3 Jul 2026 · 34m · Assaf Resnick · 20 quoted facts
Assaf Resnick argues that IT operations still run on human-heavy, reactive workflows built for a slower era, and cannot keep up with cloud, microservices and constant software change. His answer is agentic AI built on a company-owned IT knowledge graph. This lets AI agents detect, triage and help fix problems, and focus on preventing outages rather than reacting to them.
The central tension is ambition against caution. Resnick describes agents that spot a root cause in seconds instead of hours and could replace large outsourced teams. Yet he insists companies adopt them gradually, starting with a few safe workflows, keeping humans able to take over, and owning the knowledge graph so they are not locked in to one vendor.
Resnick says cloud, hybrid infrastructure and CI/CD mean operators must absorb about a thousand times more data at a thousand times the speed. Throwing more people at it is, in his words, a growing data problem met with the same workflow.
Resnick says outsourced operators average seven months before leaving. Companies keep paying to train people who don't know the business and then lose them, which he calls a poor fix for cost.
It combines live environment data, past incidents such as bridge-call transcripts, and lessons inferred by watching the best operators. In his ATM example, this lets an agent recall a past security change and propose a rollback in about thirty seconds.
Resnick says the graph should not sit in one vendor's walled garden, or switching vendors means losing the accumulated context. Consultants can help build it, but it should remain the customer's asset.
He compares it to how CIOs moved to public cloud: start with a handful of real, low-barrier workflows, win, then expand. He mentions agents offloading around 30% of work early on, with a roadmap toward much more.
It means AI agents that do IT operations work, such as spotting incidents, working out causes and recommending fixes, instead of people doing it all by hand. Resnick describes a team of specialist agents backed by a shared store of company knowledge.
It is a connected store of everything about a company's IT environment, including current alerts, past incidents and what its best engineers did. Resnick says it gives agents the context a veteran employee has, and without it the agent is still just an intern.
In Resnick's bank ATM example, a human-driven process took up to eight hours, including a six-hour bridge call. He says an agent could spot the issue and recommend a fix in about thirty seconds to minutes. This is his description, not measured data.
Resnick suggests the workforce shifts rather than disappears: instead of 300 outsourced operators reacting to issues, two or three site reliability engineers (engineers who keep systems running) manage the agents. He also says humans stay in the loop where judgment is needed.
Resnick recommends an evolution, not a revolution: pick a few real, low-risk workflows, let the agent join incidents and learn, then publicise the wins internally. He warns against changing tools, processes and org charts all at once.
Stated by the speakers. Each quote matches the transcript, which shows what was said, not that it is true.
Assaf Resnick says BigPanda's customers include banks handling half of European financial transactions and customers handling two thirds of US stock trading.
“we have banks that manage half the financial transactions in Europe, we have bank, we have customers that manage, you know, uh, two thirds of the stock trading in the United States of America.”
Assaf Resnick says a major US airline has around 57 critical applications whose outage beyond fifteen minutes would ground its entire fleet.
“You have something like fifty seven critical applications that if any one of them is down for more than fifteen minutes, your entire fleet globally has to be grounded.”
Assaf Resnick says IT operators now must consume roughly a thousand times more data at a thousand times the velocity than before.
“Cause they have to consume a thousand x more data at a thousand x more velocity to be able to make decisions around.”
Assaf Resnick says the enterprise tech stack has transformed over fifteen to twenty years through cloud, microservices and CI/CD, while IT operations methods have barely changed.
“Cloud, microservices, CICD, and the approach to IT ops has not changed dramatically at all.”
Assaf Resnick says thirty years ago most enterprise systems lived in physical data centers where the pace of change was slow.
“if you rewind thirty years ago, you know, most of that stuff was living inside of a data center.”
Assaf Resnick says a Fortune 500 healthcare company using BigPanda halved its self-inflicted incidents, which he valued at $1 billion.
“They've reduced their number of own goals and incidents by half. $1 billion.”
Assaf Resnick describes replacing roughly 300 outsourced operators with an agent layer managed by two or three site reliability engineers.
