What Companies Should Be Doing With AI and Agents Right Now

0:00Today on the AI Daily Brief, I am joined by Steve Chase, Vice Chair of Artificial Intelligence and digital innovation at KPMG. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. To join the conversation, follow the Discord link in our show notes.

0:18Hello, friends. Excited to do an interview show today, departing from our normal format. Steve Chase is someone I've gotten to know over the last year. He is deep in the AI space. His role is technically the vice chair of KPMG's Artificial Intelligence and Digital Innovation Organization, which, as they describe, involves leading a firm-wide initiative to integrate AI into every aspect of the business. And what I've seen and what makes Steve have a particularly interesting and unique perspective is that not only is he thinking about how AI helps KPMG do their jobs better, because of KPMG's role as a professional services company that works with clients who are also thinking about their own AI transformations, he almost gets to have a second perspective on the AI transformation as well that's informed by all those customers and conversations.

1:02Steve's background is on the consulting side as well, having previously led KPMG's U.S. consulting practice. We talk a lot in this conversation about lessons learned from 2024, the state of play for big enterprises in AI, and of course, quite a bit about agents. It's a great conversation, so without any further ado, let's dive in. All right, Steve, welcome to the AI Daily Brief. How are you doing? I'm doing great. I'm excited to be here. Yeah, happy 2025. Lots of great stuff to talk about. But I'm sure we're going to run up against time very quickly. But for background context, for people who aren't familiar, I gave a little bit of an intro to you before.

1:39But just tell us a little bit about your role at KPMG, the landscape or how that helps you kind of understand the landscape as we dig into the conversation. Yeah. So my title is the vice chair of AI and digital innovation. And I've had a lot of different jobs at KPMG. I really love this particular one at this particular moment. My job is to think about AI and other digital transformation or other digital capabilities that can lead to transformation both of our business and our clients business and help navigate that journey that we're going through to take advantage of these new innovations at scale and at pace.

2:14Yeah. So one of the things that I think makes your perspective extra valuable here is it's almost like a two for one, because you're thinking about AI innovation transformation in the context both of this big, large company that you're working with, as well as the plethora of clients that that company has to go serve, which is a little bit of a different stance than I think a lot of folks who just have one of those perspectives. Yeah, I mean, listen, I think that it's partly because professional services firms are, you know, have a lot of opportunity with this technology. So a lot of our clients are also interested in our journey.

2:46I wouldn't say they've always been interested in what we're doing internally, but they're really interested right now because, you know, we are, as we say, we're heavily impacted by the opportunity, both potentially positively and potentially negatively. We've certainly seen that. So then as others are thinking about framing their journey, they're very interested in how we're doing our journey as well. So it's been very symbiotic in that way. Yeah, this is great. So I think most of our conversation is going to be very forward-looking, but we're still in January. I think it's a good chance to kind of broadly catch up a little bit.

3:24Overall, what were your biggest learnings around AI, whether it's in the context of the enterprise or just generative AI more broadly last year? And, you know, maybe maybe to get refined, is there anything that you thought about generative AI coming into 2024 that didn't play out the way that you had imagined? I'll give you a market perspective. I actually thought there was going to be more regulatory response, some more court cases and other things that didn't happen last year. We were sort of expecting the emergence of some type of AI certifications or other things that just – and more in the enterprise.

4:01I'm, you know, I'm going to be more in the enterprise space, but like enterprise would be asking for that or or whatnot. And that just really it wasn't something that we had seen on a like a little bit more sort of like something I experienced. You know, I always say we were giving access to supercomputers on people's desktops. And I've been surprised, even though it's it's something that we've been we're talking about right in the beginning about how hard people change is. I've been surprised by the resistance to like people picking that up and just immediately gravitating to using it. And so, you know, it's like it seems so self-evident.

