How FinTech Govern AI — Mounika Paruchuri, Best Egg (now part of Barclays)

0:00I presume that any AI-based product launch is sort of like a rocket or something, but essentially it looks more like a kitchen remodeling because there is the inspectors watching behind you and you have to finish the kitchen remodeling in the right way.

0:16Shamil Malachiyev: My guest today is a Senior Director of AI and Emerging Technology Transformations at Bestag, now part of Barclays Bank US, where she works directly with the leadership team and executive board to set enterprise AI strategy. Before that, she launched a credit card platform that originated over 300 ,000 cards in 20 months and pushed underwriting precision on 40 % with Agentec AI. So please welcome to the studio, Monica Parachuri. And the model is only 20 % of the output. And the drift that happens once you deliver and the loops that need to be happening to keep in check. We want our people not to tiptoe around.

0:56We want them to feel free and fully empowered, empowered to go and implement the solutions. We want to make the entire building secure and we put a front desk and a security guard and a badge system at the front, which is going to be uniform across the floors. Every time an AI challenges me, I have to think higher than the AI. So I went somewhere from two weeks to two days for a prototype. Now that is a massive unlock. The ROI unlock in AI is the problem that people are trying to insert AI into places where it doesn't belong. This is not RPA. This is not robotic process automation. There's an intelligence.

1:41So you have to rethink that entire process in terms of an intelligence sitting in there. AI unlocked my productivity. So now I'm spending less time at my computer and I'm enjoying my evenings with my family. And I really hope that's where we need to. We have to learn. We have to be smarter than the system. AI is pushing a human, whether they want or not, to be smarter than us.

2:04Shamil Malachiyev: Hello, everyone, and welcome to this week's episode of The AI Floor. My guest today is a Senior Director of AI and Emerging Technology Transformations at Bestag, now part of Barclays Bank US, where she works directly with the leadership team and executive board to set enterprise AI strategy. Before that, she launched a credit card platform that originated over 300 ,000 cards in 20 months and pushed underwriting precision up 40 % with Agentec AI. She spent 25 years in engineering leadership across fintech, insurance, retail, and telecom. And I'm pretty sure we will have a lot to learn from her today.

2:42Shamil Malachiyev: So please welcome to the studio, Monica Parachuri. Hi, Monica. Hi, Shana. So can you give us a little bit of a background about the current work that you do and a bit about the experience that kind of shaped you across the journey? Sure. As you have introduced myself, I am a senior director of AI transformation at Best Tech under Barclays. It's called Barclays Best Tech now. We're getting used to the term. So I've been with Best Tech for five years. And before that, I have been working with USA. My career spans about 24 years. And it's been primarily fintech, e-commerce, and payments. And currently, I started on the AI journey along with everybody in 2024.

3:34before, and I found that there was a gap in the company that everybody was doing AI, but there was not a defined way to engage with AI, and there were pockets of innovation happening. So I saw a gap that we need to streamline and put that under strategy, and that's how I started on this role. and currently I have just finished onboarding Databricks for the company and we'll talk more about paved pathways as we go so that's a little bit about me and you've been uh busy part of the

4:14Shamil Malachiyev: heavily regulated side uh of the tech industry as well so yes I've been working in the regulated industry for 10 years and AI just enhanced the need for regulation. So I'm very much involved in the AI governance and guardrails too. Does it make your job a lot harder? Because I know that the more regulations you have in the company regarding AI, the more you have to pay attention to compliance, to the legal, to the risks, because you have like all the banking data going through your systems. And does that actually like stop the innovation because people are more kind of care to try? Or how do you go about actually allowing for that innovation to happen, even though it's heavily regulated industry?

5:00Very good question. And 100 % you're right that the importance of the governance and guardrails has just become higher on the strategy with the AI models in place. So people assume that any AI based product launches sort of like a rocket or something but essentially it looks more like a kitchen remodeling for lack of a story because there is the inspectors watching behind you and you have to finish the kitchen remodeling in the right way and the model is only 20 percent of the output and the drift that happens once you deliver and the loops that need to be happening to keep in check for the drift the hardness and all of that which comes under the governance and guardrails is the most important part and i want to split those two terms a little bit and talk governance is like teaching the people on what to do what not to do how you set your pathways like your scene logging, your, you know, your activity tracking and all that.

6:15But guardrails is more of let's protect beyond the human because we want our people not to tiptoe around. We want them to feel free and fully powered, empowered to go and implement the solutions. And hence, the guardrails is the most important part that an AI strategy leader is going to be involved in. Because once they create those paved pathways that I'm talking about, it's very easy for people to feel secure enough to go implement solutions.

