Life SciencesSeason 1 | Ep. 630 min
AstraZeneca’s Kabir Patel and Infocepts’ Shashank Garg on what agentic AI is and is not, where it applies, how to start and scale, and what may come next.
Kabir Patel
Global Head of IT, AstraZeneca
Shashank Garg
Co-Founder & CEO, Infocepts
Also on
Many vendors now say they offer agentic AI, but the term is used loosely. Kabir Patel, global head of IT at AstraZeneca, says some of what arrives as agentic is really an AI workflow with a user interface, and he offers a plain test for what the term should mean.
For enterprises the practical questions are where to start and how to scale. Patel and host Shashank Garg discuss which use cases are within reach, why data, trust, culture and sponsorship matter as much as the technology, and whether to buy, integrate or build.
The conversation covers a working definition of agentic AI, its building blocks, use cases from service operations to self-healing supply chains, the near-term path through back-office functions, how the vendor landscape may change, and what could come next.
Five Things Worth Your 41 Minutes
Kabir Patel describes agentic AI as a solution with enough agency to operate independently toward broad objectives: it assesses situations, chooses actions and learns from experience. His crisp version is a solution that is able to reason, act and learn. Shashank Garg adds that, compared with assistive AI, it is higher in complexity and autonomy and lower in human interaction. Both say chatbots, rule-based engines, automated trading bots and a plain API call to an LLM are not agentic, and Patel notes that some products presented as agentic are AI workflows with a user interface.
Patel lists memory and context, a reasoning engine, a foundation model and tools, meaning integration points and APIs that let it act. Building one from the ground up also draws on machine learning and deep learning, understanding natural-language commands, recognizing patterns and refining decisions over time. He adds that chaining, the ability to break a complex request into subtasks, prioritize and execute them, is a fundamental piece that is not yet fully there.
Patel says use cases are everywhere. In service functions an agent could autonomously detect system failures, start corrective actions and predict failures, which means faster resolution, less downtime, lower operating cost and more personalized service. In supply chain he imagines self-healing supply chains, with agents detecting failures, simulating scenarios, prescribing actions and learning from past recommendations. Garg adds that these are the same problems earlier automation and predictive AI tried to solve.
Patel suggests the areas that were always low-hanging fruit for automation and assisted AI, such as procurement, finance, HR and IT back office, where a lot of human effort goes into little value, and then building toward higher-value decisions, like a pyramid. Garg agrees: after about 18 months of assistive AI with clients, production use is more likely in internal functions where risk is low. He also advises groups that have not tried assistive AI to start there, with a human in the loop.
Patel says the considerations have not really changed: get the data right, prioritize use cases and secure strong partnership and sponsorship, have the basic technology and engineering base such as AIOps, build trust in models and data systems, and work hard on culture and adoption because of fear and lack of understanding. The big difference between generative and agentic AI is the technology curve: generative AI is getting there in maturity, while agentic AI is a very new idea for complex use cases.
Patel notes that vendors are launching agentic products, but winning products are not clear yet, and he expects foundation models to become a commodity, so the value is in what is built on them. He says only about 1% of enterprise software had agentic capabilities then, and the prediction is 30 to 40% in two to three years. Enterprises will probably buy pre-built products, integrate market offerings with their own platforms, or build custom, most likely a mix, judged on speed, cost and risk. Garg calls it more of an integration play and says vendors will have to reimagine their products. Patel puts it as today’s SaaS being tomorrow’s agents.
Patel says the space moves fast and his answer could change in two weeks. He sees agentic AI moving into physical systems such as robotics, progress toward more general intelligence, enterprise software changing how the workday starts, and possible step changes in discovery of new drugs and materials. His closing thought is that personalization will become cheap, where standardization used to be the way to cut cost, so experiences can be tailored to the person logging in. Garg agrees that this can change how roles, teams and enterprise software work.
So, if you want to define it very crisply, I would say a solution that is able to reason, act and learn is agentic AI.
