[00:00] Welcome everyone. I am Niharika with me. I have Sanjay. We both are cyber security and AI educators and we have a course on AI adoption which is secure and compliant on May 1.
[00:12] With us, we have Ashish Jagarwal. He's the former CEO of Indigo Airlines which is one of the most profitable airlines across the globe and he's also advising various companies. Welcome Ashish.
[00:27] Thank you. Thank you. It's the pleasure to be here.
[00:31] Thank you. And then one thing I would like to add here that we are fortunate that we're starting a new course on May 1 on how the project management is going to be changing
[00:40] in the AI realm and Ashish is joining hands with us so that's a good news we just wanted to share.
[00:48] Thank you. I could. There was a question I think they wanted to know where the speakers are joining from so we are all in free month California.
[00:57] Thanks Ashish. Thank you. Yes, we'll start with the question and which is we would like to know what's your level of expertise with agentic AI. So let us know on a scale of 1 to 5. 1 is beginner and 5 is expert.
[01:13] Please so while you are putting in your answers, we'll kind of keep going.
[01:18] So what's the question which is facing enterprises today? Like people companies are building AI. They are investing billions in
[01:26] pilots and proofs of concepts. But very few are building agentic AI and there are many pilots which are really not scaling and companies are also getting confused between AI project and agentic AI project.
[01:42] And very few really know that if they are ready because agentic AI it represents a fundamental shift. It involves new capabilities and a new mindset all together and that is what we are going to discuss today.
[01:57] So I'll start with a basic fundamental question is and this question is to use Sanjay that what is the difference between traditional and AI projects and what's really changed.
[02:10] Oh, actually this is a very good question. A lot of people are asking these questions to us. So traditional project that you have been working on so far is your requirement, your design,
[02:20] then development, QA and then cycle of your buck fixing and then it's in deployment and then you have
[02:29] your good until you have some new features to be added. So then you have some DevOps, CICD, 5 minus all those things.
[02:37] AI projects is totally different. It is everything in traditional projects management but few more things are additional.
[02:47] Which are your data that you train on your model. So there's something called model like
[02:55] which at GPD is one of the models. So if that model is not trained on the right data then it
[03:00] cannot give you the results, for example, if you have trained it to get a
[03:07] recruitment in HR agency and they are trained only for getting only people from USA.
[03:14] But now somebody is applying from sick Canada and your model is not trained to get somebody
[03:20] from Canada.
[03:21] But even if that employee is really good with all the ticks, he will not be selected.
[03:27] So there is a bias.
[03:29] So these are some differences that we have to now take care of it as part of the project
[03:34] management.
[03:35] I don't want to go very long about that.
[03:36] But there are some bias and there are some data privacy beaches.
[03:40] So for example, if you are developing a project in AI in the current realm, it developer,
[03:48] if he goes to check how is my code really written.
[03:51] It is really nicely, it is really following the rules.
[03:54] Then it will go to chat GPT and then find out.
[03:57] So what happens?
[03:58] There have been cases like this.
[04:00] So now that code proprietary, code now goes public and that can be a detrimental thing
[04:09] for your company.
[04:10] So these are the few examples of how the AI projects are different from traditional
[04:15] projects.
[04:16] Here we come.
[04:17] So thanks, Sanjay.
[04:18] So again, I will just recap that if you have been managing traditional projects, the first thing
[04:23] is to understand how to manage AI projects and how it is different from traditional
[04:27] projects, keeping in mind things with Sanjay said.
[04:31] So now from traditional to AI.
[04:34] Now let's move from AI projects to agentic AI projects.
[04:38] And this question is for you, are she issues that how these two types of projects
[04:42] are different?
[04:43] Yeah, and that's a big shift in Harka.
[04:47] I think in a very short period of time, companies have moved from AI projects to agentic
[04:52] AI projects.
[04:53] People currently not even differentiate what those are and feel that you know, agentic
[04:59] AI is also a sub-adventive AI.
[05:01] But there's a radical difference in terms of how agentic AI projects need to be thought
[05:06] up and manage.
[05:08] Some of the things that are really important are a traditional AI project, which is something
[05:13] you know, now done this for at least three or four years is something where you would
[05:18] predict an outcome or you would classify an outcome or you would recommend something.
[05:23] So and we're exposed to this on almost every day basis.
[05:26] If you look at the movie recommendations that we're seeing on Netflix or you know, in stores
[05:32] in warehouse is for upcoming Thanksgiving and Christmas day holiday, their demand forecasting
[05:37] happens to be expected to sell.