6:51Shamil Malachiyev: And can you talk a bit more about the actual guardrails? So for example, if we take, for example, a company with all different departments, and I guess URL is to make sure that, you know, the finance department uses this AI model, let's say the harness with certain tools, but these tools don't interact with HR systems, like that maybe don't have access. How do you go about like setting up all of these guardrails? Very good question. There are some parts. So before that, I step back and I just want to, I simplify language. So let me just simplify a little bit about harness, right? What we call harness in AI is essentially an integrated unit testing, your runway testing, and what that would be of older software terminology.

7:43So we are constantly evaluating the decisions made by the AI as to be fitting certain criteria. And so, as you aptly described, harness would mean different things in different domains. So a harness on the business side example and underwriting model is also about our proprietary knowledge staying in-house and not getting leaked out. Whereas a HR system is more about the PII. I don't want the personal information getting into a model. So it's different aspects of what we want to govern and guardrail against. But there are certain common pathways which we would do. In essential, would we want the PII to go into the model in underwriting?

8:40No. That's a common path, right? So to explain to you, the guardrails we define, there are certain layers of common guardrails and harness testing. And there are certain specificated testing and harnesses going for each. So usually we do this using gateways. so whatever platform you pick up we have to have the guardrails which is the gateways by which we protect them and hence I separated that from governance governance is also a lot about auditing the SOC compliance you know the NIST AMF now that's AI we are still getting used to it But all of this is two different things that there are two different pillars of AI.

9:37I'm going to step one more step back and then explain that when I started on the AI journey, I started with the concept of a four pillar enablement. Education, enablement, governance, and guidance. And why is this important? Right? I want to tell the story of a front desk. This is how I explained it to my team several times. And I think this is a beautiful analogy to make it simple enough. There is a building. You need security. Are we going to implement security per floor of the building? And everybody gets to determine their own security? Or do we want to make the entire building secure and we put a front desk and a security guard and a badge system at the front, which is going to be uniform across the floors.

10:35Now, just compare these two things. If you are enabling it per floor, yeah, there might be advantages, but one of the floors has a badge system. One of the floor is just manual. They're going to just look at the person and allow them in. And another is going to be a QR code. So every system is different. Now, imagine the confusion for somebody who is an auditor entering the building and trying to understand, okay, what is going on for security here? now imagine this one where we have a paved pathway that i talk about which is the central desk and they are allowing people into the building right at the door not after so people don't have to figure out to go to each floor and figure out where they're going what they need to do they can just walk in they're gonna just keep going and they know what they should do they should not do where they are allowed and all of that and that auditor is going to be a very happy person He knows exactly what's going on.

11:38Same for companies. And it may sound very simple in comparing this analogy, but you understand that if there is no central behavior, central education, central enablement, and the governance and guardrails, the confusion that comes after is going to be massive. And then you have to pull brakes, stop things, go back, reconfigure, and then enter. So that's how I would explain the AI strategy overall too.

12:18Shamil Malachiyev: I love simple. Like whenever I go to explain something to like some of my employees, also like the first job is to create a picture that's going to be simple to understand. because I think in reality, when it's the only way to show, like when you really understand the topic, you should be able to put it into simplistic terms for other people to kind of start the basic neural connections, you know, before building up on top of it. And I know in the US right now, there is a huge kind of battle between, do we use open source? Do we host it locally inside us? Or do we use one of those like Anthropics, open AI, like one of those models for the company, or we have all of these Chinese models coming up and they're like cheaper, you know, quite powerful.

13:06Shamil Malachiyev: Do we use them? How do you make the decision? Like maybe like financial, like token economics, but then security wise convenience. Do you think the future is open source or do you think anthropics and open AI have their place within enterprise world in the future as well? I think the models are not the issue. Okay. And what you're asking is absolutely correct, but I don't think the answer is a straightforward this or that. Why? I am a strong believer that one size never fits all, even in software. We cannot build one solution that is going to be uniform across. And I've seen many, many, many projects fail with that in Algy, because here it might be just one step on and off, but here it might need four steps, which is going to be first get in, check the plug, and then maybe you need to see there is a, you know, dimmer there or something like that.

