Redesigning Processes for the Age of AI Agents

[music]

0:00:03.4 Satyen Sangani: Welcome back to Data Radicals. Today, we're talking with Tom Davenport, one of the world's top experts in data and AI. We'll dive into Tom's bold prediction, that agentic AI will be the biggest trend of 2025. Could AI really replace traditional software? What does that mean for Chief Data Officers? And how is generative AI shaking up industries like healthcare? This episode is packed with insights on the future of AI, what's hype, what's real, and what every data leader needs to know. Stick around.

0:00:36.6 Producer: This podcast is brought to you by Alation. A platform that delivers trusted data. AI creators know, you can't have trusted AI without trusted data. Today, our customers use Alation to build game changing AI solutions that streamline productivity, and improve the customer experience. Learn more about Alation at alation.com.

0:00:58.2 Satyen Sangani: Today on Data Radicals, I'm joined by Tom Davenport. Distinguished professor of Information Technology and Management at Babson College. Tom is the co-founder of the International Institute for Analytics, Fellow at the MIT Initiative on Digital Economy, and Senior Advisor to Deloitte Analytics. Tom has authored and edited 23 books and is a regular contributor to the Wall Street Journal and Forbes. As the pioneer of competing on analytics, he's been named one of the 100 most influential people in the IT industry by Ziff Davis. And one of the reasons that I'm personally most excited to speak to Tom, is that he's seen data over multiple generations and is just a natural teacher and sharer of knowledge. And so, I know everybody that's listening is gonna be super excited to sort of glean a lot of that from this interview. Tom, welcome to Data Radicals.

0:01:46.8 Tom Davenport: Happy to be here. Thanks for having me, Satyen.

0:01:49.4 Satyen Sangani: So, let's maybe jump in. You recently authored an article in the MIT Sloan rag that basically was about five trends in data science and AI. And you didn't stay away from sort of some strong takes. But one that I think a lot of people would at least talk about or agree on is around agentic AI. You basically said that agentic AI is gonna be the most significant trend for 2025.

0:02:10.2 Tom Davenport: I think I was the only person in the world who said something like that.

0:02:14.2 Satyen Sangani: Yeah, you're probably the only guy talking about AI. Yeah, exactly. But maybe say more, dig into where it's working, where maybe it's not working, which jobs to be done are gonna be augmented, and which ones are probably a little bit farther away.

Agentic AI use cases: Is AI the new software?

0:02:27.2 Tom Davenport: Well, as you know, it's very early days still for agentic and most of the enterprises that I talk to about it say they're either not using it yet, or they're using it for quite simple, relatively inconsequential things. Somebody in an insurance company said, we certainly wouldn't use it for underwriting or claims, but we might use it for employees registering their PTO preferences or something along those lines. 

So, I think we still have to figure out what it means to have an agent that is probabilistic and sometimes picks the wrong word or does the wrong thing. And there may be, I think a return to some deterministic aspects of AI with rules or something like that, that guide the workflow. But I think we'll see a gradual introduction. 

It's funny some very well-known leaders of software companies, one of them Microsoft, has said, "Oh, this is the new software, it'll replace spreadsheets." But I think it's unlikely that Microsoft will take Excel off the market or that the vast majority of people who use it, will give it up anytime soon. So it's like the installed base is quite large.

0:03:54.9 Satyen Sangani: Yeah, there's a lot of people who do things with software as software is obviously defined. Where do you stand on this question of like, there's been these debates of sort of software is dead, AI is the new software. Where do you stand in terms of that question and how quickly do you think any transformation will happen? 

0:04:14.5 Tom Davenport: Well, I think AI may write a lot of the software, but I think there will still be a place for custom software. There will still be a place for packaged software from vendors for transaction systems and so on. So I think it's a bit much to say, it's going to somehow replace software. And it is software itself, of course. 

And I think we need to remember that this has been around for a while in some form or another. Last week, I was meeting with some people from the AGI Group at Amazon last week. And they reminded me that Alexa/Echo was an agent. It could actually do things like turn on your lights or turn off your heat or set up your alarm system. And they said, the really challenging thing is being good at both doing things and knowing things. And they're working on the next version of Alexa to do that. So I don't think we can minimize the challenges involved in making really smart agents who also do things well.