4:38And yet there's a resistance in there. And so kind of leads you to that narrative around enterprise AI is adoption is really challenging, right? It's really challenging topic. Yeah, it's fascinating. I mean, you know, this is something that you and I have talked about in the past. And obviously, we think about a lot. But I think that to the extent that there's a positive with it, I think that it's a great reminder that as fast as the technology is moving, there is a certain amount of human, social, and organizational inertia that gives us a little bit of a chance to catch our breath and adjust.

5:10The upside of what's challenging for a lot of folks who are trying to drive adoption and new opportunity with this technology. Yeah, absolutely. Absolutely. Absolutely. It's interesting also that the number one request you hear from users in our client work and what have you is it needs to show up where I am. Like it needs to show up in my workflow. And it's funny because one of the things – probably the first presentation I gave on this topic was about how effective this was in sealing up gaps in between workflows, right? Like it's just so useful as a thought partner and what have you and to work on things where we haven't built systems.

5:47So it's funny. That's like a funny thing about like, well, you know, emerging in my workflow. Yeah. So this actually you segwayed great perfectly into a set of questions that I had around kind of the state of AI adoption in general. Yeah. You know, what are what are some of the biggest and most common challenges that you remain? Obviously, we're talking about this, this, you know, just the adoption itself, utilization. But is there anything beyond that that you guys see or encounter frequently that is holding things back or just is a challenge that a lot of organizations are facing? Well, there's a long list of those things, right?

6:25I mean, I suppose you'd expect a consultant to say that anyway. But here's a couple that jump to the top, right?

6:36is the technology moves so much faster than people are moving that anything that somebody believed about it, especially if they started out with it and then, oh, no, it didn't really do that much. Well, six months later, it's so much better. Keeping up with that and people, well, they just don't move as fast as the tech does. So that's one. Two, you inherit everything that's great and everything that was on your to-do list in these programs. This is enterprise transformation. It affects every part of the organization from the front office where you support customers all the way through the back office where you do your accounting and whatnot and the operations in between.

7:21And so when you do enterprise transformation, it's everything you're good at and maybe you wish you were better at. and one of the things a lot of companies wish they were better at is data, knowledge management, what have you, and AI is really valuable out of the box, but it gets a lot more valuable as you start to get your data into it and what have you. Then understanding the data use, the availability access. Most people are managing their data. They have it reasonably well managed. but not available to the AI. And then, you know, I think that we're talking about a new function around responsible use that needs to be very flexible, that helps create guardrails around what I'm going to do, why I'm going to do it, how I'm going to do it.

8:15I think that most organizations, that for them is legal and risk, but not opportunity. And like, you know, so there's just a mind shift change, Nathaniel, when you think about like, like, how am I going to think about ethical security, like the whole range of these things and design it into the program that I'm doing so that because because one of the I mean, I was going to finish on one of the number one things people just are are not always sure what they're allowed to do, even when you tell them, I want you to work on this. And, you know, I think sometimes it's because of the mixed messaging that comes up in terms of of the kind of training they get or what have you around data privacy and other things are super important issues that need to be need to be netted through.

8:59Those are the things that need to like those are the guardrails need to let you go faster, not bumps that try to slow you completely down. Yeah, I think so. I completely agree. In fact, I have long thought early last year. there was that interesting Microsoft LinkedIn study that found a huge portion of people that were using AI weren't telling their companies about it. Yeah, right. And to me, my speculation, and they had some data analysis around this, but my thought was that probably a big portion of that was people who were exactly in the situation that you just described, where they just weren't quite sure what they were and weren't allowed to do.

9:40And once you use these tools, boy, you do not want to go back to the old way that you did things. And I think a lot of people probably just, you know, they were trying to do it in good faith, but they weren't sure how to do it. And they just didn't want to be told that they weren't allowed to. I think it would maybe made some progress on that front, but it's still a big issue. I know you quoted our, um, the poll survey that we did like the last quarter. Um, you know, you highlighted that point. I think you highlighted that point, which is, is that the perception of leadership of the people who were interviewed, for our poll survey said that they, you know, that the executives were the main users of AI in their organizations.

