. If our audience actually remembers that, and were consistently helping them with a similar thing, which is all these technologies are very powerful. They help you do processing of variety data for a lot of different types of outcomes. But they were rough around the edges. And to scale them inside an enterprise, you really needed a complete product, something that makes it simple enough for anybody who wants to employ this to be easy.

And that's really what we're seeing now in the market as well, where these analytics and AI platforms are going outside of the IT department and in fact every department, marketing, finance, hr, are starting to use these platforms and we're super excited to help them on their journey.

Matt Pacheco
Pretty cool. So you started Unravel at Duke. Very, very cool. I'm, I'm in North Carolina, not too far from there. You started at Duke University with a computer science professor. How did that kind of academic meets entrepreneurship dynamic kind of shape the way you've built the company.

Kunal Agarwal
It's a very interesting story. So it was really, you know, a you know, two people who were working on similar problems but had never met each other, coming together to figure out that hey, we should actually create a company around this and take it to market. So my previous endeavors before Unravel, I was actually doing consulting for a lot of large scale enterprises and one of them was this leading pharmaceutical manufacturer out in New York. And they had just started using open source Hadoop that Google had created back in the day. And I was exposed to the kind of challenges that they were facing in running these large scale systems at scale, you know, just efficiently because it was really for hackers and tinkerers back then. And that was the light bulb moment that hey, we should simplify certain areas of this.

And of course we chose the performance and efficiency and reliability pieces of it. And then during my MBA at Duke, I happened to meet with now my co founder, CTO Shifnath, who was actually looking at the same problem, but from a research perspective in how we can actually auto tune these thousands of configuration settings that these environments run on so that it becomes simple for these users to actually use it at scale. So yeah, it was one of those by luck moments that we actually got together. But then when we did, very quickly we saw that we see the world in a similar way, that these technologies should be in the hands of everybody.

But to get it in the hands of everybody, we needed to make sure that all of these things that hold a regular user back can be simplified in such a way that it becomes plain English for anybody to go and act on these kind of things. And that's really where we got started. So we both quit our jobs. He, he was just a tenure, he had just become a 10 year professor at Duke University at that time as well. And we moved from the east coast to California and haven't looked back since. And honestly, it's been an incredible fun filled journey, I would say, throughout all the years in Unravel.

Matt Pacheco
Very cool. Can you explain Unravel and talk a little bit about what you guys do and what services you provide?

Kunal Agarwal
Yeah, absolutely. Ultimately what Unravel does is ensures that all of your data and AI endeavors work to double click on. What that means is we provide automated AI solutions that help you ensure that your data pipelines, your AI projects are running on time every time, and that they scale in an efficient manner, that you don't have a lot of cost wastages. And these things can give you the Best ROI on top of your investment to click even further than that. Why is there a need for doing something like this? Running these large scale data pipelines, which now serve AI models, goes through multi stages where you've got the producers and consumer pipelines. And now that it's becoming a very business critical asset, these pipelines need to not only work, but they need to work on time every time.

Because if they go down, there's actually an impact on revenue for these enterprises now. But as it turns out, it's not very easy to maintain that reliability day over day, hour over hour inside that particular day. So we saw a lot of users firefighting and troubleshooting these pipelines more than they were actually creating net new impact for their particular employers. And that's what we really help out with, which is, hey, instead of you having to triage and figure out where that bottleneck is and for you to then try to root cause it and then try to do trial and error to go in fix that particular problem, can we use these amazing AI advanced analytics techniques ourselves and get you to those answers faster? Meaning let us find that needle in the haystack.

Let us tell you, if something is broken today or something is slow today, why is it slow? Which of those 60 stages that you have are the culprits today? Where is the bottleneck? And then not just leave you there, but even try to resolve that problem for you and tell you, look, today's problem could be hundred of these problems, but we've narrowed that down for you and told you don't bother about all those hundred. This is the two that you should focus on. And then what can you do to go and resolve that in plain English? And then in certain cases we now automate that full loop as well, where it's completely hands off unravel. Is auto healing any of these different inefficiencies that you may have and the way it really manifests itself as value and impact for the customers are twofold.

