AI agents are starting to act on their own. They answer customers. They route work. They make choices that used to belong to people. It feels fast and efficient until a decision lands wrong and no one can explain why it happened.

There is a line between automation and autonomy. Most organizations have crossed it without defining who is accountable. So when an AI agent makes the wrong call, who gets the first call? How do you decide what an agent can handle today and what must still wait for a human? And what happens when the model is confident but the outcome hurts trust?

This is not about slowing progress. It is about leadership keeping pace with the systems they put in motion.

Transcript

Sanjog Aul [00:00:00]:
Hello and welcome to CTN. To learn more about the show, please visit ciotalknetwork.com and today’s topic is about AI Agents in the Wild, which means AI agents are all over the place but let me tell the topic again, AI Agents in the Wild, who signs the incident report. So what are we talking about here? It is basically AI agents and accountability. So these systems that we are seeing today, they are to be run via AI agents and it has to make decisions itself.

Sanjog Aul [00:00:44]:
So while it might give you the speed and the performance and everything else that you would expect from AI agents, but bottom line is if something doesn’t go as you expect, who is accountable? Who will sign the incident report, if you will, and who’s going to take ownership of getting it all solved? So that’s what we are here to talk about. Now my guest is Aarti Singh, who’s the enterprise CIO with the Mahindra Group. Hey Aarti, how are you?

Aarti Singh [00:01:12]:
Hi, good evening. I’m doing well. Thank you for having me here.

Sanjog Aul [00:01:16]:
Absolutely, the pleasure is ours and as you see that I’ve set the stage for something which everybody’s talking about and everybody’s equally concerned about tackling and maybe winning over this. So if you were to give your perception and what your read is on the AI agents, not the buzz but actual reading on as AI agents getting deployed and adopted and being successful in the enterprise setting, what would you say to that?

Aarti Singh [00:01:46]:
So I think obviously, a lot of benefit comes by using various AI models and AI agents and at least from my experience, the first ones that we really tried to go after are the ones where the task is predictable, the outcomes are also predictable and you can verify and you can trace back to, why AI agents took a particular path or not. So if I just look back, the service desk area where you’re trying to do self healing, etc. That was a very easy one to pick up because of the reasons that I just laid out earlier and we are seeing success with that in earlier organizations as well. We saw success along those lines. So I would say where it is low risk, that’s where you start. Where you have good data, that is obviously a must, you have accurate data and you have predictable outcomes.

Aarti Singh [00:03:00]:
The more outcomes are predictable if you have different data sets, different days and still you kind of predict what that outcome would be and the confidence level is high of those predictions, that is when you would try and go, in creating AI agents. There’s a lot of speed, etc that you get with it, obviously, lot of scale. So to do or not to do so I would say to do, but you have to pick out the areas where they can be fully autonomous, areas where you deploy AI agents with human assistance or a human in the loop and some which would need more tweaking and more learning before you can actually go in some areas where you would not want to touch because the risk could be very high for any organization.

Sanjog Aul [00:03:53]:
See, I would like to build on this word predictability that you mentioned and frankly, as humans, we need predictability in our life, perhaps that’s where this all starts and even in an enterprise setting, in a business, we want that predictability. So that’s a great start. Now, if we start talking about predictability, can we vouch for it that it’s going to happen every time because you got an agent which is making decisions in itself, we can’t be 100% sure. So does this become a function of the potential risk tolerance of an organization for them to say, it’s not 100% predictable, but at least we have seen enough that it may be predictable, say 90% of the time. So would you still take a leap given the stakes?

Aarti Singh [00:04:44]:
Yes, it would also depend on what kind of scenario. So like where the risks are relatively lower, you would definitely pick it and then you also have to build guardrails in the system. So like, there is a degree of drift that maybe you would allow an agent which is acceptable, and after that you will want to maybe hand it over to a human being to finally conclude but in certain areas, will you be good with 90%, the answer is yes. In certain cases, will you be okay, like because I also came from aerospace.

Aarti Singh [00:05:26]:
Will you be okay with 90%, in some cases, the answer is absolutely not. So I would say for any organization, you’ll have to. It’s all about the risk appetite and the area where you’re implementing the AI agent. So you would first learn to implement the AI agents in areas where the risk is low and you’re able to absorb that risk and where the stakes are very high, you would not want to implement that without any human intervention.

