Alter Everything Podcast
Episode 204
Episode Chapters
0:00 Opening Question
2:01 Why 2026 Is a Building Year
7:59 The Value Ceiling
14:40 Measuring Real AI Progress
17:56 Why AI Governance Is Different
20:06 Governing Agentic Systems
27:34 Original Thought as Competitive Edge
32:09 Designing for Augmentation
35:36 Predictive Analytics and Agentic AI
38:54 Lightning Round
42:08 Moving Beyond Productivity Theater
44:17 Key Takeaways
Joshua Burkhow (00:05)
Here's a question that I can't stop thinking about. If every company in your industry is using the same AI, trained on the same data, producing similar results, where does the competitive advantage actually come from? Well, my guest today might actually have the answer and it's the opposite of what most executives want to hear. Hello, this is Alter Everything. It's a podcast about AI analytics and the future of work. I'm Joshua Burkhow.
My guest today is Dr. Fern Halper, founder of the AI Foundations Group, VP of research at TDWI, and author of Data Makes the World Go Round. She spent 30 years inside of data and AI, and she has one of the clearest reads I've found on what's actually working inside of enterprises right now and what isn't. Most AI coverage fixates on what tools can do next.
What agents will automate? Which model wins? Sort of want to ignore all of that today. The more interesting question sits underneath. Most organizations have deployed AI in some form, but few can point to what's actually changed. And the reasons have very little to do with the models. So three things I want to hash out with Fern today. One, where companies actually get stuck.
Something that she calls the value ceiling. Second, what predictive analytics adoption curve from the last decade tells us about where agentic AI lands in this decade? And then lastly, with the question that I opened with, in a market where everyone uses the same tools, where does that competitive edge come from? Let's get into it.
Joshua Burkhow (02:01)
Fern, welcome. Do I have this right? You started in oceanography. You spent years at Bell Labs on continuous monitoring and led research at TDWI through the last three major data waves. And this year you founded the AI Foundations Group. I really can't imagine how many times you've seen the same cycles sort of come and go.
Fern Halper (02:25)
Have it right.
Joshua Burkhow (02:26)
Time flies. So instead of going back over all this history and data world, which I think you could cover in droves, I'd actually like to start where we're at. You've written that 2026 is an inflection point and not a sort of breakthrough year that a lot of people are talking about. What is that sort of signal that you look for that tells you that this is really an inflection point versus
Sort of just another round of hype.
Fern Halper (02:57)
Right, right. I was writing about the inflection point in my blog and I said that organizations are trying to make their way towards agentic AI and that 2026 was actually going to be a year of building. So I think that that year of building is around the foundations that are needed for both generative AI that uses company data and then agentic AI, which of course uses company data also. And
In my research at TDWI, see that organizations are sort of slightly more than midway through this journey, although it's obviously a continuous journey because everything is always changing. So what I'm looking for to tell me whether or not the inflection point is real is around both data readiness for AI and then the operational readiness for it. So the data readiness,
I'm looking at data around data silos, organizations consolidating, unifying, as well as unstructured data. I'm seeing that in my research that organizations that have a significant impact from AI, they believe that the data foundation is table stakes. So there's this group that has actually made progress with AI, and they obviously view that data foundation.
as table stakes because when you ask them what their early mistakes were, their early mistakes were around data silos and thinking that they could just layer AI on top of fragmented data foundations that weren't governed. And they realized that that was a problem. But I do think that they still now this year have to deal with unstructured data and putting more mature handling of unstructured data in place. So those are the two signals in terms of the data.
Joshua Burkhow (04:21)
Right.
Fern Halper (04:41)
Foundation and then in terms of operationalizing AI, I'm looking at whether organizations have the right skills and tools and roles in place. You know, so a lot of organizations have thought about new roles like ops people and engineering roles, you know, so are they putting those in place? You know, are they also doing things like building or buying tools around agent registries or tools that's going to help to monitor agents? You know, basically are they taking steps?
To get themselves into production. So those are the types of signals that I'm looking for.
Joshua Burkhow (05:17)
Yeah, those I mean, those are pretty clear if you're paying attention to them, right? I think you're a student of history as well as we've both been in been in this game for a little bit. Do you sort of chuckle and smile and when you see sort of these patterns happen again where people are, you know, looking at and say, hey, you know, data is really important. We sort of comment around the
Big data wave that came in and then 2017 when machine learning wave came in. There were a lot of these where again, the data foundations were so important. Do you see that as well?
Fern Halper (05:52)
Well, I mean, in terms of the big data wave, think data people got excited about it, but not necessarily consumers of software to the extent that they are with AI. So in some way, this is a bit different. But certainly like the big data wave was about getting people excited about what they could do with a lot of data, and how you can enrich your data and...