“instead of having, you know, 300 operators reacting to issues that are outsourced in a low cost labor environment, uh, we're moving towards. Hey, that's an agentic layer that is managed by two or three site reliability engineers”
Assaf Resnick says major outages often pull around 300 experienced people into a bridge call lasting about six hours.
“Instead of calling 300 of your most experienced people into a bridge call from hell, uh, for six hours”
Assaf Resnick says a fifteen-minute triage target for a network issue can stretch to two hours when handled by inexperienced remote staff.
“that fifteen minute SLA turns into a two hour comedy of errors, with lots of phone calls and missed signals and drop balls.”
In Assaf Resnick's ATM example, the root cause was a network security update made three weeks earlier, found only after six hours of bridge-call investigation.
“only six hours later do we realize, ah, someone made some security update to a network three weeks ago, and now it's having unintended consequences.”
Assaf Resnick says an agent with an IT knowledge base could detect the same problem in about thirty seconds.
“Doesn't take me fifteen minutes to see, it doesn't take me two hours to see a problem, takes me about thirty seconds to see a problem.”
Assaf Resnick says an agentic solution has two layers: agents that reason, and a context or knowledge layer underneath.
“any good agentic solution typically has two layers. There's the actual agents that have the reasoning and the logic, and then there's the underneath that, there's the context or knowledge underpinning that.”
Assaf Resnick says an agentic coworker can grow to offload about 30% of operations work, with a roadmap toward all of it.
“Until it can. You're looking around, it's actually offloading 30% of the work. Now there's a roadmap to 100% of the work.”
Assaf Resnick compares agentic adoption to CIOs' gradual, workflow-by-workflow move to public cloud about fifteen years ago.
“think of how CIOs moved to public cloud fifteen years ago. They didn't come and say, Hey, we're 100% of data centers, we're gonna go to public cloud overnight.”
Assaf Resnick likens agentic operations to an autonomous car where humans can still take over, and agents hand off when below a confidence threshold.
“you know, driving an autonomous car, there's still a steering wheel, there's still brakes.”
Assaf Resnick says an enterprise's IT knowledge graph should be owned by the enterprise and not locked inside a specific vendor.
“It shouldn't sit inside of the walled garden of, you know, a specific vendor, right?”
Assaf Resnick says some consultants have taken about 200 other large enterprises through the agentic IT ops transformation.
“have done it for 200 other large,”
Assaf Resnick says BigPanda offers agentic change-management specialists that review potential enterprise changes to block risky ones.
“We're also bringing our change management specialists, uh, to help make sure that, you know, we can look at all of the potential changes an enterprise may be making”
Assaf Resnick says a typical large enterprise has hundreds or thousands of people in IT operations, split into L1 generalists, specialists and change-management teams.
“they have hundreds, if not thousands of people working in it operations, and they have different teams.”
BigPanda is described as an agentic AI platform that helps enterprise IT operations teams detect, triage and resolve incidents faster.
“BigPanda is an agentic AI platform that helps enterprise IT operations teams detect, triage and resolve incidents faster.”
Resnick defines IT ops and explains how cloud, CI/CD and faster change flooded operators with data they cannot process.
“they have to consume a thousand x more data at a thousand x more velocity”
Downtime hurts everyone, but banks, airlines and healthcare face the highest stakes, where seconds or lives are involved.
“you have something like fifty seven critical applications that if any one of them is down”
Resnick argues that cost-cutting through outsourcing created high turnover and lost knowledge while the tech stack changed radically.
“it's the definition of insanity doing the same thing, expecting a different result”
The core idea: agents need context and reasoning layers, and the graph captures live state, history and tribal knowledge.
“imagine taking all of that tribal knowledge, institutionalizing that into an operational knowledge it knowledge graph”
Resnick contrasts an eight-hour human response with an agent that finds the root cause in seconds, and cites healthcare outage savings.
“Doesn't take me fifteen minutes to see, it doesn't take me two hours to see a problem”
He recommends starting with a few workflows, keeping humans able to take over, and compares the path to cloud adoption.
“driving an autonomous car, there's still a steering wheel, there's still brakes”
The knowledge graph should belong to the enterprise, and agents can extend beyond L1 to change management and major incidents.