[05:39] They're all using AI models, which is the traditional AI.
[05:42] So you know, this is where the technology is helping you understand and predict what the
[05:49] forecast is going to look like and then there's a person, a human in the loop who
[05:53] actually goes and looks at that prediction and forecast and then start sticking actions
[05:57] based on it.
[05:58] Since we're coming close to the...
[06:00] Thanksgiving, let's just use that as an example.
[06:02] So, the ACs are in your far details, stores are people sitting in the back office who are
[06:08] forecasting how much of sale do we expect this Thanksgiving or Black Friday.
[06:12] And then there's a bunch of teams, whether it's the inventory planners or the demand planners
[06:17] or the pricing teams who take actions on those forecasts and start planning how they
[06:22] are going to maximize their sales and earnings for Thanksgiving.
[06:26] Compare that to a genty AI.
[06:30] For a genty AI solution, you don't predict forecast and those kinds of things instead,
[06:34] you set a goal for technology.
[06:37] And once you set that goal, it reasons on its own, it creates a plan.
[06:41] And it has a set of tools that it uses to execute those tasks.
[06:46] So it's almost like a shift to an autonomous team member and handing over the task to the
[06:51] team member instead of actually just defining a series of steps on how that work needs
[06:57] to be done.
[06:58] So in that same example, an agentic AI tool would actually sequence and like the goal
[07:05] is if I want to raise my revenue or sale to a certain number.
[07:11] And if I tell the tool that this is my past demand forecast, this is how you can create
[07:17] an outreach message to potential customers who visited us in the past.
[07:22] Then the agentic AI model can actually set a path, reason and decide, you know, what
[07:29] are the right things to do to maximize sales, what to tools, do I need to achieve that
[07:33] goal and that part is all done autonomously.
[07:37] So that is a very radical shift in terms of how a genty AI is coming up.
[07:42] The sequence of flow is we were to just look at it at a top level for an AI project
[07:47] would be data, a model and an output.
[07:50] But it's not as simple as that for a genty AI because you would say the flow for an
[07:55] agentic AI application would be set a goal.
[07:58] Identify the reasoning path, create a plan, identify the tools that are being used to deliver
[08:05] that plan and execute that plan.
[08:08] So it's very different in terms of how a genty AI projects need to be thought of,
[08:12] especially people who are managing these projects.
[08:15] It's important for them to understand these components and building blocks because, you
[08:19] know, I've been there, I've done this with my teams.
[08:24] And I've seen that there are so many moving parts in components to an agentic AI project,
[08:29] whether it's data team, security team, compliance team, monitoring.
[08:34] When you start to understand what goes on into executing and building an genty AI project,
[08:40] you're able to better collaborate with all of these teams and understand what are they looking
[08:45] for, how you as a project manager can provide those inputs to them and drive your project
[08:50] to success.
[08:51] So, I know it's a long answer, but really there's a lot that has changed with genty
[08:57] AI and hence it's really becoming a part and part.
[09:00] to understand what the shifts are. This is just a recap of what we just discussed and we can
[09:07] also take a moment of pause and see if there are any questions or comments that anybody has.
[09:13] But essentially, you know, in a traditionally I mode you prompted response,
[09:17] the third GPD is also very similar to that. But in a genetic area you're setting a goal at plans
[09:23] and attacks on it on synotonms to execution become its problem. But since execution is
[09:29] the agent's problem, then governance and monitoring becomes your problem. Because you want to
[09:34] make sure that whatever execution path the agent is taking is correct, is within the guard
[09:42] rails or that you want to establish and it's not going rogue. So that's where there's a big shift
[09:49] between traditionally I and a genetic area. And so these new aspects of implementation that we
[09:56] all need to understand. And this is a paradigm shift. So we have listed here, we have summarized it
[10:03] on the slide and we're going to send the slide to everyone, the deck to everyone.
[10:10] So we said that it's definitely a paradigm shift and an agentic AI project has to be managed
[10:16] with a different mindset altogether. So my next question is, why agentic AI changes everything?
[10:23] And we're talking about fundamental shift. But what is that fundamental shift?
