Sponsored by: Nagarro | The Agentic Enterprise: What has to be True Before AI can Act Autonomously

Summary

Most companies have working AI demos but the transition to autonomous production systems stalls because of missing orchestration, governance, and organizational design — not model quality.
Nagarro's Fluidic Intelligence framework treats AI as a continuous system, and Oversight Systems demonstrates how a traditional SaaS company transformed into an agentic, AI-driven finance risk intelligence solution running on Databricks.
Four prerequisites for the agentic enterprise include AI explainability and auditability, clear ROI measurement before building agentic interventions, proper debugging through reasoning traces, and defined boundaries for where human oversight still belongs.

Chapters

FAQs

What is Nagarro's agentic transformation model?

Nagarro's agentic transformation model follows three stages: Run, Grow, and Transform. Run addresses immediate operational improvements, Grow scales successful AI interventions across the organization, and Transform focuses on fundamentally changing how the business operates using autonomous AI agents — treating AI as a continuous system through what Nagarro calls the Fluidic Intelligence framework.

What did Oversight Systems change about their product to become an agentic company?

Oversight Systems transformed from a traditional SaaS and private cloud-based company to a modern agentic, AI-driven solution running on Databricks. This video covers how their finance risk intelligence platform now uses agentic capabilities to detect and respond to financial anomalies, with a live demo of multi-agent ticket resolution with a human in the loop.

What are the four prerequisites for the agentic enterprise?

Based on this video, the four prerequisites are: a clear agentic transformation model (run, grow, transform), AI explainability and auditability as non-negotiables for regulated environments, measurement of ROI before building an agentic intervention to justify the investment, and defined boundaries for where human oversight still belongs in the autonomous workflow.

How does debugging change when AI agents replace hard-coded logic?

Traditional software debugging uses line numbers and deterministic execution traces, but agentic systems reason probabilistically and do not fail at predictable points. This video explains that debugging agentic systems requires working with reasoning traces — the sequence of thoughts and decisions the agent made — rather than stack traces, which demands new tooling and a different mental model from engineering teams.