14:13So it's never going to be one size that fits all ever. And why I'm again saying a story here is that the model we use really depends on the use case. I am putting in a solution which is going to answer my help desk with an AI query, right? Somebody wants to not click four pages and just get to an answer. do I really need to use the most secure model that I'm gonna host in my uh you know environment and all of that no I can just use an open source model and then feed a document and let the rag search happen now on the other side the underwriting that's absolutely proprietary information which i don't want leaked even anywhere in the club so am i just gonna put it in an open source model now never uh it's it's it's it's gonna kill the business now so what are we gonna do we're gonna deploy models within our vpc more secure private linked uh everything hosted by us and make sure absolutely everything sits in my network under security and everything is encrypted in and now this is getting a little technical but you get the point that will cloud be the answer will will this wars do um as harm than good well it doesn't matter in that sense because if you build the systems right and if you can plug in a model in and out easily right because to me 80 percent of the industry is not training models unless there is a specific need and there is a specific model you're you're training the model on absolutely the data right from the scratch right that's only 20 percent of use cases 80 percent of the use cases are prompt based fine-tuning a model to work with the data and that doesn't need so much elaborate thinking process on which model to pick right you can as long as you know how to swap you can always do it in the olden days we called it refactor I think it's still called a refactor now with AI you just refactor to a new model it's it's not that we have to worry so much about it

16:49Shamil Malachiyev: And do you feel like a lot of difference between like, when switching models? Like, is that a real felt difference? For example, if like on personal levels, obviously, it's probably a bit different. When you had Opus 4.8, for example, and you tried to do some jobs with that, and then Fable 5 came out. And then you tried Fable 5 for a couple of days. then you found it really hard to go back to opus 4.8 because you you've kind of felt that you know relaxation when you can really trust the model to do a lot of work correctly like do you feel that happening when like you switch models within the company as well within functions or is it all about like talking economics so it's like good enough good enough across the layer let's just push it yeah you are touching the core you're running a business at the end of the day right A human picking a model and interacting with it is going to be very different from an enterprise looking at a model and trying to run it in a production system.

17:51You absolutely are right in what you touched on the point of unit economics. Is there a need for me to pick up a new model? Yeah, there's rapid improvement, but that's what the hardness and the looping is about, right? We have built it and baked it into the system right from the start. We always are evaluating the model to be meeting all the criteria, and there's a human checking it, right? So as long as that's doing good, there can be prototyping happening on the new models and check if that output is worth switching the older model and the unit economics. So you're absolutely right, Shamal, but I think a personal productivity and a personal model usage is going to be very different from a model deployed into a production system and the unit economics of it.

18:48You're not just going to switch models every day. It has to be a planned development that happens and it has to be proven that it is worth the switch.

18:59Shamil Malachiyev: And when thinking long-term, for example, within software engineering right now we're probably like when we go to level fable five right now i'm asking myself the question is like okay this is sufficiently good do we need fable six fable seven probably not it it's already doing a fairly good job so like when we look at all of these models and like how they're developing i think right now is pretty much the time where we already found models that do 90 % of all of the job really well. So we don't really need to kind of push for those other more advanced models. Or maybe I'm mistaken. Because I think we're just waiting for the existing ones that make everything run and work for those prices to kind of slowly go down so we can start getting more benefits.

19:50Shamil Malachiyev: or do you think that we'll constantly be chasing those like oh more powerful model it needs less guardrails let's switch to that and then switch to that every year how do you plan for that in long term i look at it like a human learning a system so um in ai getting a little bit into the technical knowledge is supervised learning unsupervised learning and all of this um terminology right the morals are learning from themselves um and i i think that it is just like a human a human will become stagnant if they're not learning something new or challenging themselves more and i envision a time when ai is going to correct it itself in thinking time so So previously when we did software, it was a static part.

20:43We would generate code. It would run in the system. And you know exactly how much time it would take from, you know, from the start to finish. You could measure in milliseconds of how much this process is going to take to get a response. You cannot do it anymore in AI. And an AI model is always now, it's almost trying to replace a little bit of that human knowledge of how can I keep improving myself? How am I going to learn better? So, yes, we do need improved models that are going to learn on top of themselves. Right. And if they become stagnant, we become stagnant as humans, too, because every time an AI challenges me, I have to think higher than the AI.