0:05:24.6 Satyen Sangani: Yeah, this is pretty funny because what the users don't know is that, while we were recording this interview, there was an upgrade agent on my software that literally turned off and shut down Chrome just so it could update. And what I love about this example is that, had it known any context about who I was talking to or what I was doing or looking at my calendar, it might not have actually done that work. And so, there's so many edge cases in the real world, where these agents have to be aware of context and somewhat sensitive to it. And I think the promise is amazing, but the reality of it, is gonna take time to evolve. Are there particularly novel use cases of agents that you've seen Tom, that are ones that might have been farther out than you would have expected? 

0:06:10.6 Tom Davenport: Well, there's a company based in Toronto that I've done a lot of work with, that is very document-oriented, particularly financial documents and documents of record for a variety of companies. And, they've been using it quite successfully to extract information from all types of documents and really turn documents into data so that they're much more useful and can be reused and repurposed in various ways. And so, they said it's really transformed their business that they can do this and makes it much, much easier to create the software environments to make something like that happen. So I don't know how creative it is necessarily, but it's certainly useful.

0:06:56.9 Satyen Sangani: No, that's pretty foundational. The act of sort of taking, I guess unstructured data, whether it's a video or a document or an audio file, and turning it into something that's structured and turning into something that's got attributes and labels and metadata with it, is on some level like the act of analytics. Because once you've made those decisions as to sort of what to count or what that thing is about, then you can start drawing some questions and conclusions about it. So that, I think seems to be one of the most interesting classes of software that's out there. 

Which gets to this kind of broader question of sort of unstructured data. Most data analytics professionals would have defined their job in the realm of structured data, counting things in business and processes systems and operational systems. Do you think that job is changing now? Do you think it's more about other stuff? 

The rise of AI, unstructured data – and what it means for knowledge management and data culture

0:07:46.9 Tom Davenport: I do, and I think we have generative AI to thank for that. And I'm old enough to have been quite involved in the knowledge management movement, 20 or 25 years ago. And that was all about unstructured data as well. And it turned out that the tools we had for managing that were not quite up to the task. It was very labor intensive to sort of figure out what was important about a piece of knowledge and capture it and make it accessible. Generative AI is quite good at it, but your data have to be in pretty good shape. It's not going to tell you, for example, if you've got 10 different proposals for, I don't know, putting in SAP in the energy industry, which one is the best and which one ought to be the model that has the content that's really best for the rest of the organization to draw upon. So humans are probably gonna have to do some of that. Gen AI may help a little bit and give you an opinion, but humans certainly need to validate it.

0:08:53.9 Satyen Sangani: Yeah, for sure. If you think about sort of this validation question, the last five or 10 years, people have been talking a lot, we've been talking a lot about this idea of self-service and even more broadly this idea of building sort of data-oriented cultures. You mentioned this idea of knowledge management which is really about sort of capturing what the organization knows and about building good habits. But one of the things that you mentioned in the article, is that the interest in sort of these data cultures is waning and has cooled. What led you to that conclusion and can you say more about that? 

0:09:26.0 Tom Davenport: Well, it's a sort of a positive and a negative story. This is a survey that Randy Bean does every year of data leaders in large organizations and every year for the, I don't know, past five or six, anyway, I've with him written a forward and a kind of overview of it. So I paid a lot of attention to it. And one of the things that was the bad news piece was that every year, he would ask how many of you think your organization is data-driven or do you have a data-driven culture? And the numbers would be quite low, typically in the low 20s, saying that their organization had those characteristics. And then all of a sudden in 2023, he generally surveys at the end of the year. I think he called it the 2024 survey. But things doubled basically. We're up to almost 50% for no apparent reason other than Gen AI coming along and getting everybody excited about data and AI and so on. This year it settled back a bit. Not all the way back to the earlier days, but kind of halfway between. So that suggests, I think that our love affair with gen AI, might not be enough to dramatically change the data cultures of organizations. We probably won't know for sure for another couple of years when we look at future surveys.

The CDO's role in the age of AI

0:10:52.5 Satyen Sangani: And do you see, I guess there's this big debate around the CDO and the longevity of the CEO, the role of the CDO, whether they should continue to exist. Talked about it with other guests on the podcast and some folks are saying, "Well look, the CDO is getting encompassed or enveloped by the CIO." And then there's this question of like, well, for data professionals, there's sort of good data practice, good data discipline, there are knowledge management practices in the world of data. But it feels like there's, to your point, a little bit of waning around some of that work. How do you see that evolving? Do you think that's a temporary trend? Or is that something that data professionals need to be more mindful of, moving forward? 

