MediaMint CEO Rajeev Butani on How AI Agents Are Transforming User Interaction

Hello, and welcome to the latest edition of the Textron AI Leadership Insight series. I'm your host, Mike Baer. Today we're with Rajib Botani, who's CEO for Media Man.

And we're gonna have a chat about how AI agents will change the way we interact with software. Rashiv, welcome to the show. Thank you, Matt.

Thanks for having me. Alright. For decades now, we have struggled with graphical interfaces for humans and tried to build these things and left brain people trying to build something for right brain people, and it never quite worked out the way we had imagined.

Now we're gonna add AI agents and that's gonna change the way we interact with software. But what will that experience be like? What do you think?

Yeah, Mike, that's a great question. Um, when, when I think about AI and when, when I think about agents, you know, I think about it from a holistic manner. You know, there is an element of software and how software is gonna get invo, uh, transformed.

But then if you step back and think about what's gonna happen to businesses and how businesses are run, I think the entire transformation is gonna take place all the way from strategic level all the way to the software. You know, for example, to the question that you asked, the way we think about it, oftentimes, uh, when you are integrating software, you're bringing UI or you are introducing systems, uh, humans have to go to those systems. So if you're using Salesforce, you have to log into a Salesforce screen as an example.

If you're using SAP, you have to log into an SAP screen. Now what's happening is with AI and agent and conversations, these, these new transformations will go to where humans are and where they're working, you know, and that's radically gonna change the way, uh, functions are run and businesses are run. So if I am, if I'm entering an order instead of me going into a particular screen, uh, and AI agent could be in Slack, or it could be on email and it can take care of the entire process, you know, so there's a foundational shift that is taking place with regards to how the work is being done.

Uh, and we couldn't be more excited about, uh, you know, how this is transforming and playing out, uh, within our companies. I'm trying to figure out how that might actually play out, because every vendor you talk to now is adding AI agents to their application. And that's all well and good, but, um, am I gonna have maybe my master or head butler agent, whatever you wanna call it, that's gonna talk to all those other agents?

And that's how that interaction's gonna be. And I'll have a common experience, but am I really gonna go in and ca um, make friends with all these different AI agents and all these different other applications? Uh, you know, Mike, um, different companies are approaching it from different angles.

Uh, I'll share with you how we are approaching it, you know, and then, you know, also talk about what it means for customers. But the way we, when we think about functions and roles, we think about, uh, you know, people who are doing a particular piece of work. So, for example, you know, I'm, I'm a sales manager, or I'm an account executive, or I'm a media planner.

If I'm working for a media industry, um, the way we are thinking about it is to create augmented assistance, uh, where we are creating persona based models, which will sit side by side with humans and drive significant efficiency for those humans. You know? So that's the way we are thinking about it, uh, where they will tackle not just the work that's being done by one person or for that particular role, but everything else that other person needs to do in order to go and drive efficiency.

You know? So that's how we think about it. We are taking an augmented persona based approach.

Yeah. In order to drive that, you know, just to bring this to life, you know, if I just step back and talk about media men. So we, uh, we are, you know, we work primarily with media entertainment and technology companies.

So today we work with over 150 companies with these companies We run, uh, help them run their revenue operations, their media operations, uh, their sales and marketing function, uh, on a day-to-day basis. Uh, and uh, then I think about one of the roles, you know, of a media planner. Today, a media planner goes to different systems to create media plans.

They also analyze companies. They create reports, they update their systems. So the way we are approaching it is to create an augmented assistant that sits with the media planner side by side and drives all of these efforts with an AI first approach with this media plan of being a human in the loop in order to make sure that everything is being done in a validated manner.

You know, so that's the approach that we are taking. We quite a bit of traction in that area, but, But ultimately, as you describe it, there is kinda one AI agent that winds up being the orchestration engine or driver for all the other AI agents. 'cause it's the one sending out requests to those different platforms.

So are we essentially gonna have to figure out how to manage what may become a small army of AI agents everywhere? Exactly. Exactly.

So, uh, we've, we've used the taxonomy of an assistant and agents. So the way we think of it, an agent is doing a particular task and an assistant brings a group of agents together to drive a particular function or drive an activity. So, absolutely.

So, uh, you know, there is an orchestrative orchestration element that's there, there's an orchestration element that's there around technology. And then as importantly, there's an orchestration element that's there around the business function and the outcomes that need to be achieved. Uh, so that's how we look at it.

At the same time, not every agent is gonna be fully autonomous. I think there's gonna be like a range of capabilities that we're gonna allow them to execute, and we're gonna have to figure out, you know, just what is our comfort level with letting something go off and do something in the background and provide us some sort of result. But I guess I'm getting at the point is it doesn't seem to me like all AI agents are gonna be created equal.