21:35And where is all this learning coming from, right? It's from a human. We are supervising the systems and we are saying this answer is good. This answer is not good. Go do this, redo this again, because this is not the right way. And it's taking all that feedback and it's relearning itself, right? The intelligence is learning. So of course, we're going to need better models and it's going to be an evolution that happens. but i also think that that evolution is forcing every human to be higher than that next evolution

22:10Shamil Malachiyev: that's happening in the system so yeah i think this is a continuous process that's going to happen so it's never going to stop huh we're going to have to keep learning and learning to live with more and more powerful models and from like from your experience within when AI just came into play, like everybody was panicking, like AI is going to do everything for everyone. And then we sort of realized, okay, maybe not everything for everyone. It's good at certain points. It's not very good at others. When you're looking at different functions, and I think for a lot of people listening to this, it's going to be important for them to kind of have the sobering truth, like which functions do you believe actually require for us to evolve, to kind of take a step back and think, okay, technology is in.

22:58I need to think about what do I do differently now

23:03Shamil Malachiyev: because technology is changing my function. Which functions in your experience were undergoing the biggest change and need to adapt? Sure. So I heard your question in two to three parts. Let me first answer the psychology part of it. for any advancement in technology they've always been the forefront the middle and the resistors right so when ai came out people were like okay this is just another autocomplete right or this is not going to change my entire world right so there was somewhere between this is a chatbot versus a science fiction that's a psychology because it depends on where you are in that range how do we get to the second part of the question so now people are okay interested so now they have to rapidly learn and understand how to use it and so everybody was fumbling in the start, including me too, right?

24:15I have my failures. And I also learned something that what is engineering pipelines would not work for AI. That was a learning. I have to create pathways so people can do better. That was a learning. And as the steps were going on and I was learning on the job, I realized that there are certain things that should be left to AI in certain things that can never be left to AI. And that was a lot. That was the best part of being not between somewhere between the middle and the foreigners. That's where I would categorize myself as. So I created a nod star for my company strategy. One was like 90 % of code documentation, give it to AI.

25:08Okay. why because it's already recorded it's a concrete data right now this data yes it's concrete data because it can read code it knows exactly what it is it can produce uh 90 to 100 percent of the documentation with accuracy we don't need to go and recheck it it's coming from concrete human data now the other the second directive i gave them is testing we again there is concrete data there is concrete code there is concrete everything should we be spending writing test cases or should we be telling what we are testing right previously um i used to have my developer spend so much of time on code quality and all of that i outsourced it to the ai because Because again, it comes down to what is the concrete data you're going off and how much of the drift is possible.

26:10It's testing of code that's already in. AI is even going to find you the outliers that probably the humans didn't think because they were already in the code and already in the context. They cannot switch the context enough. There you go. 80 % to 70 % of testing is all by AI. and now coding that is again so you you see the theme there that's happening anything that's coming from concrete data is always going to get you the best results if there is data that is need to be judged need to be um you know put through a transformer model or a visual model that is going to translate it, extract the information and get out.

26:59We need a human there to check if that's the right judgment that's happening. We're not going to judge. There is hardness. We're not going to do on everyone. But you get the point that we're going to judge the quality. So to answer your second part of the question is, where would I do AI? is where I know that there's quality of data, concrete data, and a result from that. So we are not trying to replace a human judgment and intelligence, but we are trying to make it easy for human to go about doing the right judgment and intelligence and use their capabilities for better outcomes rather than the mundane and the automation, right?

27:50Shamil Malachiyev: Is that why all of the companies are running around with like data, data, we need to prepare our data, you know, standardize it, put it from paper to scanned copies, OCR, structured data. And I guess that kind of brings us to the Databricks that you've mentioned at the start as one of the latest initiatives that you've run. Can you tell us a little bit about that? but like thinking behind it why and also how was your experience actually integrating Databricks has evolved into being a leader platform for AI and I recognize that because of the automatic enablement of various AI features and models and And it sort of creates the paved pathway that I was talking about.

28:41So in the industry right now, there are cloud providers. And of course, we use Amazon a lot. And we use Bedrock, Bedrock guardrails and pathways. But then I also started to put Databricks. And there is Snowflake Cortex, Snowflake AI intelligence and all. But all of these tools, right? I'm not just going to talk about Databricks, but I'm just going to talk about the tools in general. What do they bring to the table? They sort of make the prototyping faster. And here I want to introduce two metrics called cost to value and cost to market. And I started using this metrics about a year ago. Why? It's because a prototype takes some time to develop.

29:34Okay. and that is the cost value. And cost to market is now the prototype when deployed in production at a production grade scale and the unit economics of token maxing, right? So how does that scale? So I had to create two metrics for myself on judging whether a product that is a prototype. So first prove it. don't build an entire engineering pipeline for a POC. So Databricks enabled that faster from. There is features in the Amazon Bedrock and Q. Now Amazon has come up with Q, which makes it easier. But we wanted to deploy a lot of data science models in-house and Databricks made it easy for us.