[10:34] This perashish? Yeah. So when you're thinking of shifts and we just briefly spoke about it,
[10:40] there are a lot of viewers that you have to think about. Infrastructure is one, the workflow
[10:47] redesign, the governance aspects around agentic AI and the risk profile shape. So I'll talk about
[10:53] infrastructure transformation in workflow and then we cover governance and risk, such as an
[10:59] expert on those two topics. So he'll share his thoughts. So from an infrastructure perspective
[11:04] and I'm going to try and hold myself back from getting too technical because but I do also want to
[11:10] expose all of you to the fact that you know it's a very different infrastructure that you have to
[11:14] think about. So even though you're running a program that is built around agentic AI solutions,
[11:21] you have to think from the perspective, the infrastructure teams, what are they going through?
[11:25] How is their work changing? And a better understanding of what they're dealing with that also
[11:30] help you manage the project better. So from an infrastructure transformation perspective, you
[11:35] know think about old and days when there was a monolithic application. Let's just take an
[11:40] example of an ERP, we were in an age where companies were implementing ERP solutions and ERP
[11:47] was supposed to do everything within the organization, whether it's HR, finance, operations,
[11:53] marketing and the goal was that we get reintegrated applications that work end to end for the entire end.
[12:00] price. To the point where now we are in an
[12:04] agente care world where the infrastructure is going to be very different,
[12:07] think about data pipeline because this whole technology is running on data,
[12:11] it learns through data. So the quality of data that you're handling,
[12:15] the the velocity of data that you're moving, the volume of data,
[12:20] the checks and balances around data are becoming even more important.
[12:25] Let's take an example with respect to data. If you train the model with
[12:29] pictures of CEOs that are all men, and then if you ask the model to predict who the next
[12:36] CEO is going to be, the model is going to come up with a male figure. And that is a bias
[12:40] that has trapped into the model. Companies are even taking this to the extent where they're saying
[12:46] that the team members who are actually contributing to sanitizing and curing this data,
[12:51] need to be a diversity. It doesn't have to be a group of all men or all the men or equal from
[12:58] the same ethnicity because you know just because of our background and where we come from,
[13:03] there is the possibility that we might introduce bias in the data that we generate and the model
[13:08] that we create. So data becomes a very critical component of the infrastructure transformation.
[13:14] The other thing that is changing around infrastructure is it's very dynamic. In the olden days,
[13:21] you would predict a certain capacity. And you would say I need four servers, 100 terabytes of
[13:27] storage, certain model network capacity to run this application. And this is what I'm going to
[13:33] provision. And then came cloud where you had more dynamic scalability where you said, okay,
[13:38] on Thanksgiving Christmas day, I'm going to get more hits. So I need an infrastructure that
[13:42] should be able to dynamically scale up and down depending on the volumes that I see.
[13:47] Now we come from cloud to agenda. I where I'm going to say I need an agent to do this for me.
[13:54] The agent's going to be autonomous. I'm going to set guardrails. But this is the task that I
[13:59] want the agent to perform. And the agent starts to execute those tasks. It scales up. It spins off
[14:06] more agencies. So I'm more agencies to do this kind of work or I need to sub delegate this task to
[14:11] another sub agent. And I need this tool to send an email to person next or I need an API to talk
[14:19] to my inventory team to understand, you know, what is the amount of inventory available in the
[14:24] store? Are we going to be able to ship this inventory to the customer? So now all of these things
[14:29] are happening dynamically. And the agents is making decisions using all of these tools. So the kind of
[14:34] infrastructure that you need needs one real good data quality, type security guidelines and guardrails
[14:44] API and decomposable architecture where you've got lots of services and tools available that you can
[14:50] use for building your applications. And then you know orchestration mechanism because this is
[14:58] done one agent that will be able to do every
[15:00] things like a team that you're creating. So let's say you have 25 different agents that are
[15:04] managing your store or a warehouse operation. Now you need an orchestration layer to orchestrate
[15:10] these agents. Okay, I'm going to do this and I'm going to add it off to my agent B and then
[15:14] agent B is going to do something and then delegate it to agency. So for that kind of of an operation
[15:20] to run, you need a very dynamicity scaling infrastructure. So I just want to leave you with a
[15:26] thought that it's not the same infrastructure that we use to have for a API application in
[15:31] an inflow for cloud application. But we're now come a long way to a completely dynamicity scalable
[15:37] infrastructure. And when you're building your agent decay project, these are some elements that you have to
[15:42] think about and talk about with your infrastructure teams. No who's going to give you the API.