Full transcript

[00:08] Thank you all for attending. My name is Anurag Sahai. I work for a company called Nagaro and I'd happily walk you through that in a bit. Along with me is my partner today Bala. He represents uh
[00:21] his organization oversight and together today we wanted to share this stage and talk to you all about the agentic enterprise. Uh Nagaro uh you know works
[00:33] as a vendor with several clients and we help in a lot of their transformation around AI and agentic journeys. Oversight has been one of our customers and we wanted to share the stage with
[00:46] Bala today to really sort of bring these stories to you about how some of these transformation journeys are happening. So um the way we'll do this talk is I'll walk you through uh some of the important ideas that we are thinking
[00:59] about when we're thinking about the agentic enterprise and then Bala would join me uh you know in the second part of the talk to talk about what specific things oversight's doing to really become a a gentic enterprise. So with
[01:13] that let me hand uh a couple of things. Bala maybe you want to say a few words before we begin. >> No no definitely. Um as said um I'm Balatu. Um I lead the product organization at Oversight Systems. We'll
[01:26] talk a lot about what Oversight does and how we transformed from a traditional SAS a private cloud-based company to a modern agentic and AIdriven kind of
[01:38] solutions. So I'll touch on it. I'll let Anerog touch on like some of the fewer things that he'll be talking about how Nagaru does things and I'll take it from there and talk about oversight systems. >> Wonderful. Thank you. So let's begin
[01:52] with a small um explanation. Uh but before I do that actually let me just quickly talk about Nagaro because uh we are introducing our companies here as well. Um so Nagaro is you know u sort of
[02:06] uh uh the way we look at ourselves is that we are an AI and a data engineering company but with an agenda to transform our clients to help them become more AI ready AI first. uh this is our view of
[02:20] how we like to see ourselves and as part of this activity what we try and do is work at several levels with our customers. At the at the topmost level what we do is we kind of think about
[02:32] their industry, their domain, the specific verticals they come from and bring out all the uh interesting possibilities of how AI can really help transform them. We do this because we
[02:44] work with several customers in that domain and a lot of that knowledge begins to become a part of our vocabulary, our understanding. So we bring a lot of that knowhow to all our clients. The second kind of way we work
[02:56] with our customers is we bring a lot of best practices of implementing engineering projects with them because we work across several customers in several geographies with their own little complexities. that really gives
[03:10] us a very interesting position or a seat to understand what works and what doesn't work. So that's the second kind of thing that we do. The third is of course the engineering bit like if you
[03:23] want to really implement some of these solutions properly, how do you really build end-to-end systems especially if you have a good systems engineering approach to solving problems. So these are the three four things that we do
[03:35] with our customers in different ways. We like to say that we don't have a headquarters and partly that sort of sentiment is uh you know sort of pivoted around the idea that you know we are where our customers are. Every customer
[03:49] story is very unique to us very special to us because that's in their context that we are building. So while we have a good understanding of the knowledge and the learn and and the important engineering that we that we are able to
[04:02] bring but the challenge is always is can you adapt this for for for for a specific customer in their specific way. So that's us u broadly
[04:17] one of the important sentiments right that we have been getting because we again talked to so many people and again this is a cliched now things change really quickly I mean PC's it was all about PC's a few few months ago maybe
[04:29] but well of course it has been decaying very very quickly but practically every sentiment nowadays when we are talking to customers is hey can you make it work in our context that's the most important thing that customers are really seeking
[04:42] and this is very important for us because when we are trying to build something we have to make sure that this is not out of context it's not generic it's not you know um you know uh just a
[04:55] very cliched thought process that we are bringing so with that out of the way one of the important things I want to talk to you about is transformation itself right like what is this agentic transformation I find this terrifically
[05:08] useful uh as an example to talk to people about u because again it's a word it's been massively used in many many different ways everybody is transforming broadly in you know and transformation is just
[05:21] like that word transformation uh but but what is this whole transformation about right and one of the things you'd appreciate is that there is a lot of jargon in AI and and jargon's like you know all around us agentic enterprise
[05:35] and this and that right like but do we have very specific specific mental model especially for senior people right people who are leading these initiatives do they really understand where the challenge is going to be in the coming
[05:48] years when it comes to this transformation that we are talking about the AI transformation so in order for me to show you that let me just quickly switch over to a small demo to help you u um sort of see this um broadly
[06:07] just a So I use this demonstration only to sort of like help communicate the the idea
[06:20] that is inherent in transformation and and there a couple of important things that are going on. This is again, you know, talking in talking about agentic transformation in the Bay Area is like I don't know if that even makes