Get much better results. It was about how to deal with the data. To me, this is so much more complex than that is, you know, it's about the data, it's about the organizations, about the skills, it's about the governance. You know, everything has to be part of the system to achieve success, you know, really. yeah, it's the wave, but this one has gotten so many people excited that in some ways it's a little bit different. It's sort of a sea change.
Wait, that's not a good, I was an oceanographer, wait a minute, that's not a good analogy. It's a tsunami as opposed to a wave, how about that?
Joshua Burkhow (06:46)
Yeah. Good point. I got you. I got you.
Yeah. Yeah. Yeah. We've never heard that, marketing slogan of tsunamis, right? so when you, when you talk to executives, going back to the sort of importance of the, the data foundation, do you get the sense that more and more execs are coming to the table?
And realizing that they don't have to waste cycles, you know, going down these roads when if they, really rely on, on building a good, strong data foundation. do you see that more often or is that sort of worried that it's sort of falling away and they're not, they're not sort of getting the historical lessons here.
Fern Halper (07:28)
Yeah, that's a really good question. I think a lot of them aren't getting the historical lessons. And certainly, you know, we see executives stepping down because they don't think that they know enough about AI. You know, a lot of them didn't even think about the data foundation to begin with. They just sort of jumped right into, you know, off-the-shelf consumerized.
Joshua Burkhow (07:36)
Mm, yeah.
Fern Halper (07:48)
AI without even thinking about what was underneath and just said let's do AI whatever that means. So yeah, you know, obviously there are a lot that are thinking about it and it depends who you talk to but I definitely hear you on that one.
Joshua Burkhow (07:59)
Sure. Yeah.
Yeah. Yeah. It's just, it's, it's one of those things where you can sort of have a conversation within five minutes and understand if they're, really capturing the, the essence of the fact, like you're sort of hinting at it is that there's a lot more behind the curtain of, of this, this world that we're, going into. for you, you actually have a concept that you call the value ceiling. I hope I get that right, from what I gather, it's, it's this.
Point where sort of individual productivity gains from these off-the-shelf AI tools stop translating into organizational
Fern Halper (08:36)
Yeah, you have that right. In the book that I just wrote, I talk about a value ceiling in the context of sort of an AI maturity journey. And certainly with generative AI, a lot of organizations, as I was saying before, they were trying to leapfrog over foundational capabilities that they otherwise would have had to put in place for AI, machine learning, predictive analytics. So that includes the data foundation. So as they're using these off-the-shelf consumerized
Tools that are really point solutions, they're mostly getting what they think are productivity gains from it. So in this curve, I talk about quick wins in low risk use cases, right? Things like generating code to build a website or using generative AI to write content. But then the curve actually flattens and they won't get to truly high value if they don't have the foundations in place. So I show like a plateau.
And that's where the ceiling comes in. The ceiling is the plateau of the curve, because they get stuck. And once they have those foundations in place, the organization, the data, the skills, the governance, the operational know-how, then the curve can start to grow upwards again.
Joshua Burkhow (09:50)
Got it, got it. That makes sense. Are you able to sort of elucidate this idea for folks who maybe are new to this and can see what does the company go through when they're doing it? Is it just a matter of going through and getting all those foundational elements in line or how have you seen this sort of come to fruition?
Fern Halper (10:09)
You know, it's so interesting because when you have conversations with people, and they're talking about AI, almost takes 10 minutes before you can figure out what they're actually talking about when it comes to AI. Yeah. And then, you know, once I sort of explained about the consumerized path versus the organizational company enterprise path that uses company data, you know, then we
Joshua Burkhow (10:20)
100%. I so agree with that. Yes.
Fern Halper (10:36)
Come to a little bit of different conversation. So it just takes a while and they've all heard data is the new oil or whatever, data assets. So they sort of get that and then they understand what they may be struggling with.
Joshua Burkhow (10:37)
Whole different
Yeah. Yeah.
Yeah, I,
I, I'm sort of empathetic to this idea too, because you, you talk to a lot of folks, even the ones like us have been around a while that nobody is an expert in, in every area of a business, know, governance can people live their whole careers doing governance alone, let alone having to make sure that we, get that sort of managed appropriately and implemented appropriately.
Like if you put me in the shoes of say a leader and I come to you say Fern, help me out. I'm trying my best to get out of the sort of AI hype of things and get going. What are some of the first landscape pieces? Like where would you, would you go straight to data and say, hey, how's your data foundation? Or what would you sort of advise from?
From your point of view.