One, now you can depend on these things. Your data pipelines won't fail. Your machine learning algorithms will run on time. Every time that recommendation engine that you produce or that fraud prevention app that you have running will run and generate value for your company. But the other big side of the equation is hey, all these extremely scarce, extremely expensive human resources that you have on your data and AI teams, you're not going to have them spending 40, 50% of their time firefighting issues anymore. They're going to be actually productive developing and thinking about the next big AI app or the next big machine learning technology for your company. And last but certainly not the least is these investments. They need to start showing roi. They need to start showing actual return on investment of time and money.

Kunal Agarwal
And while the value is going to be produced by companies in different ways, we also help them make sure that the cost part of that equation is optimized, that you're not wasting or you're not introducing inefficiencies in running any of these different projects. And even that we try to do it in the most automated fashion possible for the same reasons that. Leave that to us. That's our place in the world. You guys go and change the world.

10:21 - What Companies Miss About AI Spend

Matt Pacheco
With all your AI ads, that's pretty cool. You're enabling the AI use and the innovation of these companies. So you work with a lot of companies, probably different, a bunch of different sizes. So you've had a lot of experience with them. I'm curious, what are some of when companies are looking at their AI spend, what are they typically seeing and what are they typically missing when they're looking at them trying to create a plan?

Kunal Agarwal
Wow, are we talking about token maxing here? Everybody's talking maxing. You know, as we're having this conversation right now, Matt, I'm sure everybody is waiting for the clock to reset its consumption level so they can get back to, you know, AI more. But look, it's getting real. It's getting real to a point where, you know, I'm starting to get worried about people having the wrong metrics of success. Token maxing is definitely not the right success metric. I think it was great to get people started and getting enabled on AI, which was both fear and greed driven, which is, hey, if we're not using AI, my competitor is using AI and you know, the market is already shaping up to new products. So what are we doing about it?

But I think we're shifting from there to now starting to have more of a value discussion around, hey, what is this AI project actually going to yield for us? But since you asked a question on the cost side, let's cover that real quick. There's definitely the model costs and the model costs are real, but the model costs are becoming visible and everybody's starting to watch those. What nobody's talking about enough is the silent killer. That's everything around that model, meaning the data pipelines that are feeding it, the pre processing jobs that are actually doing the transformations to feed those data pipelines, the feature engineering or even the retraining loops that people are not scheduling properly. That's where the money is quietly disappearing. And unfortunately nobody has an alert on that right now and nobody's watching those pieces right now.

So while AI costs are the headline, the pipeline costs are the actual story.

Matt Pacheco
So you're getting into some of those, I would say maybe hidden costs that aren't as evident. Can you walk us through some more of that? A little bit more of what are, what's driving those hidden costs?

Kunal Agarwal
Absolutely. So if you think about it's a couple of things, right? Number one, it's reprocessing data that you've already processed. You see like when your data pipeline runs, something fails downstream, then this whole thing reruns again from scratch, including all those expensive parts which may have already succeeded. So nobody is fixing that checkpoint and you're just like paying twice for it. The other ones that we see locks in common is this feature pipeline just running steel logic. Someone built a feature engineering job eight months ago, but the model that it was built for, that got deprecated. But that job that's feeding this, it's still running. And every night. Right. The other one that we see is people oversizing the clusters for small jobs.

Some teams are provisioning say a cluster for a 10 terabyte training job, but they're now using that same thing just for a 50 gig inference job. The other 95% is idle, but that meter is still running. Or even experiments that never got cleaned up. Right. Like three data scientists, they had ran 40 experiments last sprint, nobody deleted the failed ones. And the storage and compute of those experiments, well that's still on the bill. Right? So if you think about the AI bill, think about it as an iceberg where the model is just the tip and you have this entire list of inefficiencies that's underneath it.

Matt Pacheco
So what advice do you typically give? Because that could be a daunting task to go back. Especially in the instance of talking about all those projects or those pilots or experiments they've done in the past. What kind of advice do you give to kind of approach all that when you're talking to a customer about getting AI ready and being responsible and efficient?