[15:48] I understand whether tools are going to come from, understand how you're going to manage capacity of
[15:53] the infrastructure, what are you going to do when the volume sets hard, what kind of guard
[15:58] are you going to need to put around your agent so that you know and the agents are not going to
[16:03] give you the same answer every time. And that is a big chain that has happened. In the
[16:08] olden days technology, Acculator being the simplest one, let's say you said what is 2 plus 2,
[16:14] it's always going to give you an answer of 4. Now with agent decay, you ask a question to agent,
[16:21] it'll give you an answer you ask it again, it'll give you a very similar answer but a different answer.
[16:26] So now you need to understand if that different answer is still within the realms of your guard
[16:31] where you want it to be. So all of those nuances become part of how you're going to define
[16:36] your infrastructure and those are some elements, you know, I want to leave those thoughts behind
[16:40] with you so that when you're planning your project, you're able to think through those elements.
[16:46] And then from a workflow redesign perspective, I'm going to take it from an with an example.
[16:51] I think that example will speak for a thousand words. So let's take a traditional AI project
[16:57] where I am a leader of a team where I'm very concerned about customers who leave us as an
[17:04] organization or like the and it's traditionally called children, right? So essentially I'm seeing
[17:10] what percentage of my customers leave us and then we want to understand what other reasons
[17:14] what are leaving us and you know what can we do to prevent that every company goes through this.
[17:20] It's a big concern for everyone. You want to keep your chair and as minimum as possible.
[17:24] So a traditional AI project will be a good idea to science team and say I want to build any
[17:29] AI model and I want to predict your own. So I want to understand that in historically I've had 4%
[17:35] churn but with everything that has changed, we've come up with new products, we are getting new
[17:40] feedback from customers. What does my new churn is expected to be? Do we need to be concerned about
[17:46] it? Like what can we change? So my team is now going to build any AI model which will come up with
[17:53] a forecast and will go to the draft saying that this is the trend of your churn rate. This is what
[17:58] we can expect to see if we're getting more.
[18:00] negative sentiments on social media, some may be the term will go higher.
[18:03] And it starts to forecast and predict to end and give me the data.
[18:06] And now it becomes my responsibility to analyze it.
[18:10] Understand what can I do with it and then update my CRM and other systems and then guide the
[18:15] teams on how to use that data and take actions out of it.
[18:18] Let's reverse this to the agenda AI workflow design.
[18:23] And that is by one of the recommendations that we have is,
[18:26] is NTEK AI solution should not be built as a bold on solutions to your existing
[18:30] solutions. Let's say if you have an application running,
[18:33] turn models, don't just overlay it with some agents.
[18:37] Instead, we think about what the right workflow and design for an NTEK AI
[18:41] architecture should be. And so let's we're now going to redesign this whole workflow
[18:46] in an NTEK model. And the way that would work is I'll create an agent.
[18:51] The agent is going to detect the term. So it is going to watch my data and it
[18:56] has the ability to find out when my turn is increasing.
[19:00] So it will congratulate me when the turns are shown is going down or it will
[19:05] then mean when it sees, you know that the numbers are going higher and
[19:08] as an organization we're failing somewhere. Then the next step it could do is it could
[19:13] draft an out reach message based on the turn numbers that it sees.
[19:17] And start sending emails to different teams, turning them, you know,
[19:21] this is an area that is an organization. We'd be mindful of
[19:25] it could even start personalizing offers and creating marketing offers to attract
[19:30] people to buy products and reduce the turn. Even saying emails to customers that
[19:35] we might have lost trying to understand, you know, what was the reason behind
[19:41] you not coming back and what could we do to bring you back? How can we do that service
[19:46] recovery, giving them a personalizing personalized offer. And based on
[19:50] response on those emails or from those customers, it can go back and update CRM, right?
[19:56] And then even monitor response in the CRM to see whether those updates that are coming in
[20:01] are they really helping reduce the turn or not. So think about this. This is the way this
[20:07] is operating is that you've not created a software product, but you've hired a team member.
[20:13] Right. The team member is doing this end to end workflow on its own. And then what you're doing
[20:19] is your orchestrating the workflow as a human, you want to make sure that the steps that it is
[20:24] taking are right. And then, you know, there are guardrails around the steps that are taking.
[20:29] So our responsibility becomes more around orchestration and governance and less around execution.
[20:37] The execution is handed off to the agent. So that's sort of the new way of
[20:42] rethinking about workflows within your organization. And that would also be the first step for
[20:48] you when you go back to your teams and start thinking about how where the opportunities for
[20:53] agent solutions are, especially in project management, you can always think about, you know,
[21:00] areas where you can create agents to ease your work. It should be less about creating status
[21:06] reports now, less about chasing people to get their tasks followed up. All of those things should
[21:11] ideally be done by agents. Your role we should be to martyr those agents in make sure that
[21:16] the your project is running on track and all of your agent tick team members are operating in a
[21:22] perfect harmony and orchestrating the word the way you want it to be done. So that's a little bit
[21:28] about the new way of thinking workflow design in an agent tick air in able organization.