Absolutely. Mike, the world that we see is, is more augmentation to humans versus autonomous. Uh, and like you rightly said, depending upon the complexity of the work that needs to be done, there may be, you know, certain set of functions might be completely autonomous, but a large number of functions will have probably, you know, whether it's 30%, 50%, or 80% AI or, or technology based work that will need to be done with humans making sure that, uh, they can do the work.

You know, so I see this as a journey, you know, as the foundation models are evolving, as companies are strategically integrating AI into their workforce and into their, into their, uh, functions. I see this, uh, spread that will come in with regards to what AI is going to do and what humans are going to work through. You know, I see a model where there will be a star network, if you think of it, it's star network where you've got these assistants which are at the, at the nodes and the human sitting in the middle trying to orchestrate the work and deliver the outcomes.

Mm-hmm. Will there be a certain amount of nuance that needs to be navigated in the sense that right now, if I go look at ai, sometimes I feel like it wants to be overly helpful to the point of maybe telling me something that's not true. But, um, won't these agents also be similar in the sense that they will wanna be helpful, but how do I kind of temper that down a little bit so they're not constantly interrupting my thought process with some other suggestion that they think we should be doing?

Yeah, yeah. So let me just bring this to life, Mike with an example. So for a number of companies, we, so, you know, one of the areas that we work with, with companies is to help with their advertising and revenue operations, you know, so today we work with over 60 publishers and platforms who make money through advertising, and we help them drive all of the campaign optimization and campaign activation.

So the person that's running it within a company, that person is called a campaign manager as an example. So the role of a campaign manager there is to make sure that any campaign that's being run is pacing in the right way, uh, is driving the right level of outcomes, is performing in the right way. And as a campaign manager today, I'm managing probably a hundred to 200 campaigns in a very manual manner.

So when we introduce AI, and when we think about the role of an ai, you know, to what you mentioned, uh, the role of the AI is to understand which campaign is pacing, right? Which campaign is performing, right? And then start to give some recommendations in terms of what could be done so that they could achieve the right level of impressions and return on ad spend.

So in that manner, you can see that it's very deterministic in terms of what specific actions could be done, but the responsibility for taking the actions will live with the, with the campaign manager in order to do so. You know, so it's, it's one, it's, you know, one, one part where you're looking at an answer engine. It's another, another sort of activity when you're looking at an agent system to help you complete a workflow, you know?

Um, and I think there, I think we've, uh, we are seeing a pretty strong application and delivery of benefits with genic systems. Mm-hmm. Of course, we're talking about how, um, end users will access software, but we collaborate with each other and we need to collaborate with the agents.

And I'm trying to figure out if long term, am I gonna have a bunch of agents that are kind of tied to me and they'll go negotiate with agents tied to other people on the team? Or will the team have a set of agents that are kind of assigned a specific set of functions to handle on the behalf of everybody? Yeah, yeah.

You know, Mike, I, um, what I believe is, um, as the adoption of AI increases in the enterprise, um, customers will start, or enterprises will start to expect, uh, this consolidated sec it's agents and humans to deliver work and outcomes. For example, if you look at the world that we live in today, SaaS companies, they, they provide SaaS software and then someone in the organization needs to deploy it. They need to implement it, they need to run it.

And you look at, uh, consulting companies, they will come and provide consulting. When you look at operations companies, they will run operations. I think what the way the world will move, or the way this entire ecosystem or architecture will evolve is all around delivery of work and achievement of outcomes.

And then when you start to go back to your point, there may be roles that are, that are working together. So the example that I've given of a campaign manager, so a campaign manager works with an account executive, uh, a campaign manager works with someone that's creating a media plan. Now, each of these roles will have their own set of assistance that are powered by agents, which will orchestrate with each other, with, you know, with an augmented human approach in order to go and deliver the book and an outcome.

You know, so if you think of it from that perspective, there will be a, a group of, uh, enablers in this case. And some of them may be, you know, agent, some of them may be pure automation, but they will work together to go and achieve an outcome for the business. That's how I mm-hmm.

Of course, one wa one said, you know, it's one thing to be wrong, it's another thing to be wrong at scale. And a lot of these AI agents will be moving at speeds that are faster than we can comprehend. But if something goes wrong, do we need to figure out some way to roll things back?

I mean, what level of control, or as we say, you know, can I throw the AI agent in reverse? Yeah. Yeah.

Mike, I think this is where the point that you raised earlier around, around autonomous versus human in the loop approach comes in because the whole objective of, uh, creating an agent system or an agent service where we are leveraging the power and benefit of ai, but at the same time, we are doing this in a very careful, uh, manner and in a very responsible way along with human in the loop will allow us to limit the, the particular, you know, this type of deviations that will take place. So that's how, you know, we see it, uh, and that's how we are implementing it across all of our customers. Will I need maybe another AI agent to validate what the other AI agent says we should be doing?