Fern Halper (11:45)
Yeah, well, I always say that they should start with the business need, you know, that there's no point in building AI that doesn't solve a business need. But if they were using off-the-shelf AI and then if they're struggling, to actually take the next step, like say they're trying to build an application like a chatbot that interacts with customers and maybe they had a general chatbot before.
Or maybe one that was using company FAQs to provide information to customers, but they want to expand it. And to do that, they're going to need to access customer data. And then they find out that they have 10 different ways to define a customer, and those are going to provide different data about the customer. they should know that they have a problem, right?
Joshua Burkhow (12:32)
Yeah. Yeah.
Fern Halper (12:32)
If they build a simple
Model and they see that outputs are garbage, that indicates that their data is flawed or biased. So I think that you have to get to the data question pretty quickly. what's the business issue that you're trying to solve, then do an honest inventory of what data you have in place to solve it. And then pretty quickly you're gonna see, well, do I have that in place or do I not have it in place?
Joshua Burkhow (12:50)
Yeah, yeah.
That's
Right. Yeah.
Fern Halper (12:57)
But even getting to
That point to think, where's all of that data? What does that inventory look like? And it doesn't have to be every piece of data that you have in the company, but just to solve that business need and then you can build out from there. But it's just interesting what happens with some companies.
Joshua Burkhow (13:14)
Yeah, there is this sort of trade off and you hinted at this one again is the honest assessment of it. I've been so gung-ho about the results and the outcomes and, trying to get to a specified goal that I don't look as honestly and cleanly at what's really there, right? Like, Hey, do I have all the elements? Do I have 50 % of it? Is it.
A hot mess or is it in good shape and can I sort of talk to that fairly eloquently? I think we can't shortcut that, right?
Fern Halper (13:48)
You can't, and it's just so interesting now too, because the types of that companies are building now, they're using unstructured text data. So, you know, we're so used to dealing with the structured data and now they're using the unstructured documents. And I was looking at the results of a survey recently and I was just by, the biggest use case was, basically around documents and...
Joshua Burkhow (13:57)
Yeah.
Fern Halper (14:13)
Customer notes those sorts of things. so even if they had some sort of structured data infrastructure in place, most organizations, a lot of organizations don't have the right unstructured repositories in place. So some of it's not even their fault, if you know what I mean.
Joshua Burkhow (14:13)
That's right.
Yeah, yeah, totally anything else that you find really interesting about the value ceiling and how it's sort of playing out from, from the point at which you wrote about it?
Fern Halper (14:40)
Yeah, I guess I've been thinking a lot about when organizations think that they're actually making progress because they have these copilots or whatever in place. And I've been thinking about it sort of in terms of the illusion of progress versus actual progress. So they have a lot of activity, but they don't necessarily have measurable impact from their AI.
Joshua Burkhow (14:47)
Hmm.
That's right,
Fern Halper (15:05)
I just think that that's important that you always have to measure, think about what the measurements are. those that have put the work in with the data and other foundations are the ones that I always see measuring value in production. And those that aren't are stuck in the pilots and experimentation that plateau. And think that in terms of
Any sort of gaps that organizations have to fill, know, if they actually want grow, you know, like they can keep getting, I guess, productivity enhancements, but if they actually want to grow the value of what they're trying to do, they have to move past the illusion of progress to actual progress, if that makes sense.
Joshua Burkhow (15:46)
Yeah.
No, totally. I mean, again, talk about history lessons, right? Like the machine learning and data science, when that all came, came about, it was the same thing. It was this idea that, we have the smartest people on the planet and data scientists are, are essentially gods of the universe. And they had a lot of the hoopla and the hype, but like, were they making
Of real dent in actual progress, real progress that was changing the company. And I think that was the underlying assessment that I got was that the change management still to this day is the hard thing, is the thing that I would argue most companies are not great at, is getting those things into place and doing the hard work and doing the foundational aspects to make these things.
Take hold and actually, make progress. What a concept, right?
Fern Halper (16:41)
Right, not to mention actual change management. You I remember when I was at Bell Labs and we could actually predict customers that were going to disconnect AT &T services and we were very excited about it. We had built the model, you know, this was a long time ago and AT &T I'm sure does all sorts of advanced things now because they were doing advanced things back then. But, you know, we went to the people who ran the call center and said, you know, we can predict who's going to disconnect.
You know, the service and they just looked at us like we had five heads and you know, why would we do that and how would we do that and how would we operationalize that? And, you know, certainly all of the, the compute power wasn't there at the time to actually do that. But yeah, but it just goes to, the cultural type of change that is also sort of needed. So.
Joshua Burkhow (17:25)
Make it viable,
Yeah.