“operations is a team sport”
Welcome everyone to the Emerj AI in Business podcast. Today's guest is Assaf Resnick, CEO and founder at BigPanda. BigPanda is an agentic AI platform that helps enterprise IT operations teams detect, triage and resolve incidents faster. Assaf breaks down why decade old, human driven IT operation models can no longer keep pace with today's cloud and microservices environments. And how building an internal IT knowledge graph lets AI agents take on level one detection, triage and even change management.
He also explains why enterprises get the best results by starting with a handful of high confidence use cases and expanding agent responsibility gradually, rather than attempting to replace their entire operations model at once. Today's episode is sponsored by BigPanda. Do you sell AI products or services? Emerj gives you access through trusted content and real conversations. Learn how leading AI brands like Nvidia and Google Cloud work with Emerj to reach Fortune 500 AI buyers. Download our media kit at Emerj dot com slash AD1.
That's e m e r j dot com slash AD1. Now the conversation with Assaf. Assaf, welcome to the show.
Thank you for having me.
So Assaf, really glad to have you on. You've been building toward agentic IT operations for more than a decade. Long before the rest were catching up on this, uh, for BigPanda, banner sits at the center of the operational reality that CTOs, IT ops are dealing with every day. Incidents, outages, noise and the pressure to keep increasingly complex systems running. So this conversation is going to be practical in the weeds vision of what agentic AI ops actually looks like.
So from where you sitting, what is the broader, let's say, landscape of IT ops look like and how people approach it at the moment?
Well, I mean, what is it ops? It ops is basically if you are a large enterprise, so think you're a large bank, you're a large airline, an online retailer. You have a lot of digital services that you have to keep running, and those digital services run on infrastructure. They run on servers and clouds and VMs, and they have applications and networks. And it's the job, and those things move all the time and they break all the time.
And so the job of of it ops underneath a a CIO or a CTO is to make sure that all that all those digital services keep running. And so your customers and your employees can use all those digital services, and so it's their job to keep all that running. That's really challenging these days, cause if you rewind thirty years ago, you know, most of that stuff was living inside of a data center. And you manage those servers physically, and the pace of change was fairly slow.
But, you know, you fast forward to today, and you have hybrid cloud and virtualization and uh, uh, CICD. So you have all these moving parts that are moving way faster than they've ever moved before. They're breaking way faster than they've they ever were before. And that's creating this tsunami of data that humans, you know, operators in it ops are really struggling to keep up with. Cause they have to consume a thousand x more data at a thousand x more velocity to be able to make decisions around.
Hey, what's going on with my, you know, uh, health of my my services? And if something's broken, how do I find and fix that before my customers are impacted?
and from your point of view, obviously, we've seen the stack growing. I think that's one of the reasons it's it's become difficult is there's just so many different software, so many apps that you need to manage. Where in the past, you might have had one or two, it's just grown to. To the extent where a human cannot process all of the information that's coming through. So from that shift, are there specific industries that are struggling more than others?
When it comes to this, I'm thinking that regulated industries might find it more difficult. Or is it across the board that all business is being impacted by just the growing amount of incidents and fixes, et cetera? Or do you see it in specific industries? More, let's say, poignant?
So it's a good question, you know? The problem is, first of all, universal, where if you have a large global business, any downtime where your business is not operating, your supply chain is not operating, that causes business pain. Now there are some verticals that are more susceptible to to the pain of downtime. So, you know, if you are, we have banks that manage half the financial transactions in Europe, we have bank, we have customers that manage, you know, uh, two thirds of the stock trading in the United States of America.
And so those are systems where seconds count, in fact, milliseconds count. And so when you are down, folks feel it. We have, you know, many of the largest airlines in the world, and you know, every one of those airlines think of the United States. You have something like fifty seven critical applications that if any one of them is down for more than fifteen minutes, your entire fleet globally has to be grounded. You know, and there's a very special place in hell for, uh, airlines that strand their pay, uh, their their, their passengers in the airport.
So any any company where time is important, uh, you know, we have customers that are in healthcare that are, you know, in the critical path of doing organ transplants and cancer treatments. And when those guys are down, it's not financial transactions
it's life.
that get locked, it's it's people's lives and people's health. So it it really is a universal problem.