[06:32] sense in any way really, right? Because it's all being driven from here. But it's really important to be able to see some of these very simple things that are happening when we say that this transformation is underway. First of
[06:45] all, we think that the nature of software is changing really, right? And I'll show you an example of what that transformation really means or looks like at some level. On the one hand, this is really important. The second is the nature of engineering is changing.
[06:59] We think again this is really fundamental and again I'll try and show you something about that. But both these changes are leading to something that's changing in terms of products and business models and so those are
[07:11] consequences of these two other things. The software product itself and the nature of engineering. So what does that look like? Right? So if you just think about that, right? Um if I can show you
[07:24] a quick demo uh just to maybe and this is slightly animated and oversimplified but just have a look at the demo and then I will describe that in a bit.
[08:13] >> Great. So I'll replay that and this time with a narrative. So traditionally we are used to thinking of software in a certain way. I'm using this graphical user interface as a small archetype of sort of demonstrating what software is.
[08:25] But typically what happens is when you build a software, a human being operates that software. They go about clicking things. They know what they are doing, what they have to do. They maybe have to give some information somewhere. They
[08:37] maybe have to, you know, uh click a button or two. And that that logic, that workflow is in the person's mind. This is traditionally I again it doesn't have to be a UIX you know or a UI kind of an
[08:49] experience but as long as we sort of see what that is happening right now what what's really happening is now instead of that business logic inherent in that person's mind it is in the agent's mind right and you just express your
[09:02] intention and the agent does the work it it has the logic that is needed to really get that same outcome done and so the nature of software is now more
[09:14] driven towards for an agent to be driven in or actively you know sometimes you call this headless sometimes there are all kinds of ways to frame this idea but essentially the mental model is software
[09:26] the very nature of software is no longer what it used to be about having building all these little interfaces and a person going and executing something that's just gone right and and most of us are
[09:39] now thinking in in those ways so that's um the first thing I wanted to show the second important thing is around the engineering like what happens to
[09:51] engineering as a result right if that's how software is going to be what's about what what what what's going to be the nature of engineering that we do so imagine that you have a piece of software and at some point there is a
[10:03] bug that comes by now think of that engineer who has to go and figure out what that bug is so that person will go look at the code and they'll come come across maybe you know some line 147
[10:17] where they think that there is an error and that error is causing that problem to happen right so they go and they go fix that error based on the best understanding they have of that piece of software so they go to line 147 that's
[10:30] what the diagnosis and again I'm massively oversimplifying but that's the core idea right essentially now what happens in these kinds of new softwares that we're writing right you go you have a problem. You open the code
[10:44] and what you see is that there is no line 147. There is just a reasoning trace. You see a bunch of things that the agent has done. There is no 147 line where you can go
[10:58] and change a piece of thing that will change the behavior of the software. What you have to work with is you have to work with the reasoning. No, you've got to change the behavior of the agent
[11:10] in such a way that it does what needs to be done, right? And this is very significant, right? Like what's going to happen is that the way we kind of think about engineering is is very very different as
[11:23] a result, right? And there are of course very interesting nuances, right? You you fix the reasoning in one case because these are stochastic systems. they don't behave consistently all the time, right?
[11:36] You got to understand what this casticity is like to really get the kind of behavior that you want. So the new engineer or the new engineering must incorporate all of these ideas. So I'm I'm sort of showing you what this
[11:49] transformation really means because how we've been thinking about software industry primarily is very different in the days to come. It's already beginning to become different and you know the engineering around it is very very different. And what this leads to is all
[12:03] kinds of situations uh that we think are going to play out in the in the days to come right like one of the important things is that you don't since you know again this this demonstration is not really that significant but what happens
[12:16] is that you are paying for you know um I can yeah there you go that you're paying for a software seat um you know or a software license uh because you bought a
[12:30] piece of software that a certain person is going to execute or how many seats that you ever got from the piece of software. But when you have agents building things, then you're not really paying for the software. You're paying
[12:43] for how many what all activity that that agent did, whatever outcomes, how many outcomes did that agent deliver. So you're actually paying for outcomes more than what you're paying for software,
[12:55] right? And so that's the consequence of the the first two things the software changing and the uh engineering around it changing and and and so the consequences then you know sort of like