30:28So hence we chose Databricks. it took away the the hard work of setting up pipelines and the CICD pipelines and a lot of stuff just to prototype an idea so I went somewhere from two weeks to two days for a prototype now that is a massive unlock right so uh and hence that but also it also created paid pathways of governance and guardrails. Now, I created entire guidance documents on how people can just come in, prototype, just do this step one, two, three, four. And then now you tell me how is this unit economics going to work in production? Let's measure again, is this one, two, three, four? And now let's talk about how is this looping going to happen?

31:23Because AI development is never done. AI is a constant loop. You always put a prompt in, check the hardness, loop it back, check your quality. So this is a constant loop. So how are we going to loop this? And these are the steps that you're going to provide on the quality of it. And now let's talk as a team whether this prototype that you have done makes sense to exist in production. So that went from somewhere being two weeks to get a pipeline in, one more week for development of prototype, so three weeks, and then unit economics decision making, which is another two to three weeks with the harness, the looping and all that.

32:13So we cut down project time from three months to a month. and hence it made every bit of common sense it made so much sense to just deploy Databricks as our platform but Databricks is a leading one but it's not the only one and I want to enhance this to all the leaders is that it's not just the tool but it's the strategy behind so what are you looking at is how easy can your teams do a prototype? What is your cost to value? And then your market to value is how easy can they understand what they need to do in order to set up the harness, your gateways, your guardrails, governance, and now also loop back into a human judgment.

33:02And now you have production pathways, paved pathways to production for AI.

33:10Shamil Malachiyev: and from that i have uh two further questions like one let's go with uh the human quality gate when we have like after loop completes like in maturity how does that usually look like do you have like a predetermined like document that should be generated in response or an i don't know a file or a piece of software and then somebody sits down checks it if it's still as good as it should be continue and how often do you do the quality checks so the this is gonna be um again part of every kind of project i'll give you an example of say a chat a rag bot not a chatbot a rag bot that i was talking about the help desk that is going to be user feedback so there can be you can put in buttons or something on the UI that you build and let users tell you they like or don't like.

34:12Okay, simple. And you run a query, say, every week and see if things are derailing and the value is gone.

34:23Shamil Malachiyev: Everybody's putting it. That is not, yes, yes. But that is not a customer-facing bot. It's an internal customer and you're monitoring it, but that's not the highest priority where I'm going to be like, oh my God, a human has to sit on this and watch for it. Whereas now you have a customer feedback prompt engine and that is looking at the customer's feedback on what we are giving back as responses. And if now I'm hearing the customers come back and complain a lot more, of course there are buttons, but there's also feedback text. And now we are summarizing it. So now that is something I'm going to watch for more closely.

35:05And I want a human in there every day. A human in there every day watching for it, analyzing it, and seeing something derailing in real time. And the minute they see something derail, they're going to engage the team again, the development team. But the people who are engaged in it are going to be the agents or the customer service who are close to the problem, right? So it is not one and done again. But the people who are going to be looking at the feedback are going to be separate from the people who are working on it. And so there is context of judgment that is happening. And so we are not blind to everything because it's just one human doing everything.

35:53end-to-end right um but that's how i would go about it and that's how we have implemented

36:00Shamil Malachiyev: and i think a lot of people will remember like back back in the very very early days of chat gpt um you would usually get two variations of the same answer uh to you and then it would ask you like please pick which one do you like better like do you like this response and i've just checked in claude and it still has at the bottom of every response good response bad response and I guess these are the triggers that they used to kind of see like is the new model performing better whereas like based on what people say okay I think that's going to be easier for people to relate like once they look at that is the essential part of going back to your question of why we would be evolving models is they're going to collect all the data and then going to improve it the same as what we are doing for our enterprise projects we're collecting the data we are we are also monitoring the kind of emotion the customer is putting in the chat and now we are recalibrating what our answers are and what what it has to be happening in that situation right the prompt is going to be enhanced to reflect that and so that's the looping we're talking about another topic related to this you you've mentioned that like previously we had two weeks to prototype and now we do it in two days.

37:21Shamil Malachiyev: And I feel that a lot as well with my teams, everything's accelerating so much. I do feel like time is moving seven times faster, like within one day, you can get like something that half a year ago, well, yeah, half a year ago, you would do within a week. But the mental capacity, like the biology of the body, it doesn't evolve that fast to be able to like contain that much information and to make that many decisions within two days. How do you, like, when you're moving seven times as fast, how do you avoid just getting absolutely drained by the end of the day and getting burnt out by the amount of decisions of things they have to see?