And what would you say is the biggest blind spot? Where are IT ops and CTOs struggling the most? What is the biggest blind spot they're struggling with? Or is it just the triage? Trying to figure out what to hit first when when there is, uh, an incident coming up?
Well, the biggest blind spot or challenge is that, you know, your tech, the tech, the enterprise tech stack has radically transformed in the last fifteen, twenty years. Cloud, microservices, CICD, and the approach to IT ops has not changed dramatically at all. It's still throwing a bunch of humans and human workflow. Now, in the last thirty years, the trend to try to reduce costs has been, hey, let's outsource a lot of that operations to, you know, GSIs and low cost labor environments, try to reduce costs.
And that has not been a very good fix, because yes, it's reduced cost, but then you're introducing, you know, outsourced employees, where the average attrition rate is seven months. You have this constant stream of young engineers who don't know your business, making a lot of mistakes, take a lot of time to get ramped. By the time they get ramped, they get a new job. And so it's the definition of insanity doing the same thing, expecting a different result, throwing more and more workflows at a data problem that's growing exponentially.
That is broken,
And I don't want to put words in your mouth, but that sort of gets us to what is the what is the alternative to that? So if we're not outsourcing it and we're not getting costs down in that way, because it's clearly not the solution, the solution is moving to s towards something new. And that is agentic, and you've sort of just hit on the most difficult problem when it comes to agentic, but it's not an agentic alone problem.
You're looking at the same problem you just mentioned. When you're going to onboard a new employee, there is a lot of work that goes into it. Onboarding takes time, giving a new employee context, that takes time. So maybe lay out to me what that shift is. It is agentic, but aren't you sitting with the same problem? You still have to onboard a new employee when it comes to agentic. But you are saying this is a new way forward, and and tell me what that new way forward looks like.
you know, it's basically saying, Hey, thirty years, the approach to IT ops is to build a a highly manual set of workflows that are reacting to problems, right? Now it's how do I use agentic automation to a automate intelligently a lot of that workflow and b focus on prevention rather than reaction, right? So ai, so so how does that start? Onboard a new employee? What do you do when you're onboarding a new employee? You're basically downloading into their brain all of the context, all of the knowledge, access and awareness to all of the tools and all of the resources inside of, you know, the enterprise environment, all the things this employee needs to do their job well.
And so, you know, any good agentic solution typically has two layers. There's the actual agents that have the reasoning and the logic, and then there's the underneath that, there's the context or knowledge underpinning that. You know, agents and humans need to get awareness and and make smart decisions, and so for us, it it starts with really harnessing all of your enterprise knowledge. I mean, the nice thing about is so when humans, I with human operators, hey, you can only hold so much in your brain.
But when you're building a context graph or a knowledge graph, we call it a nineteen knowledge graph for agents, you can put a lot of stuff in there. And so one is you can give put inside that, not that it knowledge graph, all of the awareness of what's going on right now in your environment. At a scale that a human being just couldn't connect all those dots, that's one, but you can put so much more. You can say, Hey, let me give you the knowledge of everything that's happened in my environment.
So think of, you know, every it ops team in a large enterprise has that twenty year veteran who understands, you know, what's going on. They've seen this problem three years ago, they understand this messy environment that hasn't been well documented. Well, imagine taking all of that tribal knowledge, institutionalizing that into an operational knowledge it knowledge graph, and then making it available to all of your agents. But you have a, you know, a number of agents that are specialists that have the history and the intuition, uh, of your best employee.
And then finally, uh, there is a lot of knowledge that is already trapped and that's just trapped in people's heads, not written down anywhere. And how does that come out? Well, it comes out by, hey, when a problem happened and your employee, uh, uh, uh, the your best it operations person jumped in there. Well, what did they do? Can I observe what that person does? Can I infer knowledge? can I infer learnings from there? can I get better every day?
And the answer is yes, you can. It's not written down anywhere, but when you have an agent who's a coworker to your best teams, you can learn from your best. And so make sure that you're taking all that learned experience, institutionalizing it into an it knowledge graph.