Sanjog Aul [00:05:56]:
So do you think we are at the level we are today with the agent side of it? AI, we have been around the block. We have been tinkering with it. Agents are relatively new and the explainability is also another thing that comes up to say, I want to look within, under the hood, but not every time that’s even possible, especially at runtime. Do you think we have reached A level of maturity that we could look at that risk appetite. When you have enough data for you to say, okay, this is the likelihood of this working versus not working, that becomes the basis of somebody saying this is the amount of risk we have and when you quantify the risk is when you can take a decision. What do you think is the level of maturity in terms of the autonomy that agentic world offers, regardless,

Sanjog Aul [00:06:51]:
but the same path will be taken every time for you to say yes, it is not going to have that much deviation for you to be able to base your risk tolerance and risk measurement on.

Aarti Singh [00:07:06]:
So I would say sometimes explainability, you could do it when it is in a controlled environment and in certain cases, like see everybody is building AI agents, so you may land up using a SaaS product where it comes from a provider and you don’t own it, you don’t control it, you don’t have visibility to it. So what we’ve got to do is really have an agreement and understand with the partner as to what is the level of traceability, etc, that they have to offer and then contractual obligations, you could have indemnification and things like that and then you see like what is that risk as a collective organization, what are you willing to take? So that is where you don’t have any control or it’s like a black box on how do you really handle it. Coming to say an enterprise where you are building agents on your own.

Aarti Singh [00:08:07]:
Traceability is very important, as I said. Can you trace everything? The answer would be not maybe 100% but the results that the AI agents finally say the task that they do, if you look at the outcome and how correct it is, the moment you feed different data types, etc to it and it’ll also learn and the more data types, newer data types that are thrown to it, you will see that what is the level of drift that is coming? You will see in how many cases, like if somebody answered correctly, sometimes where it was not so correct and third, very important, is it hallucinating, guessing or trying to improvise on its own, it should not do that. So a very clear, at least in scope, out of scope. So if this is your boundary that you have set and what are the things that you’re able to do in your boundary, as long as it’s playing within the boundary, it is okay and the moment it goes into out of scope areas like typically an agent should understand that this is not my area and then hand it over.

Aarti Singh [00:09:23]:
So if you’re able to do some of those things, definitely, that you will find, I would say an indirect way to trace, how your agent is doing or given different scenarios, how is it reacting. So sometimes you can verify it very clearly and sometimes you’ll have to look at all the outcomes to see that how is it behaving. So you’ll have to have, there’s no like one way and how you solve for it. So you’ll have to have different ways in which you can see like how a agent is performing and then are you able to trace, different data sets and its outcomes based on, what the agent finally produces and especially decides not to do?

Sanjog Aul [00:10:12]:
So, like as you or you mentioned about some partners like who you would work with and it’s you getting comfortable with them that are they predictable, are they delivering quality to some extent that our human, again I bring the human element because that’s what is essentially evaluating the effectiveness of these AI agents and they’re almost thinking, oh, you’re my partner, maybe a digital partner versus a physical or a SaaS partner and are you going to behave consistently? Are you going to give me the quality if you’re going to go out of bounds where I can stop you and at some point we get comfortable and we keep doling out projects to our partners because we have kind of established that trust, this word trust is what we have to think about is on like a human to human discussions. The trust could be established because they will not totally do something crazy but here, in case of agentic AI, something could go totally out of whack. Now that is an assumption that I’m making, but is that truly an assumption? Is it paranoia or is it a reality that you see something working on a regular basis properly and suddenly it goes out of control? Have you seen such behavior being demonstrated by somethings which even offered by an outside organization or something your team would have developed and it still behaves erratically at times.

Aarti Singh [00:11:36]:
So I have not seen something which has been consistent to go totally out of whack. What I have seen is that something just not performing, like it’s able to perform with a given data set, but the moment you introduce new elements, it’s not able to react. So when you actually test an agent, you will be injecting different situations, etc and trying to see how you react. So in an enterprise setting, you want to be very sure before you go into production. So whatever we’ve seen, we’ve seen it in the development phase and we try and get it as right as we can before we move into production. I’ll just give you like a simple example, so in like call centers, typically, so the first agent that say a customer talks to you would want like simple queries, etc for that agent to be able to handle some of that.

Aarti Singh [00:12:45]:
One big thing is the recognition of different dialects, emotion and so you will have an agent which gets better with time as far as understanding emotion, understanding dialect and reacting to it and what is the right level at which it needs human assistance. So when do I really transfer it? So something like that, now we have something which works very well, but to be able to get to that was a journey in itself to understand like all the nuances and then for that agent to have a conversation and now if you hear some of that, nobody will guess that you’re actually talking to an AI agent. Like it is just so real because you’re able to understand the pause and the way the agent talks to you, it’s with the pause, it is with an understanding, it is with empathy. So to bring out all of that was a journey.