38:04Because when you have two weeks, you think about it, you let your brain do the work, you know, map the neurons.

38:11Shamil Malachiyev: And, you know, two weeks is a great time to kind of get your head around the full concept. within two days it's a full prototype like how was that for you like and what kind of life hacks do you use to to stay sane i don't know absolutely i think the burnout is real and you see it everywhere in the industry everybody is you know either in love of ai when they start and in the hate of ai by the time they get to the end of a project and it's it's a natural cycle that sort of happens um there are days when I just I just I'm like so burned out I'm so tired because that loop failed and I've been at it for many hours and I'm like hell with it I'm just gonna walk away now but in I'll just explain with the story that I think I might have mentioned before previously if I'm creating a document I would just open up a blank document and I would start putting my mental information in and then yeah of course we used to do google search we used to validate it we used but what would take me a week now is a prompt that i put into the ai i tell it what i know and i say go look at this go look at that go look at something and this is and in 20 minutes at the top right in 20 minutes at the top I have to go through all of these three paragraphs of dump that happened at me from AI and now is it is it burning my neural pathways yes 100 percent so at the end of the day am I tired yes so it it essentially comes back to the fact that AI is unlocking productivity.

40:02But at the other end, tail end of it, when that productivity is unlocked, how are we managing that time unlock for human to do better, right? It's not like a human is a machine and they can just keep going at it. So now the fact that we have to step away and, you know, do more of physical activities or things that are going to take away your, you know, your exhaustion by doing something else is going to be way more important than before. And I think that should be the human story we write is AI unlocked my productivity. So now I'm spending less time at my computer and I'm enjoying my evenings with my family.

40:55And I really hope that's where we need to instead of, okay, AI has replaced 20 % of the workforce and I just need this 80 % of people to do the job now. That's probably not the best way to write our history.

41:16Shamil Malachiyev: So as long as companies are staying profitable, we can look at like Europe a year ago I was hearing about oh we should do 4 hour work days and I was like what are people going to do within 3 days and now I can see that if we actually increase productivity within some functions like even 3 times, 4 times as much maybe it would make sense to kind of work 4 days but still do 3 times as much as you would normally do. I think it's kind of a bit worrying when you hear a lot of technical leaders giving their idea of what the future might look like. And they're like, we'll have, I think Elon Musk was saying that, we will have a lot of commodities.

42:08Shamil Malachiyev: Everything will be available at very cheap prices because of competition. So you don't need to make much money to do work. And we'll have robots and everything just producing 24-7 non-stop. What is your view on how this might look like within 10, 15 years? World has always been shifting around us, right? And it will continuously shift. But I think there is also positivity that can come out of it and negativity that can come out of it. What we identify with and what we want to do about it is what matters so if i'm if i'm the leader and i have six people um on working on a project and they're able to do that work in four hours the next four hours i'm going to ask them to do something creative something they want to do something they are going to do because they love it and this could be a bigger unlock for the companies and for the projects right i wanted to talk about 10 percent projects when i worked at southwest airlines i loved the culture and there was something called 10 percent projects so 10 percent of the time people can do whatever they want even in that day and i had a colleague who was building airplanes in his garage and he would use that 10 % time to build.

43:38It must have been a big garage. It was doing some sort of tweaking on plumbing. Okay. So I have never seen any company embrace like that. And that's why you have some craving for Southwest Airlines. But you see how they are doing innovative, creative work, which actually makes them better at their job. it's very hard to measure that sort of productivity. So when I took that to my team and I started giving them, you can do whatever you want for 10%. Of course, I told them not to stick to software, not about garages and everything.

44:21Shamil Malachiyev: But they produced some amazing results. I had somebody who had automated the entire selenium testing and I'm talking five years ago. It's not with PAI. I had somebody who solved how to help the internal customers look up information, which was in multiple systems. So what I'm trying to get to is what Elon Musk and everybody is talking about is there is a good part of the story also is that the automation is going to get better and the mundane things are going to be done better. So now a human is unlocked in terms of time and energy and to do more creative work. We are going to just keep innovating because that's what humans do.

45:12Then, you know, if you have a paycheck and you have time and now all you have to do is do something to keep yourself happy, oh, we're going to unlock so much, right?