Kunal Agarwal
Exactly. So it's just about getting the visibility and transparency at the base level, just knowing what's going on. And the tools in the market, they either see one site, one part of the problem, or one stack of where the problem is manifesting itself. For example, you have cloud built tools, they can give you the cost of the infrastructure or the cost of the models, but they're not built for this, they cannot help you understand all of these different issues that we spoke about because they don't have visibility to say, data pipelines. And you know, how our feature engineering job is actually running for an example. Right. So a couple of things are, you know, get that visibility, get the right tools in your stack so you can see those different environments end to end.

And then you start, you know, your process of, number one, audit your retraining schedules, find every job that runs on a timer and ask if it still needs to. Half of them usually don't. The other one you should do is, you know, take a look at your cluster utilization, not your cluster count, which means that, you know, you may have a team with three clusters at 30% utilization. They may have a bigger problem than a team that has 10 clusters but had 85% utilization. So like, what's the actual utilization of all these know, environments and ecosystems that we're setting up? I mean, we've heard about the shortage of Nvidia chips, but you really need the, you know, the large scale H1 hundreds or, you know, an all L60 or something like that would work better for you. Right? So really understand the workload patterns.

Like do you have a compute incentive job, do you have a memory intensive job and do you have the right type of skew on your machine's aside, you know, to that job at hand? Right. And then the third easy one is figure out if you have zombie pipelines. Do you have any job that's been running for the last six months that does not have, say, a code change, which shows you that nobody's actually looking at it? And that's probably worth a conversation with your team to see if that needs to stay or go. And what we've seen is probably several of them. A good percentage of them can usually go.

17:31 - Cost Cutting vs. Cost Optimization

Matt Pacheco
That's great advice and a good place for people to start because there's always, there's so much involved. Like we often hear we have to adopt AI. You said it earlier, competitors are starting to do this thing. So being able to do it the right way is key. Addressing your cost, your data, everything else you piloted or tested in the past, you've talked about kind of a difference, a distinction between cost cutting and cost optimization. You made some really cool points of that. I've heard you say in the past. How, how would you explain that difference to, let's say a CFO who just wants the numbers to go down?

Kunal Agarwal
How do you, how do you do that? So look, doing anything with AI will cost Some money. We've got to be honest about that. And the difference between cost cutting and cost optimization is really you cannot cut your way, you cannot cut your cost way through innovation. So innovation is going to cost dollars. And the difference between the two is are you using those dollars efficiently or are they going to waste, or are they just being idled away because of all of these types of reasons that we're doing right now? So the way to understand if you are getting the best out of the dollars that you're spending is to ensure that you're able to measure everything end to end from that data source to that AI model and then everything else in between.

So you really understand if the cost going through this entire pipeline is actually useful or not. So once you do assess this, what we usually see is people get 20 to 40% improvement by just putting enough attention on this problem. And what really the side effect of that becomes is now you have more headroom open for real AI use cases to go and deploy those save dollars on. So that's what we mean by cost optimization rather than cost cutting, which is it's not about just slashing your bill, it's about getting more yield for that $1 that you're spending on these platforms. And I think now is a good time for companies to really start doing that because we're in this inflection point between experimenting with AI to running production workloads on AI.

We're also at this inflection point of certain teams in my companies are using AI, with everybody in my company is going to start using AI. So now that you have settled down and understood what your AI posture looks like inside your company, it's good to start doing those measurements and governance on your AI projects right away. Because this is only going to explode over the next several months. And once you open this up to various business units and various other projects, you may see a 20x30x40x increase in its usage and then trying to reign everything back in becomes a nightmare situation. So you may already have good opportunities to put checkpoints and guardrails in certain places. So that becomes a standard and that becomes a policy for your companies to then start operating on top of.

So those are the kind of conversations that I'm having with C level executives across all organizations around. Yep, we have gone from experimentation to, you know, rapid production and it's going very fast in this field especially. Let's, let's put a framework around how we measure value for cost. And I think that's the key is once you have the unit cost identified, then being able to attach that to your actual business value. Like, is this helping me generate new revenue? Is this helping me improve my operations? Is it helping me improve customer churn? Whatever that value of the AI product is. If you attach that, then you'll be able to understand the margins. And you don't want the outcome of the investment to be 1x. You want the outcome versus the investment to be at least 10x, 20x 30x.

