Next in Tech | Ep. 205: Agentic AI Impacts

View Full Transcript

Eric Hanselman: Welcome to Next in Tech, an S&P Global Market Intelligence podcast where the world of emerging tech lives. I'm your host, Eric Hanselman, Chief Analyst for Industry and Company Data at S&P Global Market Intelligence. And today we're going to be talking about the impacts of agentic AI and some of the thoughts that have been kicked around out there in the marketplace about what its impacts are going to be to the broader market and with me to discuss it. 

Our two returning guests, Chris Marsh and Sheryl Kingstone. Oh, welcome back to the podcast. You both.  

Chris Marsh: Yeah, thanks for having me back.  

Sheryl Kingstone: Yeah, thanks, Eric. Love the topic.  

Eric Hanselman: It is a lively one. As we have been discussing not too long ago, Microsoft's CEO, Satya Nadella, made the comment that agentic AI is going to fully disrupt software as a service providers and some of the applications that are so widely known. 

And I guess I wanted to get a discussion going about what we think about this, some of the aspects of it. I guess to begin with, let's dig into the general idea or the proposition that he's putting out there and what that really means for agentic AI and Chris, maybe we could go to you just to lay some groundwork here about why this idea has been cast out there. 

Chris Marsh: Yeah, I mean, I think there's sort of two parts to this sort of argument. One is the adoption of generative AI and the growing interest in everything assistants and agents. And then there's the impacts that may have in the longer term when we think about SAS and just the sort of wide landscape of different business applications. 

So maybe I'll start with a little bit of our data. So, we collect a lot of data around These topics, we have two very good surveys that go to AI decision makers within organizations of different sizes across different verticals that gives us a lot of, you know, interesting sort of market data and perspectives. 

And what we know is that generative AI has already accelerated past longer standing firms of AI in terms of adoption. Right? So, 27 percent of organizations say that it's now fully integrated across their organization. So not just in specific pockets or specific departments, but broadly deployed. And that's higher than for traditional kind of rules-based AI and pattern recognition forms of AI. 

And the expectation is that by the end of this year, by the end of 2025, that 27 percent is actually going to be 40%. So, 40 percent of Organizations will have generative AI widely deployed. The interest around assistance and agents is really high as well. So, 58 percent of organizations say they're actively seeking opportunities to implement AI agents and assistants in their organization. 

And then 40 percent say that they're open to exploring that. So put those together, it's pretty much everyone who's either actively already looking for the use cases, the scenarios specifically for agents and assistants. Or they're saying, yeah, this is something we're really open to. And we're thinking about the rapidity of adoption, the profundity of this technology waves. 

is inevitably leading to lots of discussion around what the longer-term impacts may be on business applications. And with that has emerged this narrative that you've kind of outlined that this could potentially have, you know, significant implications for how we think about SAS business applications going forward. 

And Satya Nadella is one of like a growing number of voices that are beginning to sort of. put some thoughts and perspectives around it. So, the basis of that argument is the SAS and what's been 25 years of SAS sprawl is very inefficient that you have lots of single point solutions that basically function as like a crude database with business logic presented into a custom UI. 

And the organizations have had to do lots of work to integrate across. those apps to deliver the workflows that their employees actually need. And it would make sense for the business logic to actually transition to an AI layer and to have conversational AI assistants, essentially design and orchestrate agents that can touch and manage multiple. 

databases on the back-end sort of simultaneously and seamlessly. And with that we'll see more AI native business applications that operate across various systems and with-it reduced reliance on those traditional categories that we've always thought about when we think about SaaS, CRM, ERP and other SaaS categories. 

So that's the kind of thesis based on. that crazy adoption we're seeing of generative AI, the growing interest in assistants and agents, and yeah, the potential impacts that may have longer term on SaaS as we've known it for the past couple of decades.  

Sheryl Kingstone: I was just going to add on to what Chris was saying, right? 

The original discussion was, is it going to be the disruptive to SaaS? Is it going to be the death of SaaS? Right? And if you hear what Chris is saying is, yes, we do have fragmentation and pain points across the application. And everyone under the sun in the tech industry seems to be focused on big statements. 