45:24Shamil Malachiyev: So you think we'll have a lot more hackathons, basically? 100%. the jobs are going to get very interesting, right? I know it won't apply to every industry that's not the same for maybe call center employees or call center or something. But you get the point that as much as innovation is happening, the story can be two ways. One is negative and the other is positive. so we as leaders what do we want to enhance what do we want to put our time and energy into is going to decide the better part of the story so we should try aiming for utopia version for the good one rather than reading all dystopian um and let's say that uh a lot of people are still working and you know right now it's sort of like uncertain time a lot more people are starting to understand what AI really is and how it works.

46:26Shamil Malachiyev: What if somebody's in their job right now and they get their manager, CEO, saying like, we need to, like everybody's adopting AI and getting all this like three times more productivity. We've heard the podcast with Monica and she said that everybody should prototype within two days and, you know, deliver product within a month. We want that in our company too. Show me some ROI. What would you advise to those people in the position of responsibility when they have teams and they have to deliver from that? You've mentioned the four pillars. Maybe there is some extra to kind of fight the resistance, maybe promote like AI months within company.

47:11Shamil Malachiyev: Or what else can people be trying to do to you? the roi unlock in ai is the problem that people are trying to insert ai into places where it doesn't belong okay to me there is huge unlock and i want to explain this again with another story um you know um from the past when steve jobs was first um creating a hard desk for for the apple um he assigned a few designers to go and check on how to make it better and you know make it simple and they were working off a huge prototype huge document which is like 30 pages and they designed a solution which was five steps and they were so proud of what they did and they walked up to the meeting and he listened to them for 15 minutes and he said so no we are not going to do this and he went to the board and he he drew a box and he said i want something that people can pull this onto this box and they're done you see what what the strategy is going on the he was not trying to take something that's as it is and reduce you have to rethink so why is the roi unlocked so hard in ai is people just want to take ai and insert it into something that's that's that's uh you know already complex yeah it's not going to work that way we have to rethink that entire uh project the entire process the steps in terms of a because if we if there was a human who was looking at a call center feedback and then just you know saying yes no or something like that um maybe you can still insert it but there might be complicated steps where you need to redo an entire process and do the mapping first before you ever build it in AI and hence that unlock is very hard for many many many people that's what is happening in the industry and yeah to any CEO that I'm going to walk into right now the biggest unlock of roi is protecting yourself against cyber security yes because there is a lot of like with fable and everything there is a lot of chance that your systems are going to get attacked and that is an easy roi i can i can sell to any ceo hey we need to product our system so let's go and build these models that are checking in that we have to build better firewalls, we have to build better bot detection systems and all of that.

50:22Now that is an easy ROI. But if we are trying to insert AI in some sort of system and make it an automated system and reduce the burden, this is not RPA. This is not robotic process automation. There's an intelligence. So you have to rethink that entire process in terms of an intelligence sitting in there. And that is the thing that people are just now wrapping their head at.

50:54Shamil Malachiyev: That's my answer. I'm going through, like, a year ago, we've got ISO 2701 certification, and right now we're going through SOC 2. And when talking to a lot of auditors, I'm asking them about all these things like, okay, so we have an open AI model just breaking into a hugging face and just, you know, just like, doesn't make sense. You know, we're trying to protect here against something that I could just look at the system, see all the 20 loopholes and then just inject itself. Is the whole security space going to like do an evolution into something completely different? Because I don't have an answer for that.

51:36Shamil Malachiyev: I'm like, how is that going to look like? Maybe you have some guidance. i don't have an answer to because how how far deep are you going to look into your cyber security because i might be using some vendor who is using another vendor and the vulnerability sits in their code or their system how how deep do you go it's a very hard space right now i think that that space of uh ai is gonna be the one that is gonna the biggest unlock of roi right now um because we We have to protect our system. And that's the nightmare that's keeping all the CEOs up in the night, to be honest. But I think there is also enough regulation that's being kicked in just for that.

52:22Because we have to regulate some parts of the AI aspects also in terms of the damage it could produce if we don't put, again, the guard rates, right? uh so yeah i i don't have an answer to it a lot of enterprises are you know they they are trying to do the best firewalls the bot detections and it's constantly evolving but are we under threat yes but are we under under shutdown now because you're gonna you you're gonna sit on the uh the vulnerability and the legal languages that are provided by your vendors. And it's, yeah, it's, it's, he did and she did kind of situation, but that's what it is right now.