The use cases exist. These are the same problems that we tried solving using automation at some point. These are the same problems that we use predictive AIs for.
The Infocepts Point of View
Two ideas run through this conversation. First, agentic AI is a matter of degree: more autonomy and complexity and less human involvement than assistive AI, so the first job is to know where each of your own use cases sits on that scale. Second, much of the work is the work you already do: data, integration, and people and processes that are ready for change.
Shashank Garg’s advice is practical. The use cases already exist, so ask whether the technology, the data and the willingness to reimagine the business process are in place, and if a group has not yet experimented with assistive AI, begin there with a human in the loop.
For leaders planning agentic AI, three questions follow from this conversation: which low-risk process will you test first, what data, integrations and trust does the agent need in order to act safely, and which parts will you buy, integrate or build?
About the guest
Kabir Patel
Global Head of IT, AstraZeneca
Kabir Patel is global head of IT at AstraZeneca, where he oversees technology capabilities in drug development, manufacturing, digital and sustainability. He has driven digital transformation in operations, and before AstraZeneca he held consulting and engineering roles at Oracle and Thomson Reuters.
Facing this problem?
Talk to our Life Sciences experts about where agentic AI fits your roadmap and how to start with low-risk use cases.
#1 on Gartner Peer Insights, 3 years running · 97.2% client retention
Bring This to Your Team
A private briefing on agentic AI for your enterprise, covering use cases, building blocks and how to start, set against your own data estate.
Shashank Garg 00:00
Good morning. Good afternoon, everyone and welcome to the Intelligent Leader podcast. This is your host, Shashank Garg. And with me, we have our guest, Kabir Patel. Hello, Kabir.
Kabir Patel 00:19
Hi, Shashank. I’m thrilled to be here.
Shashank Garg 00:22
Kabir is the global head of IT at AstraZeneca, overseeing technology capabilities in drug development, manufacturing, digital and sustainability. Kabir, I’m told you have a strong passion for innovation, and you’ve been playing a key role in driving all the digital transformation, especially in operations. And with operations, I’m sure you enjoy solving real-world problems, building high-performing teams and delivering tangible results. Prior to AstraZeneca, you’ve had, you know, held various consulting roles and engineering roles at Oracle and Thomson Reuters. Once again, welcome to the podcast, Kabir. And as I know you’ve been deeply involved in driving IT transformation at AstraZeneca. So, can you share a little bit more about what excites you the most and how are you doing what you’re doing? Yeah.
Kabir Patel 01:17
Thanks Shashank. My background lies in leading technology teams with a strong emphasis on pharma tech, manufacturing, digital and sustainability. Now, in today’s rapidly evolving landscape, tech is a truly center of everything we do, is driving significant change, is driving innovation across industries. I’m particularly excited about how advancements in technology, especially AI, the emerging ideas around agentic AI is not only changing the game but is unlocking immense value for all the businesses and as society. So, the most exciting part to me is to be part of this wave that we have with us and contribute towards making a value created.
Shashank Garg 02:06
Awesome. Awesome. You spoke about AI especially agentic AI. Why don’t we begin that? Why don’t we begin there? Let’s discuss what you believe is agentic AI, and what it is not. Just love to get your thoughts there.
Kabir Patel 02:22
Thanks Shashank. I think it’s an interesting question, right? Like every time we get a new definition, a new concept, I think initially there are multiple ways people look at the same idea. Now, to me, AI solutions with a degree of agency that enables them to operate independently, to achieve broad objectives are fundamentals of being agentic AI solutions. Right. So, unlike traditional AI where you have predefined task and you respond to specific commands, agentic AI models need to function as digital agents capable of independently assessing situations, choosing actions and learning from experiences.
Shashank Garg 03:04
Yeah.
Kabir Patel 03:05
So, if you want to define it very crisply, I would say a solution that is able to reason, act and learn is agentic AI.