0:11:31.6 Tom Davenport: I have a bit of disagreement with my co-author of this article. I just already mentioned him Randy Bean, who believes that Chief Data Officer role is very critical. It's critical that it's a C-level role. It's critical that it's a business role rather than an IT-oriented role. And I just don't think that's very feasible. I've done another survey of senior tech people in organizations. And I was very surprised to learn, that they think there are too many of these C-level tech roles. And that causes a lot of confusion in organizations. When you have a CIO or Chief Data Officer, Chief Digital Officer. We don't even have a unique acronym for digital and Data Officers. The Chief Technology Officer, the Chief Analytics Officer, blah blah, blah. So I think a better future is if you have a very business-driven executive who gets along well with other senior people in the organization, then you can have one person who's likely to report to the CEO and oversee all of these things. Many of the functions would continue as before. They just might not be called, CDO or Chief Digital Officer or whatever.

0:12:49.8 Tom Davenport: And there are a number of those already happening, particularly in financial services. They manage all of those tech functions and in many cases even things like customer service or operations in addition, and they're very much a part of the senior management team. So, I think that kind of senior spokesperson for all of these issues can really have a big impact on making the culture more data-driven.

0:13:17.5 Satyen Sangani: And what does that look like? In this idea of making cultures data-driven, there has to be sort of this buy-in. You have to start measuring things that aren't the outcomes of the business as an intermediate sort of touch point. And then often that's where people get lost because they're like, well we're maybe more data-driven, but maybe it doesn't actually impact it, or maybe we don't even know how to measure being data-driven. In this new world where everything's moving to AI and there seems to be some consolidation to the CIO, how does that spokesman show up? 

0:13:44.4 Tom Davenport: What do they do day to day? 

0:13:45.5 Satyen Sangani: Yeah.

0:13:47.2 Tom Davenport: Yeah. I think it's gotta be multifaceted. And to be honest, I think a lot of the early efforts were not terribly impressive. They're kind of like what we did to make people more DEI-oriented. You watch a video and you take some questions afterwards to see, did you pay attention to that particular video or instructional medium. And obviously that doesn't really change much if you spend a few minutes on any subject. And a lot of companies did that with data as an important asset, blah, blah, blah, but didn't get into any depth. 

So I think from an educational standpoint, you have to pay a lot of attention to what's the context, what business function is the intended recipient from, what's the level of that person. And if it's a senior person and you can justify it, maybe it should be one-to-one instruction. So that's just on the education side, we need to put a lot more effort into it. On the issue of marketing data initiatives, I think most senior data executives ought to have a marketing person whose job it is to kind of trumpet the improvements that have been made in data around the organization and showcase great examples of work with data and analytics and AI.

0:15:13.1 Tom Davenport: I think there's probably establishing some new organizational structures for data as a part of this. Ideally, it's always been hard, but getting people in the business side to be data stewards or at a minimum to create roles like data product managers or digital product managers or something like that. That oversee a broad range of activities that are trying to create useful products for internal or external customers rather than just an IT project.

What is a data product and a data product manager?

0:15:46.8 Satyen Sangani: Yeah, I think the data product manager outcome, we believe that's a pretty strong direction. Because, on some level, most data gets created as a result of some business process. And so it's exhaust or it's an externality to something that somebody else is trying to do. But then the idea of a data product manager, is basically to create something that's more usable and that can be reused for a variety of different purposes. And at least that gives a thrust to that individual that they've got a job and that job is to make this data reusable, which and many times what happens in data initiatives is they get lost. And talking about knowledge and cultures is often soft, but it doesn't necessarily always produce business value.

0:16:27.6 Tom Davenport: Yeah, I agree. Although, we have a terminology challenge in that. For a long time, I talked about data products and data product managers and included the analytics and AI that makes data meaningful as a part of that concept and that job. Now I've kind of concluded, well, if we do that, then it's confusing to people. Maybe data products should just be those reusable assets and a digital product can be the combination of all of those things coming together that make it useful. 