53:20Shamil Malachiyev: Because to me, sometimes it seems like we're almost like 15 years ago, like everybody was hosting mostly on-prem, you know, we didn't have as much like cloud compute. Then everybody just switched to cloud computing. And now I feel like I'm seeing more and more people start talking about hosting things on-prem with a new initiative from, I think, OpenAI or Anthropic and about forward-deploying engineers within companies. And that all to me, you wouldn't need forward-deployed engineers if you're not planning to do something on-prem. You can just work for them remotely. so it seems like everybody is leaning towards you know people who have their own like ai hubs in data center you would have like separate rooms and everything just do you think that is kind of the world we're going towards and people should start already thinking about like where are they going to host all of the service the models and uh getting people to leave you know next to the kill switch at all times um highly regulated industries yes it's only happening right um if you look at any big banks and i saw this trend happen and a year and a half ago and i made a comment so there was this funny uh thing about when um open clock came out and everybody was mini max and i was laughing that okay now we are good we are back to on-prem now and somebody made a comment that that's not so true and i was like okay but but i saw that happen and it's actually happening in how the ai models proprietary models are being hosted for the highly regulated industries and i just mentioned that even our proprietary models are are cut away from access for anybody, even though they are on cloud, they are on our VPC, on our premises and all of that.

55:30Big banks have big pockets. They can deploy in total data centers, right? And so I agree with you that that's the trend that's going to happen is people would want more control. And if they're developing proprietary models, which is from scratch, model training, supervised learning or unsupervised learning, they would not want that to be exposed. So pretty much on-prem is going to happen, but it might not be on-prem as in physical hardware. That's a funny story. the first job I stepped on something and pulled out a server for my startup. We're not going to that. We are definitely doing some sort of between cloud hosted systems, but they would be in your control, in your proprietary, you know, shell and sort of secure from external hacks and attacks.

56:43Shamil Malachiyev: that's what we're good so i think that kind of goes to what you were saying about buildings you either put if one good security on top of the whole building and that i think what what amazon is doing like they're just taking it very good security and putting it out the door so everybody doesn't have to worry about yeah like how do you secure something myself that's how amazon was even with cloud right forget ai even with cloud that that's what amazon was very good at is the sort of security that goes into every bit of information that's encrypted in and out. That's most of the cloud providers.

57:21Shamil Malachiyev: And as a final touch, anything that you want to kind of advise all of the listeners to, like what should they be doing in order to make sure that they secure a good understanding and a good place in the future that is getting built right now? humans are going to evolve um with technology so we were we were i think for our generation to if i start talking about steam engines and electricity it would not make sense so we are talking about going away from retail to online e-commerce or something like this right the internet so that is more relatable so if you were somebody in 2000 looking at this and if you are in 2025 looking at ai the pace is even more um higher in the current year or 25 26 the pace in 2000 was a little uh softer you you had time to react and why i mentioned that is we have to evolve as humans i know there is resistors that they're gonna see at all the negativity that's coming out of ai and say yeah it's just another autocomplete it's just it's reading off the data it's only it can only produce what another human being has already documented so how is that any different but also there are people who are like I'm going to do everything and anything in AI so in between is there is a there's a huge range and for most of people I would want them to be somewhere in that range leading towards the front because this is going to be altering our lives and getting into the bandwagon and understanding what it can do for us and what it cannot do for us is going to be a very wise learning to stay current in the industry and and and it it is not about an age factor and i i sense i have heard this a lot that the younger folk are better because they're coming with the learning of AI and they can understand everything faster and the seniors are now in the resistor category and they know that's so untrue.

1:00:00It really depends on how well a person can learn and how dynamic they are about change. That is what it essentially boils down to because change it has always been the constant thing in our lives and we always our neural pathways are always built against that we don't watch it but the change is the constant thing that's happening so on that spectrum i would advise people to be somewhere towards the leading part so they can be ahead and understand where they are going to be at a disadvantage with an ai product or ai enhancement or anything to do with ai coming into their lives versus where they're going to be at an advantage.

1:00:49So we all have to learn it just like how we learn how to, you know, work with computers. So that's my advice is don't resist, but you don't have to be the forerunner and put everything in AI. You can be in the middle, you can observe, you can learn, stay up to date. And this is going to change our lives. so get mentally acclaim that this is going to change we have to learn we have to be smarter than the system ai is pushing a human whether they want or not to be smarter than it

1:01:31Shamil Malachiyev: yeah so it's coming well guys you've heard it there is no stopping it time to adapt time to change and evolve. Monica, I want to thank you so much for joining us at the studio today. Thank you. My pleasure, Shamo. Thank you very much for having me here.