Shashank Garg 03:15
Yeah. I like to add in a few things. If you were to compare agentic AI, general automation, RPA for example, which has been happening for a while in traditional AI on the key features like complexity, human interaction, decision making, adaptability, goal orientation and then eventually autonomy, as compared to the others, you would find agentic AI very high, both in complexity and autonomy. You’ll find it lower than anything else on the human interaction, which you know we’ve been talking about quite a bit, that all the AI that is being worked upon till now kind of falls in the assistive bucket. Right.
Shashank Garg 04:03
Where the eventual decision making is with humans and rightfully so. But agentic AI just takes it a notch further. And as you were saying, right. That being able to assess and then being able to reason and then make decisions, I think it takes it forward. It’s also very adaptable. It’s meant to be adaptable so you can adjust to scenarios that they were not given to it. So, it must have the intelligence and the use of the word agency, the amount of agency. So that’s really the difference. And we’ve been part of the whole technology revolution from the field of data and then the traditional analytics and AI for the last five to 10 years. I think this is where things are getting really advanced and complex, and we all need to tread carefully here. Just chatbots are not agentic AI.
Kabir Patel 05:02
Yeah. I think that’s a very important distinction to know what is not agentic.
Kabir Patel 05:11
One of the things that still amuses me a lot in my every day is a few things that comes to me with a pretext of being agentic. There is a similar things that I used to get when generative AI was a thing where in reality they are AI workflows with a user interface.
Kabir Patel 05:31
For something to be a very complex solution that can do the cognitive thinking of a human and replicating that in a difficult ever-changing scenario is not a very simple solution to build. But the promise is there. We are almost getting there in terms of how technology is maturing.
Kabir Patel 05:50
So, I think the future looks bright. But you’re right, not everything is agentic.
Shashank Garg 05:56
Not just chatbots, not just rule based engines, automated trading bots. You wouldn’t put it…
Kabir Patel 06:03
Just an API call to LLMs. Right.
Shashank Garg 06:06
Yeah. Good, good. I think we are all clear on what this means. Just taking maybe a step forward now from, before I go there on examples. Right. Do you want to talk a little bit about what you would call sort of the building blocks for AI and especially agentic AI, any comments there?
Kabir Patel 06:27
Yeah, I mean we can look at like agentic AI building blocks and how the idea can be made real and how to go about it. So, to start with, the core frame of technology still needs few things, right. You need the memory, the context, you need some sort of reasoning engine, you need a foundational model. Then finally you need some tools. What I mean by tools is ability to act. So, integration points, APIs, and way to conduct the actions.
Kabir Patel 07:05
Now in addition to that, to construct something like agentic AI ground up. You’ll end up using a number of AI training techniques, right. From machine learning, deep learning to reinstall certain things. You probably need the system to comprehend and respond to natural language commands, recognize the patterns in data, and refine decision-making over time, which is like learn and adapt to the situation that is changing. And finally, one fundamental thing that is still almost getting there, but not there in a big way is this characteristic of chaining. So, what I mean is the ability to break complex requests into manageable subtasks, prioritizing them, then reasoning them and then executing them. Yeah, I think a chaining technique becomes a fundamental aspect of any good agent AK solution till you kind of crank that they’re still not there.
Shashank Garg 08:03
Yeah, yeah, absolutely. Maybe it’ll become clearer when we discuss some examples. You know, from your experience, what use cases do you see evolving across the value chain, especially the pharma value chain, drug development, supply chain, and commercial operations and where do you see the big bets here for this one or for the technology here?
Kabir Patel 08:27
Sure. So, Shashank, I think we both recognize the impact is huge.