Because sometimes, you know, the AI agent might be misinformed or just plain old hallucinating, but, um, at some point, do I need maybe, uh, another mechanism where there's an AI agent built on a different LLM that's validating what the other AI agent is saying and wants to do? We are looking at such type of instances, you know, we are looking at different types of instances where we are using an AI agent for creating some other AI agents. So we are, we are using it for that purposes.

We are also using it for purposes where exactly to what you mentioned, we are using one foundation model for, for delivering an outcome, and we are using another foundation model to validate that outcome. So we are seeing all of these particular outcomes and these options, uh, but the, the, the end goal that we are looking at is to make sure that we can deliver the right level of service for our customers. Like the way, you know, we, um, uh, when I think of different types of companies, we have big domain experience in, uh, media, entertainment and technology industries.

We primarily focus on the front office, you know, we focus on sales and marketing and uh, you know, service functions. We are using our domain expertise to answer the questions that you're mentioning as an example, and to ensure that there is limited amount of hallucination that's done and division from the business outcome that's achieved, uh, when we are building these agents. Alright.

Of course, every new technology has some unintended consequences. And the one that I'm kind of scratching my head about lately is we have all these functions within a business, right? There's sales, there's marketing, there's finance, there's manufacturing, whatever it may be.

As we move to these AI agents, is there gonna be maybe an opportunity to start rethinking how companies are structured? Because the AI agents will be talking to each other in ways that are interesting and things that we set up as, uh, structures for humans may not apply to AI agents or just maybe outmoded. Yeah.

Yeah. Mike, uh, every inflection point, and each time a new technology has come in, whether it was during the internet or whether it was around mobile, or whether it was around digital transformation, each of these technologies, and for that matter now with genic AI has an implication with regards to the operating model of the company. You know, how the company's structured and how the company operates.

So I do expect that as we are maturing, uh, the adoption of AI within the enterprise and as it moves away from being used in simpler ways like chat GPT at this point in time, you know, for answering questions to being integrated into the process, I do expect that operating models will change, uh, to what extent it'll change and how it'll be, how it'll evolve is gonna be based on how that particular company is adopting the agent system and the outcome that they're achieving as a pop of it. Yeah. So what's your best advice to organizations about how to get ready for all of this?

'cause I think a lot of folks are, well, they're intrigued for sure. Some are even downright excited, but I also think they're feeling maybe a little overwhelmed. So how do I kind of wrap my head around all this in a way that allows me to get started without being paralyzed because of I'm spending too much time, frankly, analyzing what might be.

Yeah, yeah. You know, Mike, to me it is about, as companies are thinking of adopting ai, it is very important to stay away from the hype and focus on, uh, a specific set of how they can strategically integrate AI into their workflows, prove that out, and then scale it. You know, we just did it with, uh, with, with a fairly large linear TV and digital company where, uh, you know, they've been working on, they've been working on this initiative for a year, but in four weeks time, we picked up an area, we've been able to demonstrate how that area could drive significant effectiveness, productivity gains, and then, you know, eventually cost savings, uh, by using ai.

And I think we're gonna use that as a nucleus now, go and scale across the organization. So as long as we are staying away from the hype, and as long as we understand that this is change, and change within organization is hard, uh, I think we can do this in a very, you know, deliberate manner, uh, proof success and then scale it across the organization. Of course, AI agents are not free.

And one of the things that we have seen so far is that it can be expensive to automate a process using AI agents. So, um, do we need to have a better understanding of what the cost structure looks like? Because sometimes I talk to people and, you know, suddenly they're spending a million dollars to automate a task that's managed by somebody and makes 50 grand a year.

Uh, we, um, you know, because we come from a domain experience, so when we are looking at a particular area, uh, when I think about, you know, we come with deep understanding of what that area is, and we don't have to spend time learning about the area, that's number one. So I think that helps us optimize the way we are coming and approaching ai. I think the second thing that we've done, Mike, is we are leveraging AI to make AI far more effective.

For example, um, as long as there is a clear understanding of a playbook or a runbook for that process at a particular company, we can stand up these AI agents in three to four weeks time. You know, because, because we build the infrastructure of specific generic agents that can be customized for a particular customer, uh, in for that particular function, and then deliver the outcomes. Uh, so I think, you know, there's an efficiency that comes in, but it starts with a deep understanding of the particular customer, deep understanding of the domain and, and intelligent infrastructure that's being built that, uh, is a one-time effort to build, but that can be applied in a, in a specific way across different workflows and across different customers.

All right, folks heard in here, even in the age of ai, there is no substitute for domain expertise. So the more you know about how a process works, the more successful you're gonna be. Hey, Rajib, thanks for being on the show.

Thank you, Mike. Thanks for having All right. And thank you all for watching the latest episode of the Techstrong AI Leadership Insight series.

Been buying this episode and others on our website. We invite you to check them all out. Until then, we'll see you next time.