But I did a survey and about half of the respondents, so this is again, the senior IT people thought a data product was just the data, and the other half thought it was the data plus the analytics or the AI. So it's just confusing. I think or has been in the past.

0:17:21.2 Satyen Sangani: Yeah, it's super confusing. And there are some people who have that very rigid. Well, there's like a continuum that we've seen and actually one of our product managers running a blog on it, where on some people just think of it as a collection of data things and then other people think of it as very rigidly typed tables that are very well defined and semantically consistent and have data contracts backing them. And so there's sort of this continuum of definitions. The thing that we're ending up on is like all of those things can be true, but ultimately what really matters is reuse.

0:17:55.0 Tom Davenport: Yeah, use in the first place and then reuse after that. And yes, I do agree with that. And I think whatever you call them, your data assets ought to be reusable and publicized and you should make it easier for people to reuse than to create from scratch, typically replicating previous activities. And I think that's a very important part of not only data management, but your productivity with analytics and AI as well.

How will AI impact the data industry?

0:18:27.0 Satyen Sangani: So what happens then to the data industry, in your view, over the next... We've seen a proliferation of firms, we've seen tons of companies founded. These marketscapes are large. How do you see that changing in the next five to 10 years? Do you see fewer firms, more firms? How does this all play out? 

0:18:45.9 Tom Davenport: My guess is that there will be a lot of consolidation. AI has driven, in particular, generative AI has driven a lot of that proration over the last several years. And I think we're starting to realize that generative AI, while amazing and powerful and incredibly useful under the right circumstances, is not going to be able to drive our economy the way it seems to have done over the past year or two. I like some of these analyses that say to justify the amount of money flowing into the AI hardware sector alone, would require between 600 and $650 billion in savings on the part of the companies that use it.

0:19:35.6 Tom Davenport: And we're definitely not seeing that. And so, I think that means some sort of retreat. I don't think we're going to swing into a full-fledged AI winter by any means. And I don't own stock in companies like Nvidia on their own. I certainly, you can't avoid it if you own any kind of index fund. I haven't made sure that I've gotten rid of it all, but I think there's gotta be some retreat.

0:20:01.8 Satyen Sangani: Yeah, for sure. I guess like any Hype Cycle, there's sort of this peak of inflated expectations, and where we, it sounds like you're saying though, you feel like we may be past that peak and now starting to approach whatever the down cycle might be? 

0:20:17.0 Tom Davenport: And whether it's the trough of despair or disillusionment, I don't think we're there. We may not go all the way there, but yeah, I think we're just past peak. And a number of, there's still a lot of debates about this, I love within Goldman Sachs, you have some analysts saying, "Oh no, it's going to cause massive amounts of productivity gain and economic growth." And others are saying, "No way is that gonna happen." They can't even agree within one firm. And you have economists who can't agree at all. So, I don't know exactly, but it seems to me that things are getting a bit inflated.

What is AGI? Is it coming soon?

0:20:57.4 Satyen Sangani: Yeah. And at the same time, all of the sort of AI creators seem to be talking about sort of this AGI being imminent within years. I don't know what the Amazon team that you spoke with thought, but would love to get your perspective on that. And first of all, what does even AGI mean? 'Cause a lot of people have different definitions for what this sort of fully automated intelligence is. And then the second is, how close do you see us being based upon your survey of the actors that you talk to? 

0:21:29.8 Tom Davenport: Yeah. Well it's, I think, quite ironic that we have all these conversations about when we'll achieve AGI when we haven't even defined human intelligence very well. And I think it probably plays into the hands of the vendors who want to claim, yeah we're just about to hit AGI. Such a poor definition, it's relatively easy to say that some aspect of it has certainly been achieved. But I believe that we won't hit AGI with generative AI alone. I think we'll need something that is able to make sense of the world better than a predictive word model can do right now. That I don't know exactly what that is. There's people with varying philosophies. It seems unlikely that we'll go back to logic as a driver of AGI. But something closer to human intelligence, in which young children can make better sense of what's going on in the world than even the most capable generative AI model.

0:22:37.6 Satyen Sangani: Yeah. So given these trends and given what you're observing in the market and your own work, where are you spending most of your time and energy in research? 

What is analytical AI? How does it differ from generative AI?