To come out and generate some hype in the industry. But the reality is we have to look at this, the way organizations are going to be using this. Right. And so, yes, we have fragmentation and complexity. Absolutely. We have been for two to three decades trying to deal with those significant changes and overlays, whether it be integration buses, whether it be data lakes for consolidation, and whether we were looking at task-based UIs in the past, right? 

None of it solved the problem. Enter generative AI with some conversational interfaces. And if you remember back over a decade, we did have conversational interfaces over data exploration tools. It just never picked up. It wasn't effective, right? It wasn't an effective use of data exploration.  

Eric Hanselman: But, but now that we've got. 

Generative AI, I mean, we can make grand statements about what we're going to do next. Are you implying that grand statements aren't going to be enough?  

Sheryl Kingstone: No, they're not enough, Eric. And if you really think about it, we still need the underlying mission critical applications for certain use cases, right? But that doesn't mean it's one or none. 

Right. So, we do feel that there are some great use cases for like AI is the new UI. And now we're saying agentic AI is the new UI, right? Because what it can do is if you're really trying to put some information at the hands of the employees or the customer, you are able to create an interface that's much more acceptable, much more easy, reduces a lot of the friction points using the underlying data if and this is a big if and when we're able to fix the data that's underlying it right and we're able to use that data to train our models and we have sophisticated reasoning engines to make sure that we're actually. 

Using the data appropriately underneath. So, there's a lot of ifs of how this isn't going to blow up SAS as we know it in the next year or two. It really is not; it is going to change the way we potentially interact with corporate data and corporate systems.  

Eric Hanselman: Well, and, and I guess the question to maybe expand on this little. 

I mean, most organizations have got a customer relationship management system, an enterprise resource planning system, marketing systems, and they've got, you know, there's a lot of SaaS use in CRM, ERP, marketing.  

Sheryl Kingstone: All of which isn't going away in the near future. And  

Eric Hanselman: so, this is the question, is that they're still the gatekeepers of access to the data because the data is still in the SaaS environment, right? 

Sheryl Kingstone: Not only is the data still in the SaaS environment, think about the workflows. And the processes, let's look at CRM is the system of record for a lot of your customer data. You got marketing applications that a lot of the execution layer of campaigns and revenue and digital engagement, right? Even CRM is now more about bringing together structured and unstructured data. 

ERP is mission critical as it relates to getting things done and manufactured Accounted for. Do you need to have a UI layer that goes on top of that just to get day to day inventory measured? No, but do you want an agentic layer to tell you that inventory is getting low, and the supply chain is getting crazy because if we won't go there? 

And then, you know, we've got some frustration on customer service and support layers, right? On the call. Yeah, there's absolute use cases and they complement each other.  

Eric Hanselman: Well, and Chris, you've been talking about some of this shift moving beyond systems of record and actually getting into the systems of intelligence. 

What do you see as that transition?  

Chris Marsh: Yeah, I mean, this is something we've been sort of charting in our research for a good amount of time, probably a decade at this point, which Again, 20 plus years of SaaS, loads of new options, benefits, opportunities. But with that, Kay, has come, you know, huge sprawl and massive friction, you know, in terms of employees trying to get things done, working across a growing estate of applications. 

And so, I would say it was like a couple of years, maybe three years before the pandemic period. So, we're looking at like 2016, 2017. We began to see in our research that actually this was becoming what was considered by Enterprise leadership as a strategic challenge, right? There's now so much context switching now, so much friction that actually this is impeding us executing on core business objectives. 

And so, the sort of technology. market began to respond with more consolidation, unification of different functionalities that may be existed as separate SAS categories into newer types of enterprise platform that in our research we've called systems of delivery and others say systems of intelligence, which is new types of enterprise platform that are emerging in response to that, you know, just SAS. 

You know, sass sprawl. And that's a real thing. We see in our employee surveys that when we ask them, what are your sort of main technology obstacles around two thirds, I think it's just over 60 percent say some variation of app sprawl, right? It's too many apps, too big a learning curve to learn all these apps, can't get the reporting across these apps because data's splayed in lots of different places, can't collaborate easily across these apps. 