Yeah, I totally agree. we're trying to get this new incredible technology, amazing technology in place. But it's just the tip of an iceberg that has,
All these things together, your data foundations, you got to get the skillsets, right? All these pieces. One of the ones that I want to shift to is the big topic of governance. And I'm really interested in your thoughts and your ideas around this because I talk about governance a lot.
Fern Halper (17:56)
Mm.
Joshua Burkhow (18:04)
I never really saw AI governance the same as sort of data governance. I've always sort of saw them, okay, they have the same word in them, but they're pretty different disciplines, maybe like different foundational sort of concepts here. And a lot of companies treat them as sort of the same thing that seems like a problem.
Fern Halper (18:24)
Completely agree with you. I view that data governance and AI governance are complementary. And certainly, there's data governance for AI. But AI governance is something that's different, right? It's going to deal with the models, not the data. So data governance never had to deal with things like versioning and documenting AI models. It didn't deal with model registries. It didn't need to deal with explainability of outputs or observability.
For looking at models and how they decay. They're different, you know. If you don't have good data governance, you're not gonna have good AI, but I've seen organizations same as you, they try to treat them the same, then try to make it two completely different groups that are dealing with it. But you know, now they're sort of coming together and saying,
Joshua Burkhow (19:11)
Right.
Fern Halper (19:13)
You know, we all should be under the same umbrella. AI governance is different than data governance, but you know, there's overlap with what you have to be thinking about. But you know, it's the same thing where people built models, but they never thought about operationalizing them. And now, they're building AI models, but they're not thinking about governing them. So it's a conundrum, but organizations are at least getting on board.
Joshua Burkhow (19:16)
Yeah.
Yeah.
That's right.
Yeah.
Really.
Fern Halper (19:39)
With it now to actually think about how they're going to govern their AI models, which is good news.
Joshua Burkhow (19:45)
Yeah, right. And I think the interesting thing is,
Like we've got, you know, sort of the LLM world that we're dealing with, but now agentic AI is coming out quick and you've got all these other concepts like MCP and, and, other components, those are sort of new, new grounds to figure out what AI governance looks like.
Fern Halper (20:06)
Yeah, it's so much more complex. I mean, governing a machine learning model. That's hard enough, right? And the vendors I talked to about agentic AI, they talk about, access controls, we provide access controls. Like somehow that solves everything. Or at least that's the place to start. But I don't know that I agree with that. You know, I think if you're going to build and govern an agentic system, you need to actually start with what that system is going to do.
Joshua Burkhow (20:12)
Mm hmm. Yeah.
Yeah
Fern Halper (20:33)
So you have to map it out. You have to look at where all of the control points in that system need to be, the inputs, the outputs, what happens in the agents, how they interact with each other. And so the governance design needs to come up front with your system design. I how many people are thinking about you have to decide whether you're willing to live with the terms of the risk as you map that out.
I've been actually hearing from people who are talking about that these agentic systems to start out with should be more deterministic and just perform certain tasks, you know, rather than letting them, be more probabilistic and, you know, go a little bit wild. So I think that that makes sense, you know, in terms of AI governance. the good news though is that, organizations are getting on board with it.
Joshua Burkhow (21:14)
Yeah.
Fern Halper (21:24)
They're saying, absolutely, we have to do this. that in any conference that I'm at, in TDWI research that I do, everyone's saying, whoa, whoa, whoa, in terms of their agentic systems, we got to put the guardrails in there first. So at least they're thinking.
You know, first, because I guess some organizations sort of said, let, let's just build the agents and see what happens. And, you know, now they're stuck with tens of thousands of agents and they're saying maybe that wasn't a good idea. you know, different companies, different philosophies.
Joshua Burkhow (21:52)
Yeah. Yeah.
Yeah. I mean, it's
Really wild when you think about it because the two that come to mind is just the cost implications, right? don't know if I've met anyone that could tell you, Hey, I've got 10 analysts working and they cost me exactly this amount based on whatever work they're going after.
And use of LLMs and the cloud codes and Codex of the world and building stuff. And so there's this sort of just from a cost management point of view, that's one area. The other one that I'm finding really interesting and is tied in closely with AI governance is the legal aspects. And I'm curious if you have any points of view on that or any hard lines on that realm, because what I'm seeing is
Legal is, is as an organization is sort of having to force themselves into say, Hey guys, you got to think about this more clearly. And then there's the sort of push and pull that's happening in a lot of organizations around. yeah, just don't worry about it. Let's just try this out. Like you said, and we're gonna, we're gonna see where the boundaries are and we'll come back to you. You know,
Fern Halper (23:07)
Hearing a lot more organizations when I talk to them, they're legal is part of, the governance team they're trying to interpret the compliance obligations that they have, et cetera. And certainly, you know, you think about GDPR even, I don't think people were thinking about like AI models, but yet,
You there is a provision in there that says if someone to open a credit card and you deny them, you have to be able to explain to them why you denied them, you know, so you need to understand what you're doing and all of that needs to be governed, in some way. And now it seems like people are realizing, you know, I have to be able to explain that or whatever.