Kabir Patel 08:33
So, this is the big moment in the technology frame. Same as the Internet, same as when the smartphones became a thing, same as now, AI moving to become more autonomous in every which way. So, impact is huge. Now before I get to pharma, let me give you a way of looking at use cases, because the big reality in front of us is that the use cases is everywhere. Pretty much most things can have a significant value unlocked using the deck. So let me give you an example, like let’s start with the basics. If you are into business of any kind of a service, be IT service or any service function, what you traditionally have is a knowledge worker with experience and knowledge. Now if you add agentic AI to this and think that what AI could do is autonomously detect system failures, initiate corrective actions, and even predict failures, which means you are faster to resolution, you’re reducing downtimes, you’re having lower operational costs, and on top of it, you can enable more personalized service because your AI is learning from the user behaviour, adjusting responses. So fundamentally, the way services get delivered or supports get delivered is changing with the use of agentic AI.
Kabir Patel 10:05
So that’s more of a example everyone can relate with. But if I come to like specific to pharma or the industries we are in, let’s take example of supply chain. Everyone is in the business supply chain. What happens today is if you have an issue or you detect a failure in the network, someone must look into the problem, figure out what are the options on the table, take the actions as needed and fix the problem. Now the whole thing is time consuming, takes a lot of resources and is very reactive. But you could imagine a future where organizations are building more resilient systems. Let’s say self-healing supply chains. A series of capabilities that can see data end to end with agentic AI at the center, detecting failures, simulating scenarios, prescribing actions, learning from the past recommendations and then taking actions. So, agents overseeing the supply chain with a self-healing AI capability at the end. So, a lot of fundamentals of the business and the processes will be reimagined in the way you execute them.
Shashank Garg 11:23
Yeah, you bring up a good point. I mean, I’m also realizing, and that’s what I tell our clients, that we don’t need to look at newer use cases. The use cases exist. These are the same problems that we tried solving using automation at some point. These are the same problems that we use predictive AIs for. These are the same problems a lot of the data analytics focus is enabling for the humans to make decisions. We just have to see the current state of technology, the data that exist for the training, for the model to become autonomous and do we have the tech to make itself learning? And lastly, are we ready to imagine, reimagine the business process? I think that’s where at least we are finding across our clients, wherever we are running these kinds of pilots or trying to build agentic AI examples that not every business leader is ready to reimagine the business process. This is another level of autonomy, not get there in the next two quarters, we will not get there maybe in a year. But whatever experiments we can, whatever, you talked about the building blocks, if you are able to at least get two of the building blocks you have built this year, that gives us a higher chance to get to some of what you said a year from now.
Shashank Garg 12:52
So that’s how the journey is going to be. I’ve also realized that if you’re an organization or a business group that didn’t experiment with assistive AI till now. Let’s not start with agentic AI because you know that’s. Let’s get people familiar with the AI terminology. You know, let’s at least get them used to human in the loop. Maybe a lower-level autonomy and just the assisted pieces of AI and then they can take that next step as well. Kabir, are there things that… Because you know I want to put a time perspective to all this. Right. And you can talk about your role, or you can talk about industry in general. What are the areas that you think organizational purse in the context of pharma can look at in the next 12 months for example?
Kabir Patel 13:44
I think if you want to look at like short to near term.
Shashank Garg 13:48
Yeah.
Kabir Patel 13:49
Probably focus on the areas that were the always the low hanging fruits for automation and assisted AI. For example, any of your corporate functions like procurement, finance, HR, IT’s back office. You look at areas where you see significant human effort going in for very little value. You can then build on that foundation in that frame. Then go after the areas where the cognitive functions require certain amount of maturity, but the impact is big of the decisions.
Shashank Garg 14:34
Yes.
Kabir Patel 14:35
So, you build the pyramid like that. Right. You start where I think traditionally, you started with automations but then you end with the high value decisions.
Shashank Garg 14:45
That’s a good way of looking at it and actually it’s consistent with how we’re seeing a lot of our clients run their experimentation and some of the first use case in production after having done this for 18 months at least on the assistive AI side I can easily see the trend that chances of putting something in production is much higher on the internal function side as you’re saying back office where because the risk is low you can control it. You’re not touching your clients’ consumers business side that much. But once you’ve learned from what worked in your organization then you can much easier to apply and at that point you also have the track record of having done something very successfully, be it internal. But I think that’s a great way to start. Thank you for that. Moving on from here, Kabir, I know a lot of organizations still struggle to go from like scaling AI is a common topic of concern and especially if you talk about enterprise adoption. So, what would you share sort of the key considerations for embedding and scaling both. You know, maybe talk about GenAI as well because not everybody has been able to scale that even. And then maybe agentic AI later.