And in response to that, we're seeing the growth of like ops teams spread across different business functions as a way to try and expedite things like app integration and data governance and process automation that historically may have sat with IT. But increasingly, technology is enabling sort of low code, no code individuals to really do some of this work and take that brunt off IT to try and expedite what they're trying to do within their particular business function. 

I think that, so there's that sort of trend. But there's also, as Sheryl has alluded to already, the general trend of, like, SaaS companies creating more interoperable front ends that can connect with various back ends and vice versa, you know, infrastructure vendors making choices to support different types of UI and interactive model. 

So architecturally, you could say that we're already somewhere along the way of sort of decoupling these various parts of the stack that would be required for agentic AI and ultimately further down the line, the implementation of like multi agent architectures.  

Eric Hanselman: Well, if we start getting to a point at which the agentic part of this environment is the new front end, it seems like we get to that question of who owns it. 

And it is this another, is this one more consolidation pressure, is this one more piece, or is this something where we get towards a multiplicity of agents tasked towards individual functions around this?  

Sheryl Kingstone: How do you see that shaking out? It's a very good question, especially the one about who owns it, right? 

And today, even in the SaaS market, no one owns it, right? You got a combination of it. And line of business and the, you know, security officers and the chief information officers, and it's very fragmented even for your traditional applications, right? Your CRM, your marketing applications, your AI applications, wherever it's trying to go. 

So, I like to look at it is, do we have to worry about who's owning it? And what we have to look at is. What is the use case, where are we trying to go, and which ones are going to lead, right? So, we did survey in our digital maturity models, I mean our digital maturity survey, where businesses are prioritizing, and one of the things I wanted to do was really take a look at are businesses starting out more agentic AI in marketing? 

Versus IT help desk versus customer service and support versus sales versus HR versus some training and development versus internal finance right and when you stack rank that. It really does come up aligned with some of the other surveys that I had, you know, top use cases right now where people are starting are a little bit more internal, right? 

How we're working with marketing optimization to, you know, autonomous campaigns that help analyze data and identify different segments, right? Or employee help desk. There's a lot of great use cases there. And then customer service and support and even sales, right? It goes on and on and on. However, if you look at the data by the individual line of business owner, marketing wants theirs as top priority, IT wants theirs as top priority, customer service and support wants theirs, and so on and so forth. 

And so, it really comes down to how are we using, again, the right Data to drive the right models, because just because these organizations want to use it, the question really comes down is how are we training it? What are the use cases? And what are the reasoning engines that are going to drive it? And so, we're fundamentally, we are starting at more of where you're going to get the fastest time to value. 

So, a lot of times that's customer service and support on an agent assist, not necessarily agentic, meaning running self-service. And that's really where we are today. Beyond that, we are seeing things, you know, in the marketing optimization, in sales to help sales with engaging outbound approaches, you know, quantifying what is the right next lead, you know, and making sure they're able to handle objections, really things that are making the organization more productive, but that requires, again, a strong conversational interface, it requires a new way of looking at the reasoning engines, and it requires train the data. 

And  

Chris Marsh: I would, I would just maybe add on to that. I mean, there is that just wider challenge that, you know, not all systems are actually architected and designed for AI integration, right? So, it's just the usual legacy challenges that are going to come with this that precede, you know, basic conversations about what can we do, right? 

There's the, can we actually do it? There's also the point around, to, to bear in mind, right, in a lot of, certainly, compliance heavy industries, the legal, finance, and others, you know, there are core challenges here in terms of regulatory requirements, compliance requirements that demand strict explainability. 

And so, they require the kind of deterministic rules that currently exist in lots of SaaS apps that AI, with its more probabilistic nature, is not actually good at delivering. Right, so there's just a whole bunch of issues from a governance point of view that certainly very low tolerance as in low tolerance for risk industries, you know, it's going to be a deal breaker for them. 