And then you look at the EU AI Act, which is even more explicit about certain things. So legal definitely has to get involved. And that's other good news. I think that the organizations I'm talking to are bringing them in.
Joshua Burkhow (24:04)
I think one more question I want to shift into the next sort of theme here is
Do you see with all the new things coming out from MCP to all the new technologies that are coming into AI world, do you think this is not gonna be a sort of settled field or settled practice for quite a while? Do you think it's gonna?
Fern Halper (24:25)
Think it's early. What I see a lot of organizations doing is that they're trying to take the frameworks that are out there for AI and then they take some of the big four firms, they take pieces of it and then they try to put it together as their own framework.
They're not rushing into it, I guess I would say, not only do they have to sort of put a framework together, they have to put an operating model together about what that looks like in their company and, you know, dealing with this and how do we get the business stakeholders involved? And like you were saying, legal, is involved, IT, you know, there's just like lots of different people who are involved and I think it's just gonna take a while. And that was just for...
Joshua Burkhow (24:45)
Yeah.
Fern Halper (25:07)
You know, sort of thinking about generative AI and not even agent, like I just think that that's, I haven't seen, really good frameworks out there that look at agentic AI. Like I was saying, I think, it's case by case, mapping it out, thinking about, it from an auditing and control perspective, et cetera. And that's just going to take a while.
Joshua Burkhow (25:19)
Yeah.
Yeah. Yeah. I mean, could probably foresee entire companies coming out and just focusing on how to govern agentic AI. by itself is, is, it gives me a headache just thinking about it. Like it's challenge.
Fern Halper (25:44)
It's
Interesting because the auditing field, you know, I've kept trying to keep up with it a little bit since my days at Bell Labs dealing with continuous process auditing. You know, auditors are now talking about using agents to audit agentic AI systems. Talk about your mind-boggling. So.
Joshua Burkhow (25:50)
Mm-hmm.
Yeah.
That's right. Right?
I'm trying
To get your head wrapped around that one. Like who audits who now? Like it's a, we essentially have to get to a place where that's possible, right? Where, where every corner of a corporation enterprise can tap into the power of agentic AI, but have mechanisms in place that make it.
Fern Halper (26:08)
Right, right.
Joshua Burkhow (26:25)
Safe and reliable and, auditable, and, and, ability to go back to the regulators and say, Hey, this is how we did exactly this thing. The credit application, right? Yeah. It's, it's fascinating. yeah. yeah. I can't imagine. Yeah.
Fern Halper (26:39)
And supply chain is going to be a big one for agentic AI. So how are you auditing? Yeah, it's going to
Be very interesting. I don't think we're there yet by any stretch. So that's just my opinion.
Joshua Burkhow (26:50)
I think it's,
I got my career started in supply chain and I've always been fascinated because it's the sort of, multi-dimensional problem at all times, you know, trying to get all these moving pieces working together. And then to your point, you add, AI on top of that, it's, know, it's not the sort of segmented approach that, that could work, It's gotta be pretty comprehensive and then complex and
Agentic AI is highly complex already, right? what, one thing that surprised me Fern, in your recent writing is that the, deepest risks isn't governance failure, but it's something sort of quieter happening to the people that are using these tools every day. And I, I just did my
Fern Halper (27:17)
Right, right.
Joshua Burkhow (27:34)
My previous podcast was about sort of AI and people. you, wrote in February that in, in an AI saturated market, original thought becomes a scarce asset. Can you unpack that for me? Can you sort of give me, give me the underlying why, why scarce? Why now?
Fern Halper (27:54)
Yeah, I guess, you know, we've been talking a little bit about it, but what I see happening is that many organizations are making use of AI tools, you know, the copilots, the assistants, the ChatGPT's, the thousands of other tools that are out there in TDWI research, more than 90 % of them are making use of off-the-shelf self-service types of tools. And, you know, those models were trained on a corpus of data that was the internet, but
As more organizations are using generative AI to create something and that something is let the world like marketing content, then the models become trained on data that's already generated and the outputs average towards the center of the distribution of the bell curve. just becomes this homogenized thinking and everyone's just sort of at the middle of the bell curve. So to me,
That's not where original thought occurs. not at the center, it occurs at the long tail. So then if that tail is smaller, then the probability of original thought decreases. Because if everyone's just putting prompts into generative AI systems, mean, to me, where's the creativity and insights? I maybe there is some in some cases. I feel like we all become mediocre. I remember someone in a survey a couple of years ago was asking about
Joshua Burkhow (29:14)
Yeah. Yeah.