Kabir Patel 16:00
Sure. So, Shashank. That’s a very, very important point. The funny thing about that in my perspective is the considerations have not really changed. So, what I mean by that is whenever we worked, even in a, even in a very different frame or in projects in past, you start with the idea that what are the key ingredients I need to be successful even for Gen AI, to that matter agentic AI, you do need to get your data right. That consideration still there.
Kabir Patel 16:39
You need to prioritize use cases and need strong partnership and sponsorship in your organization to drive a change like this. Then you need your basic tech infrastructure, call it AIOps LMOps, basic frame of how tech gets engineered. You need trust in your AI models, foundational models, your data systems. The trust plays a big role in the next key consideration, which is culture and adoption. So, you need to work very, very hard on getting the culture right, because a lot of times there is a fear of this new tag. There is a lack of awareness and understanding of the tag. Probably you’re imagining the process and that’s not landing. So, the consideration of change and culture is important. And finally, the big distinction between gen AI and agentic AI in my view is the technology curve. Gen AI is getting there in terms of maturity. Agentic AI is a very new idea for some of the complex use cases that you see. So, a key consideration is how mature is the tech for what you’re trying to do with it. It would distinguish between gen and agentic AIs, but broadly for AI, data prioritization and sponsorship, change in culture, trust, having the right kind of technology, engineering frame of operations or AI offices.
Shashank Garg 18:11
Moving forward, you know, if you look at the technology landscape, I don’t know of a firm of any size, of significant size who hasn’t come and launched their own agent tick AI product or framework in the last six months. Be it NVIDIA, be it the big three, Amazon, Google, Microsoft, all the SaaS players, the large SaaS players have their own and then VC. If you follow the Y Combinator and the VC circles, you see at least 20 startups, brand new startups, whoever agentic AI framework or product platform. So, any thoughts on, you know, how will the technology landscape evolve? What do you think an enterprise like yours is likely to choose as you go on your own agentic AI journey?
Kabir Patel 19:05
Well, it’s a very interesting question, Shashank, because it is true that a lot of vendors, pretty much every vendor who is in the space is launching agentic AI product or models. But if you really look a little bit more closely, the gain of winning products is not there yet. So, this is like a step one of people trying to win, because everyone understands, in my view, that foundational models are going to be commodity.
Kabir Patel 19:38
What wins are the products you build on them and that journey is not very clear today. So that’s a data point one.
Kabir Patel 19:47
Then if you look at what an enterprise really has today is a lot of enterprise software bought in last 10, 15 years, mostly SaaS. In that play, only 1% of them are having agent AI capabilities as of today.
Kabir Patel 20:06
But in next two, three years the prediction is this would be anywhere between 30 to 40%.
Shashank Garg 20:12
Even more, so.
Kabir Patel 20:14
Even more. Right. So, a lot of that we are sitting on is becoming agentic as part of the role.
Kabir Patel 20:22
Then comes the third important idea, which is what happens to enterprise, we’ll probably have three options. We buy agentic AI pre-built products, to my earlier point, that someone will create winning products out there. The second option, we integrate some of what we see in the market with some of our platforms and create an experience. Or we build using our standard tools and engineering capabilities and deliver something that is completely custom-built for our enterprise. Yeah, I suppose the answer is all three, depending on the use case.
Shashank Garg 21:06
Yeah.
Kabir Patel 21:07
Probably we’ll look at speed, cost and risk and see on those parameters which is a good option to go. Most large enterprises at any given point in time, in three years would have a combination of all.