And of course, the the one year is number one, as you said, it can take in a lot more information than a human can. And then the second is they're not going to be leaving the company in seven months. It just keeps it, not only retains it, keeps learning, and it's not leaving the company every seven months. So so you're ramping up the onboarding, but it's also you're investing in a long term. Call it an employee, an agent that will not leave the company at any time.
And then the last point is the fact that it's on twenty four seven, whereas for humans, there there needs to be some kind of rotation between people. So there's no breaking continuity. It's a twenty four seven situation where the agent is running, which, of course, at the moment, you need shifts, et cetera. But there's always, we've heard from, even from manufacturing, that once that handoff happens, when that turnover happens, a lot can happen in that moment.
When when, when one person is leaving and the next is taking over, some things do get lost in that handover, where with agents, that's not a problem at all.
Absolutely, so, you know, the nature of global business today is it runs twenty four seven. And you know, when you have a follow the sun kind of shifts and you're doing three handoffs a day, there's always problems. And like you said before, especially when you have employees that are churning every seven months, you're losing that institutional knowledge. Uh, that's very, very painful. So something that gets smarter every day, builds that enduring knowledge base and then brings it twenty four seven across the globe.
Does it very, very quickly, does it proactively? Yeah, it's a lot of benefits to that.
And tell me if you, you've seen this deployed, this is not just a a theory and an idea. So where you've seen it deployed, what does that look like? And what gains are being made when this is being implemented and scaled?
Yeah, so, you know, we've deployed it in the world's largest, most complex environments. So, you know, it looks like, uh, it. It depends from which perspective you're looking at. So, first and foremost, you wanna look at it from the perspective of your customers, right? So you wanna make sure that your customers have less outages, less service disruptions, better quality of service, which leads to, you know, better brand and and and better and better revenue, better customer retention.
Uh, it also means a pretty radical shift in how you're thinking about it ops. So, so moving from a an operational perspective where you have to hire a very large team of operators to a systems management perspective. So think of, hey, instead of having, you know, 300 operators reacting to issues that are outsourced in a low cost labor environment, uh, we're moving towards. Hey, that's an agentic layer that is managed by two or three site reliability engineers, and they're managing those agents.
And so, you know, your cost structure is radically changed, your uh, operational efficiency is radically changed. At the end of the day, like I said, the the service and the quality you're giving your customers is much, much higher.
thinking about this, I'm thinking to myself, who in within the workflow at the moment? At the moment? Obviously, you've got your first line of agents that are dealing with the actual incidents and have to resolve them. And then you've got a management layer above that. Who's going to feel the relief first within this system, and who's going to have to put in a bit of work to get it to a point where agents can take over?
How does the workflow change? What is going to be happening with those first layer of actual incidents, the agents that are handling it now?
Hey, it's a great question. So, uh, you know, num, it's gonna impact the number of folks within it operations, uh, for the better. Cause. It starts with, if you think of agentic it ops, it starts with first of all, not better reaction, reactivity, but prevention. So first and foremost, it starts by, hey, we're gonna learn from the past, we're gonna make a lot less mistakes and have a lot less outages, right? So think of we work with one fortune 500 healthcare company.
They've reduced their number of own goals and incidents by half. $1 billion. The CIO and the C CFO both can feel that when you're reducing outages by half $1 billion, that's one. Two is when potential incident starts. If your L1 operators running around everywhere, trying to connect the dots, making a bunch of phone calls all over the enterprise, calling your network admin and your cloud architect, and your your developers and saying, Do you see anything?
That whole runaround doesn't happen anymore, and so things are much more, more, more seamless. And when you, you know if and when you do have a honest to goodness outage. Instead of calling 300 of your most experienced people into a bridge call from hell, uh, for six hours, uh, we're gonna bring our team of, you know, we're uh, we're gonna swarm. Our team of agentic specialists, we'll bring our agentic database administrator, our agentic network expert and cloud expert.
And we're gonna do that work for you, and we're gonna do that deep investigation for you. And so, you know, you're gonna feel it immediately in multiple parts of your company.
and if if I could ask you to get extremely practical. So at the moment, what happens is an incident happened, there's downtime on a system, somebody's trying to log into the UH system and it's just dead, nothing's happening. The A ticket comes through to your agent and it tells your agent, Listen, there's downtime here. We need to be up and running, figure out what the problem is. They have to go and see. Well, first of all, they have to triage because maybe they got ten tickets and they need to figure out which one is most important.