Fern Halper (29:18)
If people were positive or negative towards AI and these were data and AI professionals and you know, most of them were positive, really positive about it. But there were a couple of people who were saying, you know, we're just tending towards mediocrity and I've been thinking more and more about that. It's sort of like a bad sci fi show, you know, where generations before, wise people had created technology and now.
Joshua Burkhow (29:32)
Yeah.
Alright, ready?
Fern Halper (29:44)
Sad looking people, that are currently here can't fix it when it breaks, you know, so it just made me think a lot about, what I'm hearing, what I'm seeing and, when you read articles that are online and whatever, and you're just sort of saying, everyone's saying the same thing. It's, it's just all homogenized thinking at this point.
Joshua Burkhow (30:04)
Yeah.
Yeah. It's, it's really such a fascinating thing. when I try to share why I'm excited about AI is, is the creative part. on the whole, I don't consider myself highly creative.
What I found with AI, for me at least, is it allows me to get my messed up thoughts and messed up, not bad, but messed up, untangle them a little bit and get some original thoughts, get some ideas that might have a sort of spark of creativity, a spark of sort of, hey, I wonder if anybody else has come to this crossroads of these two ideas and then just fleshing it out.
I mean, I probably spend way more tokens on that than anything is just sort of going down these, these rabbit holes of, of sort of fleshing these things out. And I think AI has been pushed so much to the realm of, hey, it's a productivity tool. It's go in, get your work done. You can automate your PowerPoints. You can automate your Excel documents. You can automate your workflows. Let's, you know, let's go. And those are all.
Great. I'm not sort of harping on those, but are they at odds? Right? Are they are we sort of sacrificing this ability to do these productive things without having the sort of original thought to be like, maybe there's a better way. You know, maybe there's maybe we could think through this and come up with new creative ideas that, evolve the dialogue and conversation.
Fern Halper (31:35)
Yeah, I guess I think that when they talk about food groups or whatever, how to stay healthy, it's a balance. So maybe, you know, this is also a balance of things because certainly, not saying you can't get creative ideas out of AI, because if you know enough and you're critical,
Joshua Burkhow (31:37)
That's right.
Yeah. Yeah. Everything's.
Fern Halper (31:51)
You know, your critical thinking, your creativity is enough that you're putting the right thing into the prompt window, then you know, you may get something useful back. And certainly, we've all been helped, I'm sure, you know, by the off-the-shelf generative AI tools that are out there.
Joshua Burkhow (32:09)
Can you sort of help paint a picture of how you think this plays out in an enterprise?
Like, what would you do? How would you, would we sort of affect this if we, if we were so bold to think we could?
Fern Halper (32:22)
You know, from a leadership perspective, I think that effective leaders need to sort of design for augmentation and not substitution, you know, that means they have to be explicit about where
Joshua Burkhow (32:25)
Get your work done.
Yeah.
Fern Halper (32:37)
AI is appropriate and where it's not appropriate. Maybe AI can help accelerate your documentation or your initial exploration, but you still need to be expected to define the problem, validate the assumptions, defend the results, make sure that they're correct.
What I see some organizations doing is implementing practices like requiring their data analysts to explain or critique generated outputs, even like to produce a first pass answer before using AI. And I remember back when I was in graduate school, my major professor was saying, look at the data, look at the data, don't just run the analysis,
Joshua Burkhow (33:08)
Beautiful.
Fern Halper (33:21)
So you don't want to restrict it. You just want to ensure that AI doesn't replace thinking, you want to keep your teams sharp. You want, embed AI or sort of embed the thinking into the process, checking the logic and verifying the sources and
Joshua Burkhow (33:28)
That's right.
Yeah. Yeah, I, I totally get it.
Fern Halper (33:39)
Some organizations are putting peer review processes that specifically look for AI errors or your reasoning.
Joshua Burkhow (33:49)
I really like this because it's easy to, automate this process, do this thing, do this thing. essentially get my work done.
If we don't augment the thinking along with what we're doing, that's the drain. That's the drain of original thought, it's juxtaposed against a sort of
Personal experience I've had with our CEO and I work closely with him on AI projects and he's very much pushing the gamut like hey, let's think different. Let's not do the same sort of workslop that everybody else, what other new things could we build? What other original ideas could we come up with?
I wanna
Fern Halper (34:30)
Yeah, when I look at
Companies that are transforming, basically, it goes back to what we were saying before. It's not necessarily the ones who are just using off-the-shelf AI tools. It's the ones that are putting the foundations in place and doing all the hard work. But they're reaping the benefits from it. And that's not to say that someone just can't come along and come up with an idea.