[13:08] um move all the way from there to many many different things. Uh for example the way we kind of think about this is that you know um let me see if I have this here. um
[13:27] like right here like you know you have at the core the very nature of software changing as a result the engineering around it changing and that has consequences on products business the way you run your organization and so forth so with that being said um you know let's going back again to our u
[13:42] slide deck if I can just maybe
[13:54] we try and bring some amount of sense in this whole journey and how to really navigate this journey for them. We work with companies like data bricks, snowflake uh openai, anthropic and many
[14:07] others to try and make it simple for our customers so that they can use these platforms in the most coherent sensible manner. One of the interesting and important things that we do is you know
[14:19] try and figure out what are the most sensible use cases to really go after. One of the ways we do this is to really segment businesses in a very sensible manner. For example, we separate ops or
[14:32] run as one category, grow as another category and transform as another category. If I can just quickly take a minute to describe right all your day-to-day routine manual work is your run work right
[14:46] there's a great demand or a need for bringing efficiency in those categories so we work a lot actively with our customers trying to figure out how to really get a lot of these ops out of the way whether that's L1 L2 L3 kinds of
[14:59] things we are building agentic platforms to really make that much more simpler for our customers when it comes to grow it's all about new products new product experiences. So we work a lot with them trying to think about what would make
[15:11] sense in healthcare what would make sense in manufacturing what where are the lowhanging fruits how should you think about that and architecturally describe it and finally of course transformation right which is saying oh I have so many people we want you know
[15:25] we want to transform this entire organization we want to reimagine the processes around how we do things and so forth so the transformation is a third kind of work so broadly all these three are playing out in our ecosystem and
[15:37] there's a bunch of other things that I thought would be great to have. But in the interest of time, let me just stop here and ask Bala to really take this uh conversation further. Uh Oversight is a great partner of ours and it'd be really
[15:50] good to hear from him about how AI is really u playing a role in in in your business models. >> Awesome. Watch. Um thanks. That was a great setup and a great demo. Um so
[16:04] today you know let's talk about how we transformed from the traditional like uh private cloud enterprise to a modern agent AIdriven kind of a solution
[16:19] right uh the path was not easy but we got there and we are there right now and we are seeing some phenomenal success so before I talk about it let let me tell you why is that needed right um it's
[16:32] because If you think of any modern kind of enterprise, there are like trillions of dollars that are being processed or being paid out as as a normal day-to-day business, right? It either comes through
[16:45] as accounts receivable to your organization or either um either your payments that you're making or the procurement teams making some purchases. There are constant money flow in and out of an organization that happens. But
[16:58] guess what? with money flow being around like you know these kind of enterprises the challenges also have become even more complex the challenges being the the fraud sophistication
[17:11] you will see that you know there's like with the advancement of AI technology the bad news is the fraudsters have access to the same technology that we all have and they're creating sophisticated like kind of you know
[17:23] fraud for example in in like travel and expense area rece receipts can be generated with like one click or two click in in in like 5 seconds right I don't even have to do a travel I can say
[17:36] I travel to San Francisco I attend a data bricks conference sitting somewhere like in east coast and then submit an expense report and still get reimbured things do happen right and same thing
[17:48] right with prompt injection like if uh if an invoice comes in the prompt may mean something else if you're using LLM then you're paying much more than that's needed to the you know the vendor. So
[18:00] that's become sophisticated and which means the challenges also are sophisticated for the finance teams. The back office costs are constantly rising. I don't know about you guys, every client and every you know leader that I
[18:12] talk to the common theme these days is do more with less right everyone wants to do more with less because the costs have gone up and they want to reduce cost and get the maximum out of their teams and the hybrid and remote work
[18:27] kind of environment choses like you know every company chose to do that some companies do a complete remote some companies do a hybrid but it's always you know a challenge for the organization because Now your employees
[18:40] are all over the place and when like they submit an expense or create like a purchase, how do you track all of that? Is it truly your employee creating those in expenses or someone else doing it? And with modern you know technology
[18:52] where like the supply chain is distributed some of the large enterprises that we deal with have 150,000 vendors that they deal with and the 150,000 vendors are all over the world. So now how do I know this invoice
[19:04] came from a true vendor that who had supplied some kind of equipment to me right that's another one and if you look at the data the the history says whenever the consumer debt is like very
[19:16] high which is today people tend to do more kinds of crazy things right so everything is you know like exaggerated and like you know has multiplied in terms of the the impact because you know