And then they need to go and try and figure out what the problem is and then solve it at the moment. And then it escalates. If it's uh, uh, like you just said, when it's a very big problem, you bring in the entire team. How does this exactly change? Where in the process? So instead of we're talking about prevention, so does the agentic AI monitor all of your systems? It sees a alert is coming up, or it sees an in a situation or a a bunch of things that happened in the past.
And this used to be the result, there used to be downtime when all of this happens. And it already triggers, uh, the agent to go into that system and look if there's a problem. What exactly does this look like in in the workflow when we get really practical?
so let's get an example. Let's say you are a very large bank, right? And you wanna make sure that all of your ATMs, you know, are working, your ATMs in Paris or your ATMs in New York. Cause when your ATMs aren't working, customers can't, you know, take money, uh, that they need to live their lives, right? And so, uh, job number one, keep those services running, okay? And so when an issue does come up today, a, and let's say you have a network issue.
Uh, it's sending up lots and lots and lots of alerts that are surrounded by lots and lots of other alerts. And so you have humans, right in in your L1 triage center that have to take those thousands of alert and figure out what's actually happening. Do I really have an issue here? Is this a national issue? is it localized? And, you know, that work should take fifteen minutes, cause if your ATMs are down, you wanna know really, really quickly.
Now, typically, you've sent out that, you know, that job is managed by someone in a continent far away, who's been on the job for three months, doesn't really know what they're looking at. And you know, that fifteen minute SLA turns into a two hour comedy of errors, with lots of phone calls and missed signals and drop balls. And so by the time someone in your organization notices, Hey, I've got a problem. You've had customers in New York frustrated because of broken ATMs for two hours.
That's very, very problematic. That's problem number one. At that point, you realize, Oh no, we have a real business impacting issue. Now. Let's get on a bridge call and call in everyone who might possibly know something about what's going on. And then we start, you know, having conversations. Do you see any things in the systems? You own? No, I don't see anything. Do you see anything in the networks that you own? No, I don't see anything.
And everyone's protesting innocence. And only six hours later do we realize, ah, someone made some security update to a network three weeks ago, and now it's having unintended consequences. And so today, you know, and all the while, that's eight hours to fix an issue, your customers are furious. So that's where it really becomes very real. And when you have, you know, agentic IT ops that says, Hey, I've got this IT knowledge base and says, Hey, you know, I've seen a symptom like that before.
Seven months ago, I, I can see that there's a problem. Doesn't take me fifteen minutes to see, it doesn't take me two hours to see a problem, takes me about thirty seconds to see a problem. I can learn from the past. I can say, Hey, when we had an issue like this, there was a bridge call seven months ago. I looked at that transcript of the bridge call, I got the piece of information. I said, Okay, last time this happened, it was related to a chain, a security upgrade that someone made.
Let me see if something similar happened. Oh, yeah, it did. Let me roll that back. Boom. So, you know, in about thirty seconds, we've been able to spot an issue, get situational awareness and understand what the root cause is and recommend a fix. So where before it was eight hours, that sucked in time of, you know, hundreds of your most strategic employees, and led to really unsatisfying customer outcomes. Uh, agentic, uh platform can do that work in in minutes.
That that is, uh, for every part of that system, from the customer to the CFO. Like you said, when you're saving money, it really does affect every layer of the entire process. So let's say I am the CTOIT ops of a large company and this sounds promising. But I've been using an external service provider for a while. They are sort of sitting with all of the information because they've been handling the tickets. Now I need to build this knowledge graph.
What does the lift look like? What does the process look like when you want to start implementing this new agentic system within your own company and not be using that external service provider anymore? Is that an easy lift?
uh, it, it's, you know, my philos- our philosophy is that it should be an evolution, not a revolution, right? So, uh, the wrong way to do it is to do a boil the ocean strategy, where we're gonna change everything overnight, we're gonna change our, our tooling and our processes and our relationships with outsourcers and our org charts and everything all at once. No, uh, the right way to think about it is I'm hiring a co, an agentic coworker.