Joshua Burkhow (34:35)
Mm-hmm.
Fern Halper (34:55)
And say, let's create an app that does X, Y, and Z for our customers and we can make millions of dollars. Obviously, that's happening. But I'm looking at more sustained and more potential for growth. So that's how I'm looking at it.
Joshua Burkhow (35:06)
Yeah. Yeah. Yeah.
Yeah. imagine that that world where you have sustained original thought and a strong foundation that is reliable and auditable and all these things. That's perfect recipe for growth. Yeah, that's pretty foundational.
I want to get into this next piece sort of a hard transition, but it's the sort of idea of this predictive analytics precedent. And you've written that predictive analytics follows the same sort of adoption pattern that agentic AI is following.
Fern Halper (35:42)
Yeah, you when I got to TDWI, I started asking organizations whether they were adopting, if they were implementing predictive analytics and every year, know, every quarter, whatever, I'd be asking the same question. And then I'd also ask, are you planning to do this, or you're experimenting with it now? And if organizations continued on the same path.
And actually were doing what they thought that they were going to do, know, with predictive analytics, the adoption would have been at 80 % already, and it's more to So I think that in some ways, you know, maybe agentic AI is a little bit different. I can't say for certain, but, you know,
I see 30 to 40 % of organizations responding to surveys saying that they're experimenting with agentic AI, but only about 10 to 20 % have it in it's sort of the same.
Same idea, know, like where's it going to be in the next couple years, what would adoption be like 10 years from now,
Joshua Burkhow (36:37)
Right.
Yeah.
Fern Halper (36:43)
It's also more complex as we were talking about before. So I think that that's gonna strike some companies down. And I don't like to sound negative because I've always been interested in what the next thing is. But just think that people underestimate what's actually involved. So that always concerns me.
Joshua Burkhow (37:01)
Yeah. Yeah. Yeah.
Yeah. Hindsight's always 2020, isn't it?
But then once you start to see these things sort of get into the organizations and actually get put into place and then get way down that pathway, say five years, 10 years, you start to see the difference of what was really hyped to the beginning to the end. I mean, big data had it, machine learning had it, predictive analytics had it.
I mean, I even think, you know,
AI is so pervasive, right? my mother who is technophobe to the degree that I can't explain, like I can barely get her to use her phone, but she's using an LLM. Like that to me is wild that like it's pervasive It's being used in every, every person either knows of it or
Does it? It's not like machine learning or predictive analytics where you had to have some sort of quant background, some interest in it, some sort of love for data of that sort. So smaller population. Does the fact that AI expand all corners of the the landscape there? Does it? Right?
Fern Halper (38:17)
Yeah, that's what I think. I think because it's so
Well known at this point and everyone's using it. Like you're saying your mother, my sister, talks about, you know, I talked to chat, chat said this, said, you don't call it ChatGPT. And she said, no, that's too long. I just call it chat, you know, so they have their own lingo, you know, maybe agentic AI will get further because everyone's talking about agents
Joshua Burkhow (38:33)
Yeah, yeah, yeah, yeah.
Fern Halper (38:40)
And going anywhere. Yeah, maybe it's not gonna take off to the degree that some people...
Joshua Burkhow (38:40)
Exactly.
Fern Halper (38:47)
Thought, I don't think it's going anywhere and people are understanding it more, I think, in some ways.
Joshua Burkhow (38:54)
Yeah,
Right. I got a fun one for you, Fern. So one of the things that I like to do on this podcast is do what we call a lightning round where
I'm going to sort of throw few things at you and get your sort of quick, immediate reaction. nothing too elaborate or in depth. And they're meant to be sort of quick, instinctive answers. But hopefully no curveballs. I don't try to surprise you too much. You good for it?
Fern Halper (39:23)
I'll give it a try. Okay.
Joshua Burkhow (39:25)
All right, all
Right. So I have six of them here. Let's see, see how they go. give me one signal a company has hit the value ceiling realizing it. What's one thing that you just, you see it and you're like, yep, you're there.
Fern Halper (39:38)
Okay, so when I see that they're using AI, but their business outcomes aren't improving, they can't measure value.
Joshua Burkhow (39:46)
Can't measure value, that's a good one. One thing AI governance has to do that data governance never did.
Fern Halper (39:52)
I'll say monitor the outputs of models, so monitor software outputs.
Joshua Burkhow (39:56)
That's
A good one. Good. One habit that protects critical thinking on a team using AI every day.
Fern Halper (40:05)
that's a good question. Deliberately do work without AI on some days. How about that?