[19:29] there is there's no true solution which can solve for it for the office of the CFO and that's exactly what we do at oversight right we like Everest has formed a new category Everest is a you
[19:42] know analyst group they formed a new category called finance risk intelligence we are the market leader in that you know category if you think of what we do in simple terms any dollar that comes into an
[19:56] enterprise any dollar that goes out of an enterprise we are the system who will be monitoring for continuous fraud, risk or compliance related issues and be able
[20:08] to proactively solve for them through our agentic solution. That's what we do and that's the value we create for our clients and as you see some numbers we have been doing this for like more than you know 15 plus years in the you know
[20:21] industry and we have been very successful with it with like four enterprise plus large enterprise clients. How many of you had to travel for this uh outside of Bay Area to attend this data brick summit? Show of
[20:33] hand fans. Okay. Good majority of this and I'll tell you like when you go back home submit your expenses there's a 50% chance that oversight systems is going to look at your transactions and making
[20:45] sure uh everything is correct. Right? So that's that's what we do on a day-to-day basis and solve for your large clients. So we had to take care of this and we had to scale. So about two to three
[20:58] years back we as an executive team along with our board sat together and said like what should we do to scale ourselves and set up ourselves for the future. Clearly there were like four
[21:10] main things that we identified. The first one is we need an infra infrastructure that can scale as our models you know scale. We need an infrastructure that is scalable and secure where we can you know put our
[21:23] models run our you know data large amount of data sets that we have and also be elastic in nature so that if there is like large influx of transactions we are able to do it not only like you know do it well but do it
[21:37] do it in near real time. So that was requirement number one. Requirement number two was we all love AI. We all love to treat it as a blackbox. But guess what? No enterprise wants to treat it as a black box. They want
[21:51] explanability, right? They want to know if agent took an action or if an AI model recommended something. They want to be able to go back and say why did it do it and why was this, you know, um thing, you know, recommended and what is
[22:04] the reasoning behind it and most importantly what is the data that supports it for all good reasons, right? better governance, better compliance and even like you know making sure the decisions are made ethically. And the third important component is the
[22:17] traceability and auditability. They want everyone wants the data lineage into like you know how this data was used in the model version and how we can trace it back to the actual data set that led
[22:30] to a particularation point. So that is absolutely critical and again as you're talking to more and more clients one thing and I also constantly hear almost every organization today has either has
[22:44] an AI governance council or in the process of uh creating one because any software that they buy moving forward they want to make sure their AI governance council approves it and the things that they look for is these three
[22:56] things like agent uh the explanability the auditability and then you know the traceability. So these are absolute you know you like you know table stakes for you to be even in any conversation and
[23:08] the last but not the least is the what I call a govern like you know coworker digital coworker that can do a good portion of the regular day-to-day job
[23:21] and minimize the steps needed for complex actions keeping human in the loop right the reason why I say the governed is an very critical you know component we don't want agents to be running around and doing bunch of things and you don't even know who did what and
[23:34] you want a complete and you don't want your employees doing that, right? So, you want a complete govern structure where you know exactly how what happened and if you want to control what the agent output does, you want to be able
[23:48] to configure and control it. So, these are the four main things that we laid out for our journey and obviously we couldn't do this journey without you know partners. So that's when we looked at many partners. We chose Nagoro as our
[24:02] development partner and on the technology side we chose data bricks because we did our due diligence and we came to a clear conclusion. Data bricks had everything that we wanted the scalability the security the ML models
[24:15] that we can you know readily use and be able to do it in near real time. It checked all our boxes and then you know we chose you know data bricks and with that now we embarked on that journey. I'm happy to say today as we stand we
[24:28] have migrated more than 50% of our clients from the old private cloud environment that we had in the modern architecture uh that hosted in you know data bricks and AWS and um we we have
[24:42] overcome some of the the private cloud related limits that we faced before in terms of processing capacity and everything because every time we ran into a processing bottleneck we had to go run around add more servers and with
[24:54] the with the you know technology we were not able to do it at at at pace like the market was dictating now we are doing things 50% more faster than what we did using the data bricks technology and the
[25:08] nago team has been awesome and helping us now that has helped us lay lay out this AI foundation that that is helping us in everything that we do let me quickly call out some of the few things that we have launched and I've seen some