Sanjog Aul [00:13:50]:
Go ahead, finish your thought, please.

Aarti Singh [00:13:54]:
So and then so this is an example of how you got like the agent to really be human, like to a degree now going totally out of whack, at least in the enterprise setting. I have not seen that in production, touchwood, but in development, definitely, what we thought something would do, did not and we had to spend considerable time to get it to work.

Sanjog Aul [00:14:21]:
So do you see any problem areas or is this a truly a problem to deal with agentic AI, or is it more about a set of projects which you and I both have been technology. So we know as you build anything, you got those variabilities, you got those cases or use cases which it will not perform very well and then you tinker with it, you eventually bring it to a stable point and then you roll it out in production. So we’ve done with all other technologies here also we could do but is there some other unique elements to this which would defy whether you talk about agile methodologies or SDLC or whatever else that you’ve been using from a way of you rolling something out that it will still keep you up at night?

Aarti Singh [00:15:05]:
I think what would keep me awake at night is when we are using AI, especially in security operations, there’s going to be more of that. We have to really be vigilant around that space because security will keep anybody up at night. So if it is related to security, if it is related to safety, like if you talk of cars or safety of cars or aerospace, if it affects the safety, medical, those are the areas where it can really have a very high impact on a human life and I would also say that where it could have an impact is there’s always, I would say, a conflict between speed and risk. So sometimes you want to go very fast, so you may not get it to work or spend enough time in development and testing because it does take time and you may fast track the whole production of some of these agents. So it’s going to be about the conflict between speed and risk because the faster you want to go to, you’re inherently increasing the risk as well. So it would, I would say matter in areas where you are actually deploying it, where it is, where ethics, where it changes the, I would say the benefits or the rights of customers, of employees, decisions related to that

Aarti Singh [00:16:46]:
one would definitely have to be careful. So to say, will it not affect anybody, I would not agree to that because, when you go, there’ll be so many use cases and there will be times where, mistakes will happen but I think the important thing would be that what have you learned from it? How have you reacted when something went wrong? Are you transparent about it and are you accountable about it? So it cannot be that maybe as an enterprise or whoever owns the agent, I don’t think we are in a position to blame a model. So we have to be accountable for what is out there and we will have to ensure at least we have a proper governance mechanism around that.

Sanjog Aul [00:17:36]:
So see, when SaaS came out and then later AI applications came out, they were directly having impact and or benefits for the business leadership and business community and they said, okay, IT can be involved maybe at that point of looking and evaluating how it will integrate well with the stack that you have but within the organization but besides that, they said, IT could be more of a damper than an accelerator to adopting a SaaS application or AI application and when they were straightforward, then they had their own reasons to say so but when it comes to agentic AI, where there is a, not an unpredictability, but there is an unknown element that you’re grappling with, do you think IT should take ownership of what’s set up as governance and if something goes wrong, then they are the ones that is a throat to choke.

Aarti Singh [00:18:40]:
I will say this is one case. Well, accountability typically is with one person or one department but I would say in the world of agentic there are three parties that need to come together and work as one. One would be the product owner or the business owner, somebody who understands where it’s implemented, how it works, etc, in the business process itself, I think that is key. The second is an AI operations team. Whether that is IT or there’s a separate AI operations team. I’ve seen both flavors.

Aarti Singh [00:19:16]:
Sometimes IT is the operations team and sometimes they have a team that which does AI operations. So depending on who the operations the technical operations team does, that’s the second and the third, I would say it is the risk and compliance office because AI issues always relate to risks associated with trust, with fairness and could be some financial obligation as well. So I think all these three bodies really need to come together, when agents are released and when, you govern all the agents that you have created in an ecosystem and you have a regular mechanism in which you actually monitor all the agents, the drift, how they are reacting, how they are functioning and things like that. So the more you have a more structured approach to it, I think you’ll be more in control in how you manage it but yes, at the time of an incident all three parties have to come together and will it slow down if it’s with IT? I would say to be successful AI agents or even AI models, to be successful it’s always outcome driven. So the business has to have an equal stake compared to the AI or the IT.