Shashank Garg 21:21
Yeah, fair point, Kabir. On this one, I’m certainly seeing across our clients because I believe that if you reflect back on your answer on what are the building blocks of, you know, agentic AI? Right. And the data piece and the ability to have context and then the reinforcement learning, you know, continuously learning. And then integrations, multiple integrations. Right. So out of the three things that you said, which is the large platforms, then all the SaaS players are custom-built. My mind is on the custom build or the large platforms because you have to have much more than a SaaS product. This is an integration play. This is far more of an integration play than choosing one SaaS product and sticking to their play unless the whole use case can be embedded within that.
Kabir Patel 22:20
I agree. But equally, I think what has happened is, and this may sound counterintuitive, but what I think has happened is a lot of enterprise software has created a certain amount of rigidity within the enterprise. Yeah, because it creates boundaries.
Shashank Garg 22:38
Yes.
Kabir Patel 22:38
What software is supposed to do?
Shashank Garg 22:41
Yes.
Kabir Patel 22:41
We are moving towards completion of the task, a task-oriented play that is more integrated depending on what needs to be done to achieve the outcome.
Shashank Garg 22:51
Yes.
Kabir Patel 22:52
So that I absolutely agree with. However, I also feel a lot of very mature vendors in the space are also thinking the same and they will reimagine what their product really means.
Shashank Garg 23:03
Yeah, yeah, yeah. Unless that happens. Yes, you’re right.
Kabir Patel 23:07
Because they’re sitting on a lot of our enterprise information. I think they have a big chance. But yes, if you think the way enterprise software used to think, then it’s very difficult to see the value creation there.
Shashank Garg 23:21
Absolutely. Yeah. The old mantra of let’s lock this client down, this use case down in our application and then we have nothing to worry about. It’s not going to work. You have to broaden, you have to reimagine your own products. It’s a… You will see a lot of reimagining of the current SaaS of vendors.
Kabir Patel 23:37
Yeah, I mean today’s SaaS is tomorrow’s agents is just what form they reimagine themselves.
Shashank Garg 23:42
Yeah, yeah. And it has to be a broader play, unfortunately. So, agree. Kabir, moving on, just one last question. Where do you think we are headed with all this? Right. So, what’s beyond agentic AI? Any thoughts there?
Kabir Patel 24:00
I mean, your guess is as good as mine. It’s a very fast-moving space. It’s probably so difficult to kind of predict my answer today will change in two weeks’ time. So, things are rapidly moving. Now, of course, what we are seeing, we’re seeing models that can deconstruct complex problems that inquiry data sources provide comprehensive multi-step solutions very effectively and efficiently. You are seeing solutions that can pursue higher-level objectives, including human cognition in some form based on what they reason as the right next step. So, these are the kinds of capabilities that we never had before. Right. Even when we moved from automation to bots to a few other technologies that came in, we never had this idea that someone could do what I just described. So, to me, the future has so much potential and so much value that it can go in any direction. A few things that I see will happen is something like this that is in the frame of…
Kabir Patel 25:17
Agentic AI probably will move in a direction where the next big thing would be a physical system with agentic AI at the centre. So, you can see more advancement in robotics in that area where the systems become driven by agentic. But they are physical structures to get a lot of things done. Yeah, you can of course then see more general intelligence that everyone is trying to get to as the step. In terms of use within the enterprise sphere. I think enterprise software is fundamentally coming to an important work. I do not see someone starting a workday today in the same way the workday was. Right. So, it’s fundamentally different. So, enterprise software space I think is changing. You can also go into the zone of whether this can lead to a step change in a lot of new discoveries.
Kabir Patel 26:16
Right. Newer drugs, innovative materials, things or problems that you never thought were possible to solve. You can have wars without humans. Well, prefer that we don’t want wars. But I think a lot of areas are fundamentally can be reimagined. And finally, I will leave you with one thought that I think may happen. It’s like for years we almost equated standardization to the way of reducing cost anywhere, right? So, people run large programs to standardize. Slowly what we’ll see is personalization is going to be so cheap. There are a lot of experiences that can be personalized based on a persona.