10:15And that's not our take on it. Our take on it is, and all, you know, when we instrument the systems, the demographically, the more recent hires, the younger, you know, younger folks in the organization tend to be the ones that are actually the bigger users. So anyway, that's, but maybe not talking about it, right? Even in organizations where you've said, I want you to do this, you have to keep hammering the message and hammering the message. Because schools are really struggling with this. Well, AI is cheating. And I know you quote Professor Ethan Malik a lot, talking about, well, school can be better with AI in it.

11:02But yet, kids are petrified to use AI in school. Right. And then we get our interns come to KPMG coming out of college and they're like they're they're not really prepared right now. They're coming. They're like, wait a second. I'm told that this is cheating at school. And you're telling me we can't be successful if you don't use it. It's just an interesting interesting dichotomy. It's like in that that dislocation that this period that we're in right now. Yeah, it's a very, very liminal period. And those are historically the hardest for people to deal with those in betweens. Yeah. So this gets me to another question.

11:35And there may be no real answer for this. But is there, you know, two, a little over two years now from the starting gun on Gen.ai with the launch of ChatGPT. Is there a normal now? Is that, you know, the average place that an organization that you see is, is there table stakes in terms of where people are in the client journey? Or is it still so broadly distributed that it's hard to hard to pin that down? Well, I think it's like with most innovation and change, we're beginning to separate into a group of leaders, maybe fast followers and laggards, and where there were very few in the leader category, you know, ones who were moved past.

12:17and maybe put a little definition around that. Like so ones that have an actual strategy, they've put leadership in place, they've defined the program and its set of goals and outcomes that they're looking for, you know, and have moved into scaling. I think that, so I think there's starting to become some norms there, but I'm surprised in the same way that I am around adoption when you share with someone this supercomputer, I'm surprised by the number of companies saying, yeah, we're not doing anything with that right now. And the reasons they give are often regulatory and future regulatory issues and stuff.

12:58And I really think with the EU AI Act and some of the things around NIST and whatnot, I think we have enough clarity on what to do that you can't let that be an inhibitor to move forward. So table stakes for a normal organization, ubiquitous access to one or more AI capabilities in the form of some type of foundation model access. Um, something that is, um, you know, I would say a number of them beginning to move or look at, um, the addition of co-pilots or co-pilot, like, you know, personal augmentation, and then, uh, the emergence of rag solutions for, you know, some type of talk to your documents or, uh, now we call them knowledge assistance, um, that are, that are serving the single purpose.

13:51That's what that's that's table stakes. What is, you know, the majority and what's beginning to happen now is what you saw, like a number of folks being experiment with agents or agentic capability is starting to get on the, you know, on the other end of that. But listen, we're so early in the table stakes part of it in terms like the amount of that. Like, is is there a sense that they've gotten to the end of it? No, it's like all very much the beginning. It's not ubiquitous across everywhere. The so that's my take on it. And I suspect that that's consistent with whatever you talking about as well.

14:29Absolutely. Yeah, I think that's the case. Another way to frame it is I definitely think that there are starting to be leaders average and sort of people who are a little bit behind. But I think we're very early such that that could shift very quickly, given where the table stakes are, given how emergent things are. And I think that, you know, jumping ahead maybe to agents a little bit, which will probably be a big part of our conversation. Agents are almost a great equalizing force on that, again, because even the organizations that feel like they've done really hard work over the last two years to wrap their head around the co-pilot era and build systems and start to think about their data, which, as you pointed out, I was thinking about as you were talking, the survey that you guys put out last week, 85 % of people said that data was their biggest issue.

15:13That's very clearly something that's on people's minds. But even those folks who are thinking at that level, they still the vast majority of people have not even piloted an agent to say nothing of deploying. And so there's a moment now, and I think that this is a real opportunity for a lot of organizations that have perhaps felt behind to jump in with both feet to really start figuring this out. Before we move off this point, though, I just want to reflect something you say a lot, but we believe as well is, you know, that idea of leaders and average and laggards, as we use that phrasing, there's a period of awareness and cultural change and training and training people to be different kinds of leaders and different kinds.