Sanjog Aul [00:20:48]:
So I agree with you that the business needs to own it and at the same time there were so many other aspects to business using technology and legacy also that we would start, we will have a very close monitoring and or ownership established for IT to be there and that is where they said it’s a command and control structure. Then there were some organizational design changes that were made where IT was embedded into that organization and the central office of the CIO was moved to more of a program office to say okay, these guys will have an oversight. Do you think among other things like we have done in the past, would agentic AI would reach a point where business will take total ownership and IT will take more of a support role to help, maybe do investigation, et cetera, versus always be there because frankly IT has a lot to do. I’m not saying that they’re trying to shed their responsibilities, but do you think we could work towards an environment where we totally could achieve the speed when we have seen enough proof that this particular agent or this environment is working predictably enough that IT doesn’t need to babysit it.

Aarti Singh [00:22:05]:
So I would say AI agents, the AI world needs a focus now, as you rightly said, which I agree, is that IT has so many things to do. Now, it needs focus, it needs attention. Now whether should it be a central organization? I think when you’re starting off it is good to have a central AI organization. Where does this fit? It should definitely fit in the business. Sometimes it also fits with IT but the important part is the focus that, that team is only doing AI or AI agents, that is key and depending on the construct of the business, it could sit on both faces.

Aarti Singh [00:22:48]:
Now, as far as when an incident occurs, who is the first team that really understands or has the ability to troubleshoot? It would be the AI technology team/IT team, whoever is that group that is supporting it technically. So that would be the first group that will help or will be able to actually understand what went wrong and try and remediate the situation as far as business impact, etc. Then these other groups will come in that I talked about in the beginning but when that incident happens, it’s going to be the technology team that will have to assess as to what went wrong. So for example, they’ll have to look at system logs, they’ll have to look at the guardrails that were set. They’ll have to see what was in scope for the agent, what is out of scope and how did it react, what was the data that it was dealing with, et cetera, did it try to improvise, as I said, on its own or not? So I think all that technical details will have to be done by the technical team.

Sanjog Aul [00:23:53]:
Now when you talk about say agentic AI world, and we actually did a discussion on this on our forum about comparing it to the human anatomy where our own metabolic system kind of simulates or it mimics how AI agents work because it makes its own decision, et cetera and that could almost inspire the agentic folks who are trying to make this work but at the same time when they look at it, they don’t save any of a complete collapses. When you start looking at your gut and you go to the doctor, you look at some symptoms that this thing is happening and the doctor will say if you got a pain in the left side of your tummy, then that’s where you should start thinking, doing something extra to see if something else is wrong and that’s a doctor doing that because they have somehow over the years figured out that what all it could potentially lead to, if I were to compare that to where we are in the agentic AI world, what level of collapse or symptoms would indicate immediate subsequent diagnosis, which we are unable to do today because we don’t have enough data. So do we wait for the collapse for us to know that it has failed and it’s too late, or have we kind of started working or have come close to that diagnostic process where we can prevent an AI agent from failing, especially when we don’t know how it works?

Aarti Singh [00:25:27]:
So I can give you some examples, so in the beginning I did say that, when we put things in production, we actually test it, there’s enough data, etc, situations, days that it’s dealt with those kind of scenarios and then we look at what is the outcome. So for example, if you have an AI agent that handles calls in the call center, okay, so how do you really understand whether it’s working or not or how do you start getting symptoms? So, if they are simple queries, and the expectation is that 100% of those, the agent should be able to answer, so that is say one KPI. So you have some KPIs that you have set. Now if that happens 100% of the time, we know like that agent is working because you will get user feedback, et cetera, in the end and you will watch that.

Aarti Singh [00:26:30]:
So like how when you talk to a customer service agent, you get like a feedback. So that is one KPI that you will measure. The second is that when you’re not, can I help you in other ways, if your query is not answered and say a customer presses that, that means it’s getting, the call is getting directed to a human being. So then you’ll start looking at how many times that has happened and what were the queries because of which the call had to be passed on to a human agent. So that is the other one. Then you’ll also look at system logs, you look at the conversations, the transcripts to see that how the agent was, the AI agent was reacting, what was the conversation, etc. So you will have KPI setup, you will have a governance mechanism around that to keep watching it because see new data sets.