0:22:45.9 Tom Davenport: Well, I've always really focused less on what vendors are doing and more on what their customers are doing, with those capabilities. And so, I’ve spent a lot of time talking to companies about what they're doing with generative AI, how they're trying to create value out of it, how it affects particular industries. I'm particularly interested in healthcare because I think there's a massive amount of potential for improving healthcare with AI. I'm also trying to rectify what I think has become a bit of an imbalance between focusing on generative AI and other forms. I can't stand this idea of Legacy AI or traditional AI. So I call it analytical AI. There's no perfect term for it because of course generative AI is analytical as well. But the purpose of generative AI is not to analyze data, it's to create content of course. So I wrote a piece recently on analytical with Peter High, a consultant to IT organizations on analytical AI versus generative AI. And mostly, people have been around the industry for a while saying, "This has really gotten out of hand." And for a lot of organizations, analytical AI could even be more valuable than generative AI.

0:24:03.8 Tom Davenport: I'm always interested in, what does this mean for how organizations manage and the organizational structures as we were talking about before, the Chief Data and analytics officers and I've been interested for a long time in this issue of, "If this technology is so impressive in the realm of marketing and personalization and one to one approaches to customers, why do I still get so much crap in my inbox and on my mobile phone screen and so on?" 

And so I just finished the first draft anyway of a manuscript with Jim Sterne, on how do we really fulfill the promise of one-to-one marketing which was defined 30 or so years ago, but for most of us hasn't really come true yet. And will generative AI be the final straw that makes this all possible? And I think it will certainly help. I doubt that it will be enough in itself.

0:25:09.2 Satyen Sangani: And what do we need to do to fulfill that promise then in your view? 

0:25:13.5 Tom Davenport: Well, I think it really becomes a combination. It's doing things with managing both your structured data and your unstructured data relative to customers. It's just the idea that if you're only interested in selling more to your customer, you're probably not going to be successful because that will lead to excesses of trying to sell them things that they really don't need or want. You have to be genuinely interested in improving the lives of your customers if you're I think going to be successful with this. So there's kind of a purpose element to it. Lots of different analytical technologies, generative, analytical AI, even going back to more plain analytics, in a sens,e would be an improvement for a lot of organizations. And multiple channels, of course, because everybody interacts with multiple channels and I think some awareness. I wrote a book a long time ago, was one of my worst selling books sadly. It was called The Attention Economy and it argued that this was in 2001.

0:26:21.7 Satyen Sangani: The irony of that is great.

0:26:24.0 Tom Davenport: Yeah, came out the day before 9/11 and somehow the world's attention went otherwise. But even then, there was a shortage of human attention and it's just gotten hugely exacerbated since then. So I think trying to preserve your customers attention is paramount as well.

How can organizations transform with data?

0:26:43.3 Satyen Sangani: I wanna go to a book that you wrote in 2007, Competing on Analytics. 'Cause I think it speaks to somewhat of this idea of data culture and 'cause ultimately what people want in building these cultures or in building these knowledge repositories is to actually create some form of competitive differentiation or competitive advantage. What did you learn from that book and tell us what the conclusions from that work is? 

0:27:06.0 Tom Davenport: Sure, yeah. It's not rocket science. It's having senior people who are interested in analytical, decision making, hiring people who can do the work. The day-to-day work of analytics, both the data management and the data analysis. By the way, all these things are true about AI as well. I wrote another book a few years ago called, All In on AI, two years ago now. And it's basically the same set of ideas only about AI. I think because many people are not oriented to analytics or AI, devoting a lot of attention to upskilling them and helping them use these tools more effectively in doing their jobs. And ultimately having some sort of unique data that is proprietary to you, that will really differentiate you. Because ultimately data is a fuel of analytics and AI. And if you don't have something distinctive, you're gonna have the same models that everybody else has.

0:28:10.4 Satyen Sangani: And whether it's the All In on AI book or the Competing on Analytics book, is there anything that you've, given particularly what you said at the very beginning. Which was, we had a blip, went all the way up to 50%. Now that's gone back down, there's less interest. Do companies successfully affect these transformations? Is that something that companies are able to do or? And what does that transformation look like? Because there's sort of the idealized state, there's the current state, and then there's the question of like, how does one actually get there? 