[21:36] Thanks, Ashie. So it's a new perspective altogether, not a bolt on.
[21:40] Absolutely. Ashie. A bolt on, very radical, different way of thinking about workflows
[21:45] and thinking about agents as your team members, not as software solutions.
[21:51] Thanks. And I think equally important is governance and risk profile. So, Sanjay,
[21:56] do you want to take those two aspects? Yeah, before thank you Ashie, very nice answer.
[22:03] Very detailed also. So I have a question for everybody. Do you remember that you asked
[22:08] some question to chat GPT or any LLM and then you know it is wrong but it keeps saying no I'm
[22:15] correct and totally incorrect and non-sensical. How many people can relate here?
[22:21] Yeah. It's called hallucination. We call it hallucination. That means your
[22:29] chat GPT is definitely sure about something and which is totally incorrect.
[22:34] Isn't that a big risk? Yes, it is a big risk. Number one thing is that
[22:40] hallucination you don't even know if it is hallucination hallucinating. So there is a governance
[22:43] that comes in the picture. So I'm just talking about risks and governance interchangeably.
[22:48] Second thing is bias that Ashie has talked about that you say that create an image of
[22:56] SEO and then Sanly will say, a white 45 to 50 to 60 years old will be given to you. Why not
[23:06] why not a woman? So these are the bias that are the big risk for it and data privacy.
[23:15] Data privacy is also another thing because when you say agentic AI, agentic AI means you are
[23:21] deploying an employee. So AI agents are actually not only the workflow but they have their own
[23:28] brain also. So they can think and they keep learning from their past experience also.
[23:34] So that employee can make mistakes also. So there's a big risk about that. So it can end up
[23:40] giving you wrong information. So these are a few of the risks because we have left
[23:46] it only for months left. So these are the risks that we have to be cognizant about and then
[23:53] your governance comes then we have to be following some different rules there.
[24:00] If you have a different governance as to happen, your checklist have to be different,
[24:04] your employee, you have to follow certain compliances for example EU.
[24:10] Our ISO 4211, you have to follow their guidelines as to how to be compliant with the new norms.
[24:21] So, you have to only form its left eye thing.
[24:24] We can go over a little bit by, you know, 5-10 minutes.
[24:28] The bottom line is that you are not deploying a tool, you are deploying the team member with no employment contract.
[24:34] So, now I go to the next question, which is to you, Ashish, how do you ensure that we have readiness for agentic AI projects?
[24:42] Yeah, and I'm just going to be mindful of the time because I know I get look ahead of it and start talking too much.
[24:48] No, because all good information, we can definitely go over.
[24:52] Right, so we'll keep it concise and I know there's more detail course that is coming up around the stockpacks.
[24:59] So, if you have the group is interested, there's always an opportunity to dive into each of these areas, a lot more in detail.
[25:06] But there are three things to think about, is anticharrad, and as is, make sure your team is ready.
[25:11] Data is going to be extremely critical for the success of anticharrad project, and governance becomes very important as well.
[25:18] And my recommendation is not to think of governance as an after the fact.
[25:23] I don't know, but something that you have to take care as part of your design.
[25:27] We'll go very briefly into each of these aspects, just to give you a little flavor of what to think about.
[25:34] So, when you're thinking about your team, make sure that your team really understands beforehand what the agent is supposed to do.
[25:43] You also understand the failure modes of a time and system. So, think about it.
[25:47] Let's say you take a day off and your agent is working for you and running the task that you had asked for.
[25:52] How do you make sure that it doesn't do something that you didn't want it to do?
[25:56] So, those guardrails should already be built into your agent design.
[26:00] You've got to think of, let's say you are a brand that is completely doesn't want any red color shirts, right?
[26:08] So, you want to make sure that your agent is not recommending red color shirts to its customers.
[26:12] And those guardrails have to be built into the agent definition.
[26:15] And let's say who owns when your agent screws up?
[26:19] So, those are kind of things that you never thought in your old traditional software because
[26:23] you was built on a set of rules and the program exactly followed those rules.
[26:28] But since the reasoning and planning is done dynamically, you have to give it a set of guardrails that whatever rules you're creating as an agent don't go outside these boundaries.