That's really the kind of outcomes that AI should be driving for you.

22:49 - Preparing Leadership for Board Conversations

Matt Pacheco
That's a great point. So when it comes to. So you just said you were talking to C Suite a lot about this. Let's talk about the board and the people they answer to a lot of the times. Because at first were hearing we need to do AI, so we need to implement AI. We need to work it into everything we do. Competitors are doing it, we need to get on board. Now the conversation is sort of becoming more about we need. We can't spend money forever on this thing. Like, innovation's great, but like, we don't have unlimited funds. How do you prep your leadership teams in order to. Or your customers that are often leaders, how do you prep them for a conversation with their board about AI investment and framing those conversations?

Kunal Agarwal
Very timely question there, Matt. So you don't want to ever go to the board or your CFO saying, hey, we asked you for $20 million at the start of the year, which we thought is going to be good for us for the next 12 months, and we are in month six right now, and we've blown past that budget and we need more money. A better conversation to have is we know that we have spent X dollars in the last six months, but we know that our price performance for this is absolutely optimized. So we've taken part of care of that part of the equation, that we are spending money in an efficient manner and there's no wastages and inefficiencies. So we did spend that 20 million properly.

And then the second part of that equation is, what kind of yield did that have for us? And I think it's important to measure both sides of those equations, especially at this time frame right now. Because if you're not, then to a person that doesn't understand how these technologies work, they will not understand where that money is actually being spent. And then if it's being used or if that money is being abused. And you want to have that clarity in saying, were able to run this in an optimized manner because we did XYZ things that gives confidence to the board that the money is not being wasted, that the money is not being just thrown away. The other part is now you have to really start measuring your outcomes and it really varies by function.

If it's a customer facing function, if it's a operational function, or maybe a marketing function as an example. So I think people are now starting to take certain projects out of their AI workflows and say, hey, this is actually driving more signups for us. This is actually helping us sell more products, you know, to those variety of audiences. And being able to measure that in the form of actual dollars is super important right now. Because I've already started to see some enterprises where, because they're unable to prove both sides of the equation, that we're using money in an efficient manner and we're getting outcomes from this. That's actually started to stifle innovation where like you said, CFOs are coming back and saying, we don't have unlimited dollars for these projects right now. We were supposed to spend x, we already spent 2x.

And the mucus is going to stop. That's the worrying part at this particular time. And I think even with the advancements of the models and the frameworks that companies are using, we're now actually starting to see some real productivity gains. I don't know if the gains are at the levels that the CEOs of these different model companies talk about, where they say, hey, 90% of my code is being written with AI. Or, you know, this complete function has now been automated through AI. But if you're able to make meaningful progress, like, you know, a good percentage of my customer success calls are being handled by AI. My support is being handled by AI. New capabilities and features that we're coding are now being done with AI.

And we're taking our old code and making sure that, you know, the testing, the scaling, the upkeep of the old code is also happening through AI. So even if you're able to make those kind of smaller but meaningful progress, that's really what the board wants to see at this particular time because, yeah, the bills are getting higher and higher.

Matt Pacheco
Excellent points. Thank you, thank you for that. So we love predictability, the predicting predictable costs. Boards love it, CFOs love it. What about the external factors? So let's talk a little bit about everyone's using an AI service, some kind of service to build upon, build their own products with. But we're seeing now some of these companies might need to turn a profit which may introduce some unpredictability in pricing structure of some of these tools. How do you best. Because you can't predict the future. No one can do that. But how do you best prepare yourself for increased costs from some of the tools that now all of a sudden we rely on for all these innovative new products we've released. We relying on the open AI is the clause are anthropic. How do you, how do you build that strategy and account for something?

Kunal Agarwal
I think they've got us hooked Matt. And that's actually their strategy working, you know. Really? Really? Yeah. That's a big debate. It's like how long can VC money actually subsidize the cost of these kind of things? Right, but look, I think it's yet to be seen but we are seeing a couple of cost reductions. The cost per token has gone down dramatically. I think it's going to keep going down even further. I think the supply of data centers is going to reach that inflection point that maybe it's on par with demand or maybe ahead of demand, which would be great for consumers like us of using this technology. But I think what enterprises are also doing, not just to hedge their bet against the cost, but to own and not be relying on.