0:28:36.3 Tom Davenport: Yeah. Well, it was interesting. In Competing on Analytics, I was fortunate enough to have a friend, a neighbor and a fellow little league parent and fellow Harvard Business School professor for a while, Gary Loveman, who oddly enough went from being a professor at Harvard Business School to becoming CEO of Harrah's, then Caesars when they merged. And the world's largest gaming company, an industry that should be based heavily on data, but wasn't in many cases. And we really engineered a huge transformation. So in this book on being customer-focused with your data and AI and so on, we talked about that as a great example of how to make that transformation. However, the disappointing thing is they ran into some issues. One of their business units had done a private equity transaction in the last financial crisis 2007, I think it was. And they declared bankruptcy. There was some dissatisfaction with all of the management team. Gary ended up leaving and the subsequent management just wasn't terribly interested in data or analytics or AI. They did some things, but nothing like what Gary had done. So I think it really takes a kind of a consistent purpose and drumbeat of messaging over the long run, if you're gonna be successful as an organization with this.

0:30:11.6 Satyen Sangani: Yeah, one of the things that I think we've been talking about is this idea of sort of adoption and culture. And adoption is a measurement of the culture. But ultimately, all of that has to result in some set of outcomes or value. And so, this assertion from the top that this is useful and these behaviors are useful and the consistent tie back to value are the two trends that we seem to have seen.

0:30:34.9 Tom Davenport: Absolutely. And I think that's been one of the challenges with AI. Companies have experimented a lot, but not really put a lot of things into production deployment. That was true of analytical AI. It's now true of generative AI. The numbers are getting better, but I think we're still probably around 20% have a production deployment of generative AI. Early 2024, I did a survey with AWS and it was 6%. So improvement is happening, but it's still too low. You don't have any production deployment, you don't have any economic value. And so, we need to commit and spend the dollars and make the organizational changes necessary to put these things into production or we're not going to get value, and it'll dissipate over time.

0:31:25.7 Satyen Sangani: You get the opportunity to talk to a lot of senior CIOs, senior business executives. In this kind of Hype Cycle, are you feeling like they are also coming to the conclusion that they've kind of gone past the risk-reward continuum for these AI investments? Are you feeling like there's gonna be a backlash in the near term? People are talking about sort of the limits of scaling on a lot of the models and that you're gonna need 10x or 100x more computing resources to get to a better version of the model. Is the reckoning for gen AI coming soon, or is that gonna be a big adjustment in your mind? 

0:32:00.0 Tom Davenport: Yeah, well I think there are two issues there. Is there reckoning coming and how are data leaders preparing for it? And I think it's tough for them because, on the one hand, as one said to me, this was a woman. Said, "I feel like the bell of the ball now. Everybody's paying attention to generative AI, and I'm ahead at every management meeting, and the board wants to talk to me, and so on." On the other hand, gone into organizations and said, who's responsible for generative AI around here? And you get five hands going up. And so, there's a lot of contention for it. And I've seen that in surveys as well. And so, something that's really hot, creates a lot of desire to own it. Maybe the smart data leaders are trying to be sort of more collaborative and consensus-oriented because then they won't have the blame if it starts to go down somewhat. But will it go? Will it go down precipitously? 

0:33:00.9 Tom Davenport: Agents are going to give the whole domain a boost, I think. Although there's even more hype for agents before it appears than there was for generative AI, I think. But I do think there will be some retreat, and I said in the MIT Sloan Management Review article that I thought this really needed to be the year where you demonstrate value and measure it, or 2026 will be much less fun.

What is an AI agent? How does it differ from ChatGPT?

0:33:32.1 Satyen Sangani: Yeah. Well, maybe let's talk about that agent topic for a second. People use the word a lot, and they use the word agentic a lot. And I think, what exactly is an agent and how does that differ from what people experience with ChatGPT and the like? And why is there so much promise? 