I'm gonna give it, you know, it's gonna join me in my bridge calls. It's gonna take over a little bit of work, uh, I'm gonna give it access to all of our data. And it's gonna get smarter every day, and then it's gonna learn to be able to say, Hey, I think I got this, and I think I got this, and I think I got this workflow. And every day it's gonna get smarter and smarter and smarter.
Until it can. You're looking around, it's actually offloading 30% of the work. Now there's a roadmap to 100% of the work. So, uh, you know, it's it's really important to choose, you know, think of how CIOs moved to public cloud fifteen years ago. They didn't come and say, Hey, we're 100% of data centers, we're gonna go to public cloud overnight. No, they said, Hey, let me put a handful of workflows into public cloud, let me see how it works, let me get familiar with it.
Let me put some more important workflows and and assets in the cloud. And eventually, there's enough confidence to say, Hey, I'm just gonna move my entire estate to the cloud. So the same thing with with agentic operations, you know, you don't wanna say, hey, I'm gonna radically, uh, revolutionize my entire strategy overnight. No, I'm gonna give it a handful of workflows. I'm gonna get a quick win. I'm gonna make sure that, you know, I'm.
I'm pointing it to air to to problems that are very real, that have a low barrier to entry. And then, you know, once there's success, I'm gonna talk about that success internally, I'm gonna evangelize that success internally, and that's how you start getting, uh, uh, that that snowball rolling.
And I think that sort of answers my question where, where do you start? It's identify the point where you can start bringing that agent on. And th that process, from what I'm hearing from you, literally looks like when you have the next incident, start getting that agent in, start bring, uh, building that knowledge. And just for interest's sake, when? When this whole system is implemented and it's working well, where do human agents then still get into the loop?
Where? where are they still important? And and need to be part of the process
Yeah, so think about it like, you know, driving an autonomous car, there's still a steering wheel, there's still brakes. And whenever you know you need to step in and take over, you can do that. So, you know, AI, you know, AI or agentic operations is not, you know, there's certain areas where there's a confidence threshold. And the agentic platform can say, Hey, I've got this. And then there's certain areas. We say, Hey, I need to hand this off to another team member who maybe has more experience, uh, for this particular issue.
And so part of creating that kind of man machine interface is an agentic platform. That knows when it can take over for a fully autonomous workflow and understand when it has to hand it over to a t a human teammate and say, Hey, I've done my part, let me hand over the baton to you. And so a lot of the system will will automatically know that, and whether it needs to. I need to hand it off to an L1.
Or hey, I already know exactly who the specialist is in my org, and I'm gonna send a ticket to that person. Except I'm gonna save six hours of mistakes before it gets to the
before you hand it off to the right person, not the whole team, to the right person. So th thinking about this to me, it seems like, uh, we've heard this across quite a lot of conversations we've had. Is that as a business, and I think especially in this specific space? The whole IT ops thing has sort of, in a many cases, been handed off to third party service providers to handle. Whether it was for cost or whether the business just wanted to focus on its own operations rather than worrying about the IT ops.
Is this a situation where it's best to keep things in house and you build this agentic system with within your company? Or is it like, in most cases, it's go to the experts that have been building this for a while? This is their focus and bring those teams in to help you implement this agentic system. Have you seen both ways people building it themselves and what what would be the the way to go? Or is it a not a build versus buy, but rather a like we see in most cases, it's a combination?
well, you know, um, I, I don't think that there's one silver bullet way of doing it. I think the important thing is that y you build an IT knowledge graph that encapsulates all of the knowledge and data and history within your enterprise, and that should be yours. It shouldn't sit inside of the walled garden of, you know, a specific vendor, right? And so if you end up moving vendors, uh, you lose
but it would be both
all of that critical, you know, context and knowledge. Now, is there value in bringing in consultants who have expertise in building an it knowledge graph for it ops? Oftentimes, yes, cause you may have the experience in house and you may not have the experience in house. And there's also a lot of benefit for, you know, working with consultants who have been on this transformational journey for agentic IT ops and have done it for 200 other large,
Seen the mistakes.
you know, enterprises that look just like you do, and are as complex and bespoke as you are. There's a lot of benefits to that, uh. But the important thing is that they're helping you build an enduring knowledge graph that is yours, that you can build agentic workflows on top of.