[25:21] enormous success already the first one is receipt analytics I want you guys like when you go back home and submit your expense report there's more than a 50% chance that that receipt is going to come to us for an analysis that's one of
[25:34] the solution that we launched in this newer platform which is nothing but when you submit an expense report and attach a receipt to it we look at the receipt completely and then using the large language models and the you know the AI
[25:47] models that we built in data bricks we can tell you whether each line item is like uh with within the compliance limit and the policy limit that your organization has established and also are there any risky purchase item and
[26:00] trust me people buy Rolex watches using their you know corporate car programs and we see all kinds of crazy stuff you know people do with their kind you know you know when whenever they do travel and expense so those are things that we
[26:12] catch and we do that in a much more effective way and a much more scalable way and almost near real time we can do it these days. The second one is the vendor statement reconciliation in the procure to pay you know world when your
[26:24] vendors submit the monthly quarterly their list of transactions and you're trying to reconcile that with your accounts payable. It's a nightmare. People have stopped doing that because if you have 150,000 vendors and each one
[26:37] is going to send you like 50 or 100 line items of worth of statements. There's no way you can reconcile. So what happens is there lot of leakages there. If there's a credit memo that was issued, we are not capturing it. So all of that
[26:50] like was not even looked at before but now that with the technology we are able to do that and we have launched that and we have you know clients using it and seeing some phenomenal value out of it. Global risk scoring is a machine learning scoring where we look at all
[27:04] the historical data that has come through and score every transaction that comes in so that we can predict if this is a risky transaction that's going to happen that will not comply to the policies that you have laid out and as I
[27:16] said policy monitoring that is custom to a particular organization we you know monitor the policy 24 by7 so with all of this we have a phenomenal platform today where as I said we can
[27:28] predict some of the you know risky purchase or risky transaction that's going to happen based on the label data that we have and based on the models that we have built in data bricks and we have added the agentic layer to it where we have the
[27:42] orchestration layer which is much more intelligent much more you know you know capability to reason and it can do many of the work that a typical you know auditor or anyone internal teams can do wherever it needs human in the loop
[27:55] it'll bring human in the loop for that action and deliver the most important results And with all of that we are not forgetting about the explanability and traceability because that'll continue to be the focus especially in the
[28:07] enterprise world as more and more companies start to use like agentic solutions and AI solutions. This is a must have the agentic and the tability and explanability. So that's what we are focused on and we are seeing some phenomenal results already. So um that's
[28:21] the high level view of how we transformed from being a private cloud SAS company to a modern AI based company like you know AI first company with like agentic solutions and most importantly
[28:34] we are able to do it at a much more like you know faster pace that we ever imagined like last two years back and most importantly we are also solving some critical you know um you know business challenges for our clients that
[28:46] um I'll hand it over Thank you, Bala. That was really useful. I particularly like the Rolex Rolex example. Oh, is that even surprised? I mean, people buy a lot of stuff. Chanel,
[28:59] you know, perfume mirror thing. >> Wow, that's giving me ideas. But uh but very interesting. Thank you so much. And you know one of the reasons we were hoping to speak with Bala on this on
[29:11] stage was to kind of bring this whole context of the specific customer which is really important in our belief like you know there are there is so much uh when we think of AI there are there are
[29:23] we think of them in very neat boxes we think of one platform two platforms we think in API okay I'm going to use this islang graph langchain you know SDK something is going to be data bricks or
[29:35] snowflake or something Right. But the real world is very messy, right? Like I mean for us the agentic trans enterprise transformation is a long arduous
[29:47] tough task to really do. It's easy to say it. It's a very fancy word to hear and get, you know, on a deck, but to really build the specific journey for a customer, it's non-trivial, right? And
[30:01] and so having that kind of context as the basis for transformation is one of the key things that we try and, you know, uh think about. I'm going to give you a couple of more examples and one more demo to sort of walk you through
[30:15] this idea, uh the story. So one of our customers uh has um has a has a very strong ops situation. They are actively
[30:27] at any given point of time um they have many products and for all of those products they really want to run their L1, L2, L3 uh situations and they spend
[30:39] a lot of money they spend a lot of energy just trying to get that right. Right. And since AI has come by, they are very actively asking us, asking other vendors, what can they do to
[30:52] really make that a much better experience. And so one of the conversations I was in, it came very obvious that, you know, when we're talking about transformation, they were like, okay, all of this stuff is great.
[31:05] How do you think AI can really help here in a reliable, active manner? Right? And what that meant for us was to really go and figure out their specific process and interestingly one of our teams was