15:58Like that work you're going to do at some point in your journey. And there's no it's a no regrets move to get started on that ASAP. Right. I mean, that is that is clear. The other thing is, is this get a mindset that this is transforming your business. You have to be thinking about your processes differently. And I think the last time we probably did work like that, maybe a lot of the leaders weren't even in the workforce. You know, it was like late 90s. We were last really thinking about rewiring the enterprise like this into a business mall. It didn't exist. Most of the stuff after that, you know, built on that concept, but it took us, what, 10 years around that, you know?

16:35So we're trying to do that all much faster. So I just there's no regrets moves in here, improving data, getting the training and stuff rolling. Even if you believe agents, well, agents are going to and I do believe this agents will actually be adopted at a faster rate and what have you. for the employees that aren't agent, that aren't, you know, for the work that they're doing where they're not working with agents, all this AI work will still have been, this generative AI will still be an incredibly useful thought partner and an incredibly useful part of that enablement of the workforce in the future as well.

17:12So no regrets moves there. Yeah, I mean, I think that's a positive thing just to really double click on. there's very low downside to almost everything that organizations could be doing, trying, experimenting with right now relative to where the organization needs to be in five, 10 years. Let's talk about agents. So is 2025, let's start broad. Is 2025 the year of agents? How do you think about this? Okay. If what we were talking about with generative AI was still early innings, We're starting the game right on agents. Agents will be I think it agents will be the dominant topic for the next 18 months.

17:54There'll probably be a hype cycle around it where folks go through like it's way overhyped right now for what it's going to deliver this year in 25 would be my guess. But it will begin to be, you know, and then it'll go through that period. It's probably underhyped how big of a change this is really going to be in the enterprise, my suspicion. I think that this is going to be easier to adopt than other technologies in the AI space. It's going to be easier to adopt than generative AI because it's going to materialize right in the workflows and or as a teammate that sits beside it. So anyway, yes, I think it's the year of the agent.

18:37I don't think it's the last year of the agent. And I think, you know, we'll definitely start to see those counterfactuals like reports. Oh, agents were overblown. I suspect that'll be my prediction for 2025 is in the in October timeframe. Someone's gonna be start writing the agents were overblown article. Yeah. It's interesting, though. So one of the more interesting stats that I saw towards the end of last year came from this Menlo Enterprise report. And it was about the shift in buy versus build behavior. And so they found that in 23, the organizations that they surveyed, it was 80 by 20 build in terms of the Gen AI solutions they deployed, where last year it was 53-47, so almost half and half, so still slightly more buying than building.

19:24And I think that part of what it reflected was increase in confidence plus a recognition as they dove in that there were certain types of applications, maybe vertical applications, applications that use their specific data or that were specific to their industry that just weren't available yet. I think it's going to boomerang. And now we're seeing all of those sort of vertical solutions start to come on. And the thing that I wonder as relates to that question of overhyped, I think it's almost inevitable. I think it's probably hard to deny. But at the same time, I do think that for an inning zero type of situation, the enterprises that are going into this are a lot more sophisticated than they have been in the past at inning zero in terms of where they get.

20:09And so it may be that just the deployments and the experiments are more modest. And it's just what they find when they dig in that they actually can't be as hyped as they had hoped for. here's one of the reasons why i think i here's any here's a reason why i'm at i entirely agree with that point i know i just uh claimed the overhype but but like there's so much hype that it's it's kind of inevitable that someone write that article right but what's different here is the emergence of agents inside the enterprise platforms we were always talking about 2025 was going to be the year ai began to materialize in the enterprise software and they've been working diligently to do that.

20:51And that's now taken the form of agents, right? So that seems to be the dominant way that they're thinking about that that'll materialize. And because of the investments others have made in cloud or SaaS-based systems and what have you, it makes it a lot easier to adopt these things, right? And so I think that'll be very compelling to IT departments. I know a lot of the clients we're talking to that's like, this part of AI is not complex for them to think about and understand they they get it they get why it's going to matter i don't think that everybody's fully understood like i don't think there's as much sense like i don't think people because they haven't really had experience them yet this idea that they're going to be teammates and how they're going to get invoked and they're going to work and that's going to require training about like how you work with them and work effectively with them there's a whole management layer that i think is interesting i don't think that's actually going to happen in it it's going going to happen somewhere else in the organization.