A lot of these model companies is starting to embed open source models into their own workflows which I think now is a sound strategy because these models are getting better and better every day. And unlike last year where there was a big delta between what was with OpenAI versus the next model, now you're seeing models exchange the leadership position every two months if you may and so are the open source models. And then I think companies are going to get to a place where you do have to start tuning models even more for some particular use cases that your company is actually finding more valuable because your company has that proprietary data that these open source or these model companies may not have. So I think that's eventually how this is going to shape up.

But just for that part, it's kind of early to start worrying about that just yet. These companies have a very healthy growth rate and whenever you see the kind of growth rate those economies of scale is usually passed down to the customers as well. But I think we're in that right place where what I'm particularly excited about seeing is how quickly this technology was adopted inside the enterprise. I'm not saying it's fully adopted, but it's one of the fastest adoption cycles that have happened inside the enterprise, which is awesome. People are not scared, people are not keeping it at arm's length and say, I need to figure all of these things out before I start doing anything.

With AI, I think they're figuring out the security and the governance and all these things that an enterprise company really has to worry about in parallel with finding the use cases and the advantages of these things. And I think along with that, they will also have to come to a conclusion that you cannot be with one model company. There may be certain models which are best fit for purpose, depending on the use cases that you have. And then it also helps you hedge against the cost side of the equation.

30:26 - Organizational Readiness and AI Talent

Matt Pacheco
Awesome. Let's talk a little bit. So you're a leader of a decent sized organization. Let's talk about talent and organizational readiness for things like AI, the advice you so what you use for yourself and your company, but also the advice you give to some of the clients you have. So we know every company is becoming some kind of tech company. It's, it's just how it is. Like everybody. Uber is a huge, Airbnb is a huge tech company and they're like hotel temporary hotels. It's interesting how everyone shifted to being tech first. So they're doing all this innovation and doing all these cool things with AI and now they're sort of becoming like AI companies these days.

How do you, how do you, what does that mean for organizations as they're like, as they should be structured and advice you'd give to customers to make sure that they have the right staff to do all this stuff? Because that's always an interesting question. Not everybody off the bat is an AI expert. Like there's some, there's some gaps. How do you address that? What advice do you.

Kunal Agarwal
Yeah, a CEO of a company. I think about this a lot and very often what I've seen is it, you know, four. So number one, I think if you're not AI enabled, let alone AI native as a company, then you're going to be left behind. And I strongly believe that because the run rate at which you can improve things and launch new products, experiment with things and, you know, create new journeys for your users. It's just incredible with AI. And what I'm personally thinking about is, you know, how quickly can we move to becoming an AI native company. And there's a difference between AI enabled versus AI native, where AI is not just your assist, but AI becomes your operator inside that company, really. Right. And I think it really comes down to making sure that you have curious people on your team.

And you know, different people on your team come in different flavors and backgrounds and interests in what they do and how they do certain things. But I will be seeing that the folks that have always adopted these technologies faster and pushed their particular functions further are the ones that don't necessarily have technical backgrounds or just know AI better than the other, but they're just curious and they're experimenting and they're not afraid to fail and afraid to get, you know, mud on their face publicly around, hey, this is what I've done. So some practical things that I have seen, Matt, that really help in accelerating this AI enablement inside a company is like you said, it's moving so fast that nobody knows what AI can actually do practically for them.

The best thing is to have a very regular show and tell sessions inside your functions in sari companies around. This is what person A did and showcase that to the company. Because we're all catching up, we're all learning from each other. This is that era in this new technology where it's moving so fast that even if you read up all the news and all of the articles in the blogs, that's old by the time you read it. And people are always trying to understand what does that mean for the particular function that I am doing. So we like to have both of an inside view, where you're sharing the stories of what people are doing across the enterprise, but also try to bring an outside in perspective of what are the other CEOs doing, what are the other teams doing?