We also know that from our current surveys, only around 20 percent of AI decision makers actually trust the accuracy of their AI models completely. And it's about the same for third party and homegrown models. And about another 40 percent say that they mostly trust them. You know, for many industries, and certainly again, those with sort of high compliance requirements, that inaccuracy is just not there yet to even, you know, think about putting this mainstream. 

So  

Eric Hanselman: we're, we're a ways from an agent figuring out that, wow, we've got a whole bunch of high stock of widgets right now. We should decrease the price and shift the sales team's compensation plan to go push that and send the emails automatically. Autonomously?  

Sheryl Kingstone: Absolutely not. Autonomously, absolutely not. 

But can we use more of the Gen AI and the Copilot or even some of the agentic AI reasoning for really solid use cases that can go straight to the bottom line? Think customer service support. Think employee or customer service agent productivity, right? Think helping solve So chatbots, so I go back to some of the data I have right now, right on. 

Let's just take a look at customer service and let's take a look at chatbots versus assisted channels. Well, right now the satisfaction on the assisted channels are much higher than your traditional chatbots. However, when I aggregate up even today, the poor experiences, meaning only 25 percent of consumers are actually satisfied. 

Very satisfied with the current chat bot scenario. We're saving 69 billion of call deflections just in the U S due to handling certain calls that are in a lower cost channel. Think agent assistant in a social service setting or in a chatbot setting. Now if we can improve the chatbot for more of the self-service to get it up to the same equivalent of satisfaction with a live call or assisted service call, that potential is huge. 

And it's going to skyrocket that 69 billion, potentially even doubling it.  

Eric Hanselman: Well, but you then get, I think, getting back to two of Chris's points, which is one, you know, the level of organizational trust and, you know, hey, maybe in the chat bot environments, if we can improve quality, you know, we do better there. 

But the other stumbling block seems to be, Chris, the other point that you'd made, which is that there are a whole set of these systems that are not really ready for AI integration. And it seems like that's a significant challenge into and of itself.  

Chris Marsh: Yeah, I mean, there's, I think this whole debate is interesting in the sense that it's making people think about some of these issues, but there are just so many things that are going to need to happen, not just over the near term, but over the medium term to actually make this a reality. 

Then, you know, it's good, it's good that we have a sort of stretch narrative here. But there's lots of, there's lots of things at play. I mean, you know, another one is that business logic isn't always so straightforward, right? It requires, in lots of certainly more complex processes, human in the loop. It may well require, you know, lots of different stages of approval and compliance checks and AI, you know, where we stand at the moment may be good to automate some of those things. 

But in terms of like autonomously, operating and doing the advanced reasoning. I don't think it's there yet to really replace the way that many processes have been instantiated in companies to date. So, lots of just legacy, unsurprisingly, lots of legacy technology challenges that are going to need to be thought through to actually, to actually get there. 

And that is going to put pressure on, you know, SAS companies to do a bunch of things, right. To, you know, rethink their interfaces, to rethink their interoperability, too. Really think about what the strategic workflow automation capabilities are, you know, how that aligns with key use cases, key personas. How do they sort of identify opportunities to integrate AI even if not generative to actually begin to bridge into a world of, of mainstream gen AI and assistants and agents. 

So, I think one of the good things about this debate and trying to stretch people to think about what needs to happen is that it's going to apply pressure on SAS companies to actually think about sequentially what are those steps that can get them sort of agent ready.  

Sheryl Kingstone: So, you asked the Eric, you asked the question as it relates to the legacy applications and a lot of it not ready. 

Right. And then Chris brought up some of the pressures on the SAS side, right. But let's not remember that a lot of the legacy applications is not even SAS.  

Chris Marsh: Right. Yeah,  

Sheryl Kingstone: it is a lot of the mainframe. So, a lot of the SAS vendors are actually spending a lot of R and D. Money today to open up their applications with API development to make sure that they're embedding a lot of the AI into their application today. 

And it's what's driving a lot of the upgrades. So, if you're already on the SAS, you just have to make sure that you haven't customized it to depth just like you did your client server applications in the past, right? To move forward with some of these newer. Releases that have been brought out in the sass mode, but let's not forget What you originally said is there's a lot of legacy out there and that legacy is not even sass The legacy is the old mainframe side, and that's where we're really trying to understand because they've already, you know, IT has already written custom APIs to get the data out of that into more of a data lake. 