21:46But anyway, I think it'll be easier. The enterprise side will be easier. Those are the people that are those are the capabilities that materialize in the platforms. And then there's the you talked about like the enterprise wides or the vertical agents that are going to materialize that work on the platforms, right? Like that work across the platforms. I think those will be later. Like I don't think those becoming in scale and especially where autonomy comes in, you know, uh, so we're really focused this year on, on, on the enterprise agents as, as part of it, but also you need to be making your experiments in orchestration and, um, um, and the outside the system agents as well, because those would be really useful too.

22:31And there's a lot of third parties that are offering these, you know, you guys, and I think in one of the last podcasts I listened to, uh, you, um, You guys were talking about CRAI, and there's a variety of those like that, that seem like they're providing pretty good insight into where things are headed. Yeah. Well, it's interesting, too. One of the things that is fascinating about this generation of agent companies is I think they have an appreciation for, you know, one of the things that you said, which I also very strongly agree with, is underestimating in the long run just how different, how disruptive, how much they're going to change the shape of organizations.

23:06You see a little bit of a hint of this in that the way that this software is being deployed inside companies is these startups are all basically deploying engineers, kind of Palantir style from a decade ago, where they're sitting with inside companies doing this stuff. It's so different than point and click kind of software that we've had in the past, which I think is a reflection of just how big the change might be. Yeah, it's interesting, right? And so how systems integrators or business consultants interface with that as well and, like, come into that ecosystem. Because if it's only that they can deploy, which seems to be – which has been part of the model in the beginning, then the scale, which is – it can't scale fast enough, right?

23:55And so that's why we're having those conversations and talking about – and then, of course, you know, we develop those same sort of capabilities in our own platforms. like in our tax platforms and what have you. And so trying to learn that lesson too about how we bring that effectively into those departments has been really useful as well. On the idea of just the magnitude of disruption, I don't know if you caught this, but in his CES keynote, NVIDIA CEO Jensen, he said that IT is the new HR department for agents, which I thought was a fascinating way of, obviously a little bit provocative just to get people talking, but a really fascinating way to think about agents as a new portion of your workforce are these, you have digital workers now and you have human workers and you have a combination where they'll meet and that just, you know, it's a totally new management structure on top of just a totally new technology structure.

24:51Yeah. And far be it for me to disagree with Jensen. All I would say is that it's not entirely obvious to me that IT is the right landing zone for that, right? I mean, And in fact, we've been having a lot of conversations with HR departments that, you know, just like you manage contractors, employees and other labor, why wouldn't the HR systems and HR processes actually be really well tuned to the agent space? which where because I need to onboard them, I need to train them, I need to I need to think about their performance. I need to give them performance reviews. I probably need some type of management technique over top of them that may be different.

25:33So you could argue that that actually the HR systems themselves are well constructed to think about all the different things that I need to go do with agents. So but I think it's gonna be fascinating to watch how this how this works. Right. Yeah, no, I think about this a lot. And I do think one of the sneaky second order effects of AI and will be extended with agents are certain parts of the organization having even more significance attached to them because of this new set of skills they need to develop. So HR actually evolving the capability to manage agents and help individuals manage agents becomes a very different proposition that's really valuable.

26:12I think about this with L &D as well. Learning and development for organizations are all different, but some organizations kind of treat it as this. It's basically a perk, right? It's a thing where they're investing in their people being better in the future, and so go off and do your thing. I think L &D is increasing in importance as these skills are now really mission critical to the organization in an immediate timescale. And I think that it's reflective just of the sort of broad management changes that these technologies are bringing with them. So one of the things I was thinking when I was listening to that speech was, if you extend that, IT has the potential to be like everything in the whole organization becomes IT.