Aarti Singh [00:27:39]:
You’ll also want to improve the agent. So it’s going to be a continuous process. So you’ll start with maybe simple queries. So if I were to give you an analogy of maybe a bank, you want to know what is your balance, you want to know your credit card payment date, you may want to know if you’ve taken a loan, your EMI schedule, now, for example, you want to know that I want to foreclose my loan, just as an example and what are the foreclosure charges? So regular answers you can get. Now if the customer says that, can you give me a rebate on the foreclosure charge, or if I pay up slightly early, is there any benefit that you can give me? So now those, the agent, for example, has not been set up to answer. So the moment it goes into that land, did it pass it to the human agent or not or did the agent try to answer all these questions on its own, which are out of scope topics.

Aarti Singh [00:28:36]:
So you will try and watch out for some of these kind of symptoms and then you will know that what is the level of confidence that this agent is working as expected or not? So you will get some of that and even, the dialect example that I gave you when you were talking in the beginning about, how real the agent starts sounding. So how and there were cases where it would get passed on. So you knew that in certain cases it worked, but in certain cases it did not and when a customer is not very happy, it didn’t sound very human like so you will realize some of those things through some of these parameters that you have set and then you’ll be able to act on it

Aarti Singh [00:29:22]:
and then, gradually those numbers will keep improving. So it has to be KPI driven. You have to set KPIs for some of the agents and watch through those.

Sanjog Aul [00:29:32]:
And this part of watching, so so far the watching has been done by humans. You’re getting something, but then at the speed at which this works, and there could be a cascading effect if something goes south. Do you think there has been an effort made to build that AI agentic watchman or an AI watchman or watch woman, whatever, just one way to figure out what is happening, but catch it timely and that way human is moved out of the loop and that’s where you do not have human getting in the way of preventing risk and preventing losses timely is that getting incorporated where the patient itself, which is the AI agent itself is reporting back to you versus a human looking at it that you’re getting out of the these guardrails or something is going, is misbehaving.

Aarti Singh [00:30:35]:
So that is a conversation that we’ve started and so we are looking at agentic platform for governance where you kind of bring all your agents together so you’re able to govern them, you’re able to monitor them, you have visibility to all of that. So that journey has started. Are we done yet? Not yet, but definitely that is going to be an area that we will also be picking up.

Sanjog Aul [00:31:07]:
And so till the jury is out or people are working towards it, an agentic AI which was brought into being or into existence and into production is still being so called policed and monitored by humans so far and because of that human intervention, it’s not that it is slowing down the working, but it definitely keeps the risk much higher and we are living with that risk that things if they go out of hand, they’ll go out pretty quickly and human may not be able to diagnose and report on it and stop it.

Aarti Singh [00:31:43]:
I would say the way we’ve at least prevented some of that from happening is through the guardrails that we have set up. So depending on like the financial risk associated with an agent or how it could go south now if the guardrail itself crashes, there could be an issue, but at least as of now, the guardrails are taking care of, where we feel there could be a potential risk of something going wrong.

Sanjog Aul [00:32:18]:
How could we prevent finger pointing? Why because you got the risk team, you got the business team and the IT team, they’re all trying to of course, with the right intention working on this and when things go south, humans behave in interesting ways and departments could try to cover their basis and I’m just getting very real to how an organization functions. Is there a very well established mechanism to prevent such finger pointing and or friction creating scenarios in this world where every one of them are trying to babysit a piece of software which nobody knows how it’s going to behave and of course nobody would want to take ownership without them knowing that yes, truly it was a fault from my department. How do you prevent that and I’m coming into an area which is more not a digital problem, it’s like an analog problem, it’s a culture problem, it’s an organizational design problem, it’s an immunity problem.

Sanjog Aul [00:33:30]:
It’s a level of safety net that we offer to people who are willing to become custodians of these agents to say you will not be fired because an agent misfired.

Aarti Singh [00:33:42]:
So I would start at a macro that it is going to start with the culture that you’re driving in an organization. If you’re going to provide psychological safety to people, people will own up their mistakes and I think the other way also to look at it is like if something were to go wrong, your organization suffers. So it doesn’t matter whether it came from team A, B, C but it impacted your bottom line, it impacted your ebitda, it impacted your P&L. So I think once, so it is all about the culture that you’re driving. So if everybody starts thinking about the organization first and then about your team and about yourself, I think to some degree the comfort would be there

Aarti Singh [00:34:35]:
and to enable that kind of culture, you really need to drive psychological safety. A culture of where people speak up, they don’t hesitate and I think the important point also to be made is that what was the learning? So you’ve got to celebrate failure also and what was the learning that came out from that failure so that others around could also learn from it. So if you have that kind of culture where it is organization first and second, you can make a mistake but the important thing is what have you learned and what has the organization learned from that? So if you create that kind of culture, defending, blaming, etc, I will not say we’ll cease, but will reduce and if that same message comes like from the leadership is driven by leaders from different groups, I think that is a kind of culture that percolates down and that would really be helpful. Where and why I talk about that is because I’ve seen it up close and personal, not just for AI agents, but when things go wrong, if there’s psychological safety and when people think of the organization like, I’ll give you an example, like something goes wrong, you have an assembly line, okay, it could be an IT issue.