[26:38] So, that's on the team readiness.
[26:41] With respect to data, I think we covered this a lot, but the one takeaway that I would leave with the team is,
[26:47] and this is also related to governance.
[26:50] So, of course, quality becomes very important.
[26:52] The agent should know what is the level of access that it can have.
[26:57] Think of it like a person.
[26:58] Do you want the agent to have a person?
[27:00] access to everything or are there limits, right? So those limits have to be defined that the
[27:04] agent can go and look at employee data, but not SSM, no addresses. So, you know those
[27:10] guardrails have to be built and then you also have to have data lineage. So, let's say if the agent
[27:17] recommends a certain type of a product to me as a customer and if I'm guessing why did this
[27:23] agent recommend this product to me, there should be a way all the way to go back from the
[27:28] recommendation to the source of data to tell me how exactly the the agent came of the
[27:35] this recommendation, right? And this is a this is a governance requirement, it's also called
[27:41] explainability. So companies are expected to keep a trail of how what they reset, what rules,
[27:50] how did the agent eventually come to that conclusion because somebody might ask you.
[27:54] Right? So, data readiness becomes an important aspect and the last one is governance,
[28:00] but very very important. We've already talked about explainability, security control versus
[28:06] autonomy in the past of resolutions where they're built around controls, you would have your
[28:11] network policy is patching rules all of those things. Here you're monitoring an autonomous
[28:17] agent, it's like a child that you're telling the set of rules to, it's likely with tell your
[28:22] son and daughter don't do this, don't do that. This is okay, that is fine and those are the kind of
[28:27] rules that you have to tell your agents every time you find that agent is found a new way to
[28:31] break your rules, you go back and add some more insights to constrain it. And this is also not about
[28:39] audit trails in the in the past, like you would have a quarterly audit or an annual audit and you
[28:44] would see how the systems are behaving, but there in a normally it should be do something
[28:49] incorrect that but here you have to monitor it on real time basis because you cannot have
[28:55] your agent giving incorrect advice or incorrect action on a real time basis. So, the monitoring
[29:01] becomes a lot more real time and so, today talk about some of those amplified risks,
[29:07] with the exists traditional software, but by a certain nation, all of those unintended
[29:12] consequences need to be managed. So, make sure that you have all of these elements on your
[29:17] project plan and you're building your agentic acidution and then you're working the teams that
[29:23] can bring in solutions and mitigation for all of these aspects.
[29:28] Thanks Ashish, so we need to have three pillars which is one as data, one as team,
[29:33] second is data and third is governance. And as we said that its agentic air is not about a new
[29:40] tech coming in, it requires one cross functional alignment, second here readiness is not a one-time
[29:48] checklist, it's ever evolving. And third, the winners are not the one which are fastest
[29:53] movers, these are the people, these are the companies which are who are moving cautiously
[29:58] making sure that
[30:00] everyone all the stakeholders are on board.
[30:03] So your new language should be
[30:06] think of teammates, not tools,
[30:09] think of outcomes, not in sites, in sites for traditional AI.
[30:13] Think of workflow autonomy, not automating a task.
[30:17] Think of adaptive system because these are not linear.
[30:20] They don't do things one at a time, one by one in a sequential way,
[30:23] but they adapt to data and situations.
[30:26] And think of software systems that are not executing rules,
[30:29] but generating rules.
[30:31] And you want to make sure that guardrails around the rules.
[30:34] And these systems are not to answer questions.
[30:36] They are to achieve goals.
[30:38] And that is where you know, a gentry AI comes in.
[30:41] So you know, very fine boundaries,
[30:44] lot of people think of LGBT as an agent,
[30:47] excretion.
[30:48] But I think this session would have helped you better
[30:51] understand what a gentry AI is,
[30:53] how to differentiate that from traditionally I.
[30:55] What are the concepts and things to think about
[30:58] when setting up an agentic AI project?
[31:01] Thank you, Ashij.
[31:02] It was very insightful.
[31:04] And before we take any questions,
[31:06] we would like to let us let you know that we have a course
[31:09] coming up on May 1, which is, which teaches you how do you lead AI,
[31:13] agentic AI projects, which legal approve,
[31:17] sensitivity trust.
[31:18] Because we don't want to start with a project,
[31:20] which is,
[31:21] which is, uh,
[31:22] mecha meets business requirements, businesses happy,
[31:24] but then ultimately when we go for legal and security approval,
[31:28] it's, it doesn't go further.