0:33:49.3 Tom Davenport: Well, I think there's several different attributes. One is that it does something useful, rather than just inform you. It performs a transaction or as we said earlier, it turns the document into reusable data or makes a reservation for you or something like that. It is more autonomous than existing systems. Nobody knows exactly how autonomous they can be because as I said, you're still dealing with machine learning-based prediction models and they might make the wrong prediction. And the ante is raised when real money and business transactions are involved. So I think for a while there's gonna end up being some review of these transactions by humans. Maybe certainly less than doing all the work themselves, but not taking the human component out completely. I think there is a generative AI component, but it's not just generative AI, it's probably analytical AI as well. It's probably still APIs, it's probably still transaction systems, ERP and CRM and so on. So there had to be a lot of integration, which means that it's gonna be a fair amount of work for companies to pull this off. I think vendors will help and they'll provide lots of tools, but I think companies will have to figure out what they wanna accomplish with it and make it happen. And that will take some time and effort.

0:35:22.4 Satyen Sangani: Yeah, for sure. As you sort of look forward to 2026, you've obviously done a future-forward view as to what the trends happen to be. Give us two or three predictions which I know as somebody who's watching the field, you're probably very careful to not do those things. But...

The need to redesign processes for AI

0:35:38.2 Tom Davenport: It's interesting you're the first to ask me and I haven't really thought about it all that much. But what I would like to see happen, is some really major change in certain industries that can use these tools to, healthcare as I say. I think enormous potential AI in healthcare is found largely in the research lab and not at the bedside. So I'd love to see some of these things go to the front lines. I'm also very interested and in a way, for me this is a retreat to some of my older work. I wrote the first article and the first book on business process reengineering. Neither were the bestsellers in that area. I like to say, I bore the burden of academic respectability. And there's very heavy burden to bear from a sales, book sales standpoint. 

But I do think people are realizing now that to get the value out of AI-related tools, we have to redesign our processes. And so I wrote an article in a current issue, the January, February issue of Harvard Business Review with Tom Redman, on how to marry process and AI. I've written a previous article with a couple of other guys on how to redesign your processes with AI.

0:36:54.5 Tom Davenport: And so, I think lots and lots of opportunity for that. And again, economic value requires that we change the way we do our work. And there has to be some intentional design activity. It can't just evolve.

0:37:11.6 Satyen Sangani: Yeah. I do think that this, the merging of sort of traditional business process with these agents, I think has a ton of opportunity because you can construct them in a way that they are pretty narrow. And then you can also take these orchestrating agents to actually start putting together these things in a way that make them perhaps more powerful and flexible. And then traditional hardwired processes have historically been.

0:37:38.5 Tom Davenport: Yeah. And even eventually, I think, interorganizational, where you as one company that I've worked with in the past, that is called CCC Intelligent Solutions, and they have created these ecosystems of property and casualty insurance companies and body shops and automobile parts vendors. So that when you crash your car, when they can give you a quick estimate based on a photograph from your phone, but then they can orchestrate the whole process. And I said to them, this agent thing is basically created for the kind of work you do. It'll be hard to sort of orchestrate all those interorganizational ecosystems and the information flowing back and forth and so on, getting everybody to agree. But it would be really useful when it happens.

0:38:30.7 Satyen Sangani: Yeah, it'll be an exciting moment when it does. Well Tom, thank you for taking the time to speak with us. I think, as I was so patient to predict, this was an incredible conversation and pretty wide ranging. So thank you so much for the gift of your time and we'll look to have you back on sometime soon.

0:38:45.2 Tom Davenport: Thanks. I just don't know, was I radical enough for...

0:38:48.5 Satyen Sangani: I think you passed the radical test, absolutely.

0:38:51.3 Tom Davenport: Oh good. Thanks for asking such great questions.

0:38:54.2 Satyen Sangani: Great. Till next time.

[music]

0:38:58.9 Satyen Sangani: That was an incredible conversation with Tom. We covered a lot, like why agentic AI could change everything. But also, why AI won't replace software as fast as some people think. And while human oversight will still matter, it won't need to be 24/7. We also explored how generative AI is transforming industries like healthcare, but only if companies can tackle the big challenges of making it work in the real world. As the AI landscape keeps evolving, Tom's insights give us a roadmap for what's next. I'm Satyen Sangani, CEO of Alation. Thanks for tuning in to Data Radicals. Keep learning and keep sharing. Until next time.

0:39:42.2 Producer: This podcast is brought to you by Alation. Your boss may be AI ready, but is your data? Learn how to prepare your data for a range of AI use cases. This white paper will show you how to build an AI success strategy, and avoid common pitfalls. Visit alation.com/AI-ready. That's alation.com/AI-ready.