And not about how do we catch up, but can we get inspired with what those guys are doing and how we can change our functions as well. So those have really opened the eyes of a lot of people to think about the possibilities where they go, oh my God, didn't even know that this capability existed and didn't even know that my job function could change in such and such a way. But the one line that I always talk about for my particular company is, look, all of the repeated high toil tasks you should not be doing anymore. That is something that AI should be doing. You as an expert in your function, say you're a salesperson, customer, success person, product person, you should be thinking about what are you uniquely set up to do that you can do that I cannot.

Like a salesperson, for example, is about creating and maintaining and expanding relationships with a particular account and the customer that requires that human interaction that requires you to visit them and meet them and Be with them and understand, you know, their objectives and think about how we can help them meet those goals. But if you're thinking about, you know, things like hey, summarize that meeting, create meeting notes, create a follow up email, track my to do list, right? I mean those things your AI companion should be doing for you. So you see like how you just freed up, you know, half an hour times four meetings a day, you know, two hours, three hours of your day in doing those follow ups and doing notes, in doing updates to CRM and whatever else you do to become automated.

Where now you're spending your time doing more higher value tasks which are great at doing as this person in this particular role. The other thing I think about is use AI to leapfrog a certain function or a certain area. So the unraveled product of course uses a lot of machine learning from back in the day and newer AI technologies including LNMS right now. So that's how were able to create this innovative product of hey, let's take observabilities this world observability and make it actionability. So it's no longer just observing things and leaving it to the user to go and figure out what to do. But let's get them to answers because nobody's checking out observability for the fun, right? Everybody wants to have a outcome. I want to improve things, I want to make things cheaper, faster, better, more reliable.

So let's get them to the answer and not be in this observability world anymore. So that's an example of how you can leapfrog certain industries, certain products, certain functions areas. So we always encourage our team to come up with these big hairy audacious goals. Using AI in their functions can finance close the books every month in an hour after the month closes. Why not? Why does it take 7 days, 10 days, 15 days to go and close the books? As an example, can we have prediction on customer growth for each account and nail that predictably? Well, what kind of signals do we need and how do we get the right input for this particular function so that we can make this much more predictable as an example, right?

And then improve the certain recommendation engines or things that we're already doing with machine learning and how do we enhance that? So I think the leadership in every company is what's going to drive this. There of course can always be bottoms up innovation as you think about it. But unless the leader of a function, the leader of the company is moving the chains forward for their team I don't think the innovation is going to happen as fast as even in my company. I had to start leading by example and creating certain agents and AI capabilities not in our product but outside, just core functions where people are like, oh, that's actually cool and we should be employing this for our functional areas and so on.

And I think the shape and the roles and the architecture of companies will change with AI where you may start to have more generalists rather than specialists in certain roles or you may start to have more combined roles. Meaning the product person who's an AI enabled product person doesn't need an engineering partner to go and think about the next new capabilities and features. We don't need to sit down in design meetings and brainstorm meetings as meeting number one to think about a new capability and how it's going to work. Meeting number one should not be run with a working prototype where we all see and experience what that looks like and then iterate from there and that becomes a much different experience than sitting down in a blank canvas and imagining what this thing would look like. And now that's a one person task.

And then that one person can also think about what the reaction of this feature and capability would be in the market and go and test it out very rapidly. Don't need a marketing person to set up an email campaign or an outreach campaign to get these things validated anymore. So you see like how that's become a cross function and a joint function now. So you really need people who are thinking about it broadly and boldly like that to really go from yeah, I'm just using ChatGPT or Claude for asking certain questions like a search engine to okay, this has actually moved our organization forward and now we're becoming AI enabled or AI native.

Matt Pacheco
No, that's awesome. And that's a nice glimpse into the future, how a lot of companies probably will operate. So it's really cool to hear some of the things you guys are doing and how you're addressing your culture and addressing your organizational structure and addressing the projects you focus on with AI and some of these newer tools. That's, that's really cool. Thanks for sharing that. I got a few questions left for you. I like this section at the end. It's kind of looking forward at the future. Kind of. You got into that a little. But also some open ended questions and kind of some interesting stuff we could chat about quick. So my first question is, you're working with a lot of companies across all sizes what's the one thing? Just name one thing that they're most worried about right now that they're not talking about publicly when it comes to.