Aarti Singh [00:35:54]:
It could be, then yes, maybe IT is to be blamed but finally at the end of it, you did not meet your numbers of production. So it is a joint problem to solve for. So like the plant owner will also have accountability finally, it’s going to be about how he or she ensures that the numbers are produced. So it’ll come right from there and then all the parties have to play the game so that you’re able to meet the number and I think it’s going to be finally about the culture that we have in an organization

Aarti Singh [00:36:29]:
and I have seen very positive culture. So therefore I’m very confident that if you drive the right culture, it’s going to be about solving first, it’s going to be about getting right and it’s going to be about helping and collaborating with each other versus trying to blame somebody.

Sanjog Aul [00:36:48]:
So again, being in IT, you and I both have seen enough and it should not phase us ideally but then at the same time there is a lot of hype and to some extent there is a lot of disruption that is inherent in the kind of technology we are dealing with today. So the psyche of the people in IT, in business, in risk, in your view, should that be morphed to the next level or should be a different way of thinking for us to tackle this or is this business as usual and just don’t get paranoid, don’t get psyched up and make more mistakes than you’re supposed to. Is that the mindset they should embrace?

Aarti Singh [00:37:35]:
No, I think the mindset has to be about being cautious, about being careful, about dipping into the right resources to ensure that you have a right solution and should things go wrong, have the courage to ask for help and have the courage to speak up and to raise the issue. Is it business as usual, absolutely not. It is a big change. It is going to get bigger. So it is going to be very important to understand the technology, to upscale and cross skill yourself and the business around so that they understand the implications and then how do you gear up to use it positively and make a change and I would say look at the future where there could be so much more that you could do with AI agents versus being afraid of it.

Sanjog Aul [00:38:37]:
So as you see things that are happening today and how it seems to be morphing, what do you think your three constituents that you mentioned, IT, the business and the risk team should be getting ready for?

Aarti Singh [00:38:57]:
I think what we need to get ready for is that there’ll be so many choices and emerging choices. So I think for the business and the IT teams would be that there’ll be no one right answer from a technology standpoint. So how have you really geared up to build enough choice that you’re able to leverage to build the right solution, so that is one. How are you able to scale your solutions and how are you able to manage the risk? I think that is going to be key as well. The learning is going to be tremendous because the new, I would say platforms, new ways of working, it’s all getting enhanced. So you will have to keep learning and for the risk team, I think it’s very important for them also to understand, I think maybe all the areas that have gone wrong in enterprises and to keep validating, like how you have a maker checker, so not just rely on the technology teams, but also how do you really get up to see what went wrong, what could potentially go wrong and challenge the technology and AI teams as well, to ensure that, they have taken care

Aarti Singh [00:40:24]:
all the guardrails are in place, the security aspects. I think security is going to be a big one for AI agents because they are human beings finally in the way they operate. So, to ensure all the security also is taken care so that the risk is minimized. So I think for the risk team also, it’s going to be very important that they learn and understand when things go into production what would be the appetite for the risk that the organization can carry.

Sanjog Aul [00:40:55]:
Thank you so much, Aarti, for taking the time. Very interesting discussion and very timely that AI agents are showing a lot of promise, as you rightly said and there is a lot more to be done but we got to get ready with our leadership style, with our communication, with our change management. Mainly it is becoming more of a human transformation which is going to support the AI agent and of course technology will evolve. We will come to know what it is and then I like the way when you said that if we give people the psychological safety at all levels, no matter who is involved, they will own up

Sanjog Aul [00:41:34]:
and that’s where people will be volunteering to sign the incident report because they feel they are in it together and that’s what’s going to keep us going in the right direction but thank you so much again.

Aarti Singh [00:41:46]:
Thank you. It was a pleasure being here.

Sanjog Aul [00:41:50]:
Absolutely and for our audience, please find us on social media and other channels where our content exists. This is Sanjog Aul, your host, signing off. Till next time, take care and god bless.

Aarti Singh [00:42:04]:
Thank you.