That's really what we've spent the past decade on. And so now we're saying, okay, you're still going to get leapfrogged here, right? It's not just about getting the data out of it. What are we doing with the overlay? And so again, there's legacy and then there's. older SaaS applications. And then there's prior SaaS applications that need to be upgraded. 

So, there's so many different levels of where businesses are today, and it's not just about the SaaS market today.  

Eric Hanselman: So, looking ahead, is this just raising the pressures for consolidation to a new level of abstraction? Is this something because we saw You know, the CRM vendors all bought up a whole bunch of marketing companies and, and integrated them, but then there were still gaps. 

And so marketing companies still exist on their own. Are we shifting to, uh, the next position of consolidation in which there's a race to be able to be the agent that rules them all, or is there a space for there to be an overlay or integration, or does this continue to be more specialized?  

Sheryl Kingstone: Well, it depends on your consolidation, right? 

Right now, you're looking at the consolidation as it relates to SAS providers. There's also consolidation as it relates to businesses buying applications and consolidating SAS contracts. Because what they have is duplicative, right? So, as we consolidate SAS contracts, we're trying to do it to save money and reduce complexity. 

Now you also do have consolidation on the SAS market and that's been going on for forever. You're right. A sales application acquired a marketing application, which acquired a service application, which created the next gen or commerce application, which created the tech giants we have today in the CX market. 

Right. And then all the vendors in the SAS are buying little Gen AI, predictive AI, data lakes, CDPs. to make sure that the underlying data can be aggregated. So, it really depends on what you mean by consolidation and who's doing the consolidating.  

Chris Marsh: Yeah, I was just going to say, I think it's also worth just setting the context that this isn't obviously a technology only Discussion, right? 

I mean, there's a whole regulatory element to it in terms of, you know, more maturity needed from a regulatory point of view for businesses to, you know, have the appropriate guardrails to invest in this and, you know, invest in the shift and. You know, with that, just culturally, I think more mature discussions need to happen around data governance amongst enterprise leadership to actually make the right kind of progress here. 

And, you know, there's always the stratification of those companies that are already more data driven and, you know, having those discussions, but really much of the market isn't certainly not in a mature sense. that's laying the groundwork for this shift. So there's all of the technology issues in terms of like how is this going to actually happen and what are the sequential steps but there's all the culture issues the change management issue and there's the regulatory context to that that's also going to need to be put in place which is just yet more things to think about as we you know as we sort of chart the direction of how this is going to evolve. 

Eric Hanselman: All the concerns about autonomous action or the implications of autonomous action, we keep getting back to, you know, we still haven't have well settled concerns about liability, all the governance concerns, Chris, that you're pointing out, just in terms of what we're doing with all this data, a lot to answer on a whole set of fronts, as you said, that are beyond the technology piece of this as well. 

Yeah, indeed. Well, a lot that we'll have to look at in terms of where this all goes. This has been a great discussion, but unfortunately, we are at time. Thank you both for a very interesting discussion and clearly much more to come.  

Sheryl Kingstone: Absolutely. Thank you, Eric. And we will be documenting all of this. So, look for the write up of it. 

Chris Marsh: Yeah. Thanks for having me on again.  

Eric Hanselman: And thanks to you both for being here, more conversations clearly to come, but that is it for this episode of Next in Tech. Thanks to our audience for staying with us and thanks to our production team, including Sophie Carr, Feranmi Adeoshun, and Kyra Smith on the marketing events teams and our agency partner. 

I hope you'll join us for our next episode where we'll be discussing something that's a little different. We're actually going to be taking a look at the NVIDIA earnings call and we'll be actually talking about that up front. Actually, bringing in one of our colleagues from Visible Alpha to get some perspectives on what we can expect. 

And we'll actually be doing that in an episode that drops on the earnings call date itself. So, I hope you'll join us then. Because there is always something Next in Tech.