[31:18] embedded in that activity and you know one of our teams is providing some kind of L3 support to them. So we spent you know we got all of them in a room and we like okay help me understand help us
[31:31] understand how you guys do these things. And so there about eight to 10 different engineering members and they all started talking about like hey this is how I when I get a ticket this is what I do and this is what I do or this works and
[31:43] this doesn't work or I have been fixing the same ticket for the last one year and you know I just got to I keep my notes and so the you know the first time I saw that ticket um because I don't have a way to sort of store it or
[31:56] amplify it or the system doesn't learn I keep getting similar things and now I have a pattern like I have fixed one ticket three different times over the last one year for example and there is no compounding there is no brain there
[32:08] is nothing that's in the system that can recall and build this right and it's very obvious for somebody who's building aic outcomes that this should go away so so things like that right and so eventually um I want to show you a
[32:20] platform which we are trying to you know build but the idea is mostly to sort of like you know first of all capture the key elements of what such a platform
[32:32] would entail So one of the first things is that if you are going to build an agentic intervention into a customer process, you should be able to show ROI.
[32:44] This has to be the first and the most important thing. If you do not have a way to think about ROI, then we think we haven't grasped the ground yet in any fundamental way. Right? So in this
[32:56] particular case for example um there there are several metrics that are being shown but one of the top ones here is this one thing called 2.5x right we we call it like the engineering hours
[33:08] leverage. So for example what work we are doing with the help of this platform it's giving us a 2.5 times amplification over a single manual engineer. That's
[33:21] the way we kind of articulate in this particular case an ROI for the sigantic intervention. There are several ways to define it, right? And depends on the process that you're really thinking about, right? But what I'm showing you is a dashboard of how are we, you know,
[33:35] even articulating some of these key KPI metrics. Now, this is an L3 specialist augmentation exercise. A bunch of engineers who are sort of experts in fixing a sort of a software problem have
[33:50] been augmented with the help of a platform like this. Let me just show you this in action. So I'm going to go to live operation and I'm going to inject a ticket and I'm going to inject a ticket here. I've you know the ticket is it's
[34:03] completely synthetic. It says inventory uploads failing for several dealers in the southeast region and there is a certain way things are happening. Now the moment I do that what happens is
[34:15] that a bunch of agents kick in, right? And and these agents are mimicking what typically would have been done by a bunch of manual engineers, right? And and one of the first skills that we
[34:27] really need to have is to be able to look at logs and do root cause analysis. Now this is what every L3 engineer tradition you know typically does except that you ask them and they would never say that. They would never say oh I do
[34:40] log investigation or an RCA as a first thing unless they are super articulate but unfortunately eight out of 10 engineers are not so articulate. So what we had to do is we had to distill that knowledge out after those conversations
[34:52] with those engineers and say okay here are the six seven things we typically do. We do log investigation we do fixing proposals. We do a similar past incident retrieval. We do code context assembly.
[35:06] We do you know all of these little things that we listed out where the kinds of distillation that we learned from the team itself right and then we build these skill skills into the agentic platform. Now what the skills do
[35:18] of course you know there's a bunch of other efforts that goes in but then you have all this stuff right and at some point the the algor the the agentic platform gets to a place where it sort of says okay I think here is what I
[35:31] think is the problem do you agree and the human being the human in the loop in this case has a bunch of choices they can look at the hypothesis that the RCA is throwing up
[35:45] they can on the sides kick off a coding agent and look at whether that makes sense. They can modify that hypothesis in a certain manner or you know they can do a deep investigation or something
[35:57] right. So the deep investigation would mean that I am asking for a more exhaustive code assembly lookup, right? And so now this whole process goes two steps deeper. You know, now we're going to burn more tokens to really go do
[36:11] this. And maybe you know what you're really saying is asking a senior engineer to look into this more deeply before committing a fix in any form. Now with platforms like this, we think that we can really change the way run
[36:23] operates for a company. And this applies to L1, L2, L3. the specific process that we are encoding here is very very specific to this customer. So these are the kinds of transformation journeys
[36:36] that typically exist um you know which we have to craft every time we are working with a customer. And so what I really wanted to say and share with you all is this whole idea of what this
[36:48] agentic enterprise is. How many different processes can be reimagined? how many ways we could really intervene and oversight's one example which is really working in this area to try and become agentic and we have several other
[37:00] examples that we work with but with that being said let me just stop here and take a pause and if you all have any questions for either me or Bala we are very happy to answer for

Learn more about the Databricks Data and AI platform.

The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.