Shashank Garg 27:02
Yeah.
Kabir Patel 27:03
Like in your personal lives, you may have your own assistant, you may have your own itinerary experiences, your preferences, and your own movie. Right. The enterprise world depends on who is logging in the whole screen. And the experience can be tailored and different. So, it’s fundamentally marketing changes from personalization being this expensive thing or customization being a bad word to becoming the way a good experience gets delivered.
Shashank Garg 27:32
I love the way you imagined this future. I agree with many of the things that you said. You’re right. The combination of having this intelligence available to everyone in the enterprise or in our personal with a combination of self-learning and automation can completely change the way we do things today, including enterprise software, how many of our roles are structured today, and how our teams operate. And I’m certainly in the bucket that it’s a great opportunity for everyone who’s listening to this podcast for all of us to get excited and participate in this reimagining. I also love the way you describe that. You’re right. For a long time standardization, we just, you know, the way to cut cost was standardization and personalization was a really bad thing. And with AI, that’s no longer the case. That’s no longer the case. You know, four years, five years ago, if I had to support six different personas, you know, there’s enterprise software, SaaS vendors would charge you by persona.
Kabir Patel 28:53
Good.
Shashank Garg 28:54
You don’t need to. You don’t need to anymore. So great thoughts, Kabir. And I’m excited about what this future holds for all of us. So, thank you very much. It’s been a pleasure having you on the show today. Your insights on what AI, agentic AI is real-life applications and how the future can be imagined are incredibly valuable. Thank you for taking the time. Thank you.
Kabir Patel 29:28
Thanks Shashank for having me. And we’ll catch up soon.
Shashank Garg 29:33
Thank you. And to our listeners, we hope you enjoyed this episode of the Intelligent Leader with me, Shashank and our guest, Kabir. If you liked what you heard, please consider sharing it in your next network with the hashtag Intelligent Leader. And don’t forget to hit the subscribe button. And you wouldn’t want to miss the next episode. Thanks for tuning in and see you next time.
Insights & Resources
The latest conversations from The Intelligent Leader.
Retail & CPGRetailers generate more data than ever, yet many struggle to turn it into meaningful business outcomes. This podcast explores the hidden challenges…
Retail & CPG 41:16 Ali Miller, VP of Ads at Instacart, and Shashank explore how data, AI, and collaboration are transforming retail media. Ali shares Instacart’s…
Cross-Industry 31:55 In this episode, Prateek Shrivastava, Advanced Analytics Manager at Cummins and Shashank Garg discuss the transformative impact of AI and data science…
About this episode
Kabir Patel describes it as AI with enough agency to operate independently toward broad objectives, assessing situations, choosing actions and learning from experience. His short version is a solution that is able to reason, act and learn.
Shashank Garg says assistive AI leaves the final decision with humans, while agentic AI can assess a situation, reason, decide and adapt to scenarios it was not given. Compared with assistive AI and automation it is higher in complexity and autonomy and lower in human interaction.
Internal and back-office functions. Patel points to procurement, finance, HR and IT back office as the low-hanging fruit, and Garg says that after 18 months of assistive AI with clients, production use is more likely in internal functions where the risk is low and clients’ customers are not touched.
Patel lists getting the data right, prioritizing use cases with strong partnership and sponsorship, a solid engineering base such as AIOps, trust in models and data systems, and culture and adoption. He adds that technology maturity matters, because agentic AI is much newer than generative AI.
Patel expects a combination of all three over three years, chosen by speed, cost and risk for each use case: buy pre-built products, integrate market offerings with existing platforms, or build custom. Shashank Garg sees it as more of an integration play than a single-product choice.
Yes. The full speaker-labelled transcript of the 30-minute conversation is on this page, free to read with no form, with chapters that match the key moments.
The Enterprise AI & Data Transformation Partner. Delivering measurable business impact globally.
#1 on Gartner Peer Insights · 97.2% client retention