And just for the bigger picture before we end off is we've got L1 sort of as that first line where we can bring in agentic. Is there a future where it's not just the L1, but it's a whole system where it's all one integrated system that we should be thinking of in the future?
So the, you know, this concept of a team of agentic specialists that are working together, uh, is how you should think about it. And so I so, if you think of us, you know, if you think of a typical enterprise, you know, they. They have hundreds, if not thousands of people working in it operations, and they have different teams. So they have the L1 generalists who are kind of the first responders, and they have a whole series of specialists that specialize in very particular part of your stack.
And they have other teams that are doing change management, making sure that no one in the organization is making risky changes that might bring down the organization. And so our concept is, hey, that whole uh, it operations team that that folks have built over the last forty years. We're bringing agentic specialists for every one of those domains, and they're working together and they're making each other better. They're really working, you know, operations is a team sport.
And so, you know, for us is we bring those L1 agents to, uh, help automate, you know, initial, uh, detection and response. We're also bringing our change management specialists, uh, to help make sure that, you know, we can look at all of the potential changes an enterprise may be making and make sure they're not letting those risky, uh changes go through. And then we're also bringing our specialists around SR Es and and major incidents. So when there is a major incident, you're not calling in 200 of your most expensive people and disrupting their strategic work.
You're actually, we're swarming our own agents that understand, at a very, very deep level, your networks, your databases, your applications. And so we're already, you know, well beyond, uh, an an L1 agent, which is really cool by itself. We're really, uh, bringing agents across that entire spectrum of a large enterprise, uh, it ops. So if you're a CIO, uh, you can bring someone like a something like a BigPanda agentic platform and say, hey, it's got the ability to offload every single one of my critical team.
So I'm going to try and nutshell the entire episode and you'll agree with me. We are missed something. So to me, it seems like the biggest problem is the system has become too complex for for the human brain to harness all of all of the information and make fast, accurate decisions. To bring down downtime because customers just don't have any appetite for downtime anymore. They've grown to a point where they don't tolerate any downtime, so the business needs to adapt to that.
The first step would be to build that knowledge graph within your company so that you've got something to onboard your agent with. Because it is a very intelligent intern, but it's still an intern until you've given it the knowledge graph of your company. And then for IT ops, the leaders, they need to really think as system managers rather than people managers. It's how does the system work? Where does the agent come in? What is the next step?
And that is a whole, I wouldn't say, new way of thinking, but it is a shift in thinking of how you manage your IT ops. And then I love the idea of it's not a revolution, it's an evolution. You start training your agent and let it take over s the parts that it can handle, still bring the human in the loop where it's necessary. And we don't have to think about this as just the first layer, the L1 that we're covering.
Towards the future, we can look at the whole system and implement agents as specialists where they are needed. And then we'll open up the humans to the more strategic thinking and where we can go beyond that. What did I miss? Assaf, help me out.
Uh, I think you got it pretty well. I think you understood it very, very well. You Perfect. got it.
Uh, thank you so much for your time. Looking forward to the R rest of the series.
Thank you so much.
Wrapping up today's episode, I think there are three key takeaways from the conversation with Assaf. First, IT operations have outgrown human scale triage. Modern cloud and microservices environments generate far more data than manual workflows can process in time to prevent customer impact. Second, the fastest path to resolution comes from institutionalizing tribal knowledge into a knowledge graph that gives AI agents the same context and instinct as the team's most experienced operators. And finally, the winning approach is evolutionary, not revolutionary.
Start with a handful of high confidence use cases, prove the value and expand agent responsibility over time, while keeping humans in the loop for the calls that need judgment. Do you sell AI products or services? Emerj gives you access through trusted content and real conversations. Learn how leading AI brands like Nvidia and Google Cloud work with Emerj to reach Fortune 500 AI buyers. Download our media kit at Emerj dot com slash AD1. That's e m e r j dot com slash AD1.
On behalf of the team at Emerj, we'll see you on the next episode.