Kunal Agarwal
Your world governance and security. I think they are talking about it but not talking about it enough. And it comes from, you know, various angles. One is just, I don't know if you guys have noticed but there's been a lot of more attacks now with AI but everybody's all these fast moving companies, code is getting stolen and everything else in between is happening. But I also think if we start to have agents become, you know, another person on your team, just like I got Matt and Kunal and this agent and they are part of the marketing team, just understanding how that's going to fit in our human based culture is going to be important to understand. Meaning Matt doesn't want a micromanager looking at how they're doing work every day. Right. But that agent's work is going to be fully auditable and fully exposed.

Check this thing, did this thing, process this thing because we need to be able to fix it and tune it and whatnot. But that agent is probably going to have details from that and details from Canal and you know, how it's doing certain things. So what level of transparency are we opening ourselves up to if we do have agents on our team? Because of course they're exposed to audio work and would you expose your work to your boss? As much as we think about it. So there's a rapid cultural change that people will need to start thinking about and I think that's going to be one of the resistance slash opportunities for companies to really shine and how they tackle it will be interesting to see company by company and like what are we actually, you know, coming down to conclusions with?
But yeah, I think the hype is real. I wouldn't have said that if I didn't see enterprise companies actually using this for meaningful business critical use cases while everybody is thinking about hey, if you don't get AGI then this is not meaningful. I think that's the wrong construct. What we have right now is already very useful technology that helps you speed up a lot of different functions. So I'm convinced that this is not only here to stay, but this is going to become a core part of every function inside enterprises. So really excited to see, you know, the first one person, billion dollar company that hasn't happened yet. I, I'm as a Silicon Valley co founder, CEO, I'm excited to see that one for sure as a personal thing in the next couple of years maybe.

But how that manifests itself into enterprises is can we do a lot more with the teams that we have at hand rather than replacing things with AI? And for me, still, I don't think AI alone can do the things inside an enterprise that we think about. So I don't think that's going to happen, that AI is going to mass replace functions and teams. I really feel that it'll supercharge the person on your team, so it'll be something that's part of their toolkit, part of their flow, rather than that team member not existing. And AI has taken over that completely. Yeah. And I think that's going to be for the foreseeable future. I don't think that's going to change very rapidly.

45:48 - Looking Ahead at Future Tech

Matt Pacheco
Very interesting. Another question for you because you mentioned a little bit about your excitement for the future. What, what is one thing professionally you're excited for in the next, let's say, five years as it relates to new technology, anything tech related? And then what is one personal thing you're excited for?

Kunal Agarwal
Oh, in the same field, professional thing, I think we're all waiting for the quantum breakthroughs now, really. We're starting to see AI and what it can do and that's taken a life of its own. And there's a lot of exciting things that are coming beyond LLMs or even how they're learning beyond just scraping the web. So excited to see that next evolution of AI, if you may. And I think quantum is going to play a big role in that for sure, of course. But yeah, we haven't really seen those use cases. So if there's going to be a ChatGPT moment for that, I think that's going to be a hundred times bigger than what we saw in 2023, 2022. Personally, I'm a big fan of cars and I'm not excited about seeing everything becoming autonomous. So I'm in that camp. I'm in the.

I'm in the camp of I want to still drive my car. I want to still hear the sound of the engine, I still want to burn $7 a gallon fuel. So from that perspective, I'll take a contrarian view and say I still wish the V8s and the V12 engines are still around in the foreseeable future and it's not just a electric autonomous world when it comes to the road. And it really comes back to, look, if AI is going to give you a lot of time back in your hands, you're going to be doing more things that you like to do as a person, you'll get more times to do things that, you know, fulfill you more than running around every day trying to get that paycheck.

And for a person like me, that means being able to go to a winding road and drive the car a little bit longer than I do on the weekends, as an example. So I'm really excited for that. I'm really excited for, you know, having 20, 25 hour work weeks and getting some more time back to enjoy time with family and hobbies and things like that.

Matt Pacheco
Well said. I love it. And I wanted to thank you for being on the podcast today. We learned a lot, we talked about a lot. Really appreciate you taking the time.