Joshua Burkhow (40:12)
Ooh, good one. You're
Going to start a riot in the streets on that one. Yeah, yeah, totally. One vendor claim about agentic AI that you do not believe in yet, or do you do not believe
Fern Halper (40:15)
Yeah, I try to do that. I say don't use it, don't use it, don't use it. Yeah.
Yeah. I don't believe that companies are putting multi-agent systems to work in a production environment in any meaningful way. I don't think that there are fully autonomous agents, running end-to-end business processes. without human oversight, and I guess, you know, vendors are talking about everyone, you know, you need human.
Oversight, but then they also talk about, you know, that agents are your team members. It's another interesting cultural type of dynamic. So
Joshua Burkhow (40:52)
Yeah.
That's right.
Next one is one skill that stays valuable for the next 10 years, regardless of how fast all these tools move.
Fern Halper (41:06)
How about having good judgment? being able to frame problems, assess trade-offs, although I guess agents are going to help doing that, you know, and making decisions under uncertainty.
Joshua Burkhow (41:09)
Gonna go there too. Yeah
Being able to think through these things critically. All right, last one. You've done amazing so far. So one question every executive should ask before green lighting an AI project.
Fern Halper (41:25)
Mm-hmm.
Hmm.
Joshua Burkhow (41:35)
What should be that one question?
Fern Halper (41:37)
How does this change what we're doing and is it worth the change?
Joshua Burkhow (41:42)
Which probably shouldn't be much different than the other projects that they're green lighting, right? I mean, that sounds pretty straightforward. Cool. so I'm going to, we'll, ask a quick question to sort of to round it out. And then I think we can, we can let you go but if a listener runs a data or analytics team and is, is listening to, to this podcast and they want to move.
Fern Halper (41:46)
Yeah, yeah, yeah, yeah.
Joshua Burkhow (42:08)
Beyond the sort of productivity theater into sort of real value. They want to actually, you know, enact that change. What would you advise them to move forward on? Well, you know, we're not talking to the executive per se, but talking to the data analyst or the analytics team or that newly minted AI.
Engineer, what would you have for them?
Fern Halper (42:28)
AI, I guess I was,
I guess if their organization thinks that AI is just an off-the-shelf tool,
And they want to move further and push the needle further in terms of thinking about their AI programs. then they might start with an AI incubator team. It's a concept that is a small group. They listen to business problems. They try out new ways.
To solve it. I mean, like in some ways, the center of excellence that I was in at Bell Labs long, long, long ago, you know, was we wanted to solve business problems in new ways.
You know, and they follow a framework to solve certain problems, And one thing to be sure is that you set metrics.
Success metrics as you do this, because I always say that success begets success, because I've seen that over and over again, and it's really a virtuous circle. If you can do something where you can show that you delivered value and you measured then people start to get excited, that worked. Look at that. Let's do the next thing.
Joshua Burkhow (43:15)
Hmm.
Yeah.
Mm-hmm.
Yeah. Yeah. Yeah. Yeah. Yeah. And
Keep it going. Exactly. Yeah. Well, Fern, thank you so much. I appreciate your time. I'm very grateful for your wisdom here. I think this has been immensely valuable. If people want to follow up with you, follow sort of what your book and...
Connect with you, what would be the right place to do that?
Fern Halper (43:54)
Well, they can connect with me on LinkedIn. They can send me an email, go to aifoundationsgroup.com and all of my information is there.
Joshua Burkhow (43:57)
Got it.
Cool. Connect with you.
Perfect. Thank you so much. I'll make sure that we post that in the show notes, but with that, I just want to say thank you so much for your time. I appreciate you.
Fern Halper (44:12)
That was fun to talk to you. Thank you.
Joshua Burkhow (44:14)
Thanks, take care.
Joshua Burkhow (44:17)
Well, that was a fun conversation. Here's a few key takeaways that I took from my conversation with Fern. The first one is that the value ceiling is real. You know, most organizations, they're using AI, but they can't point to the business outcomes that actually improved and they can't measure the value. That's the ceiling. And most companies are actually sitting right under it and they aren't even realizing it. The second thing.
Is that AI governance is not data governance with a new name. Fern points out that a key difference is that it has to monitor the output of models, not just the inputs. And with the agentic systems, auditing becomes a whole new problem. Even Fern pointed out that auditors are now using agents to audit agentic AI. It's a field that's still being figured out.
Now, the last one is the one that I'll probably be chewing on for quite a while. You know, original thought is now becoming a scarce asset. And I don't say this tongue in cheek or being snarky, but you know, as more of us use the same tools, trained on the same data, the outputs drift towards the center of the bell curve. The answer isn't to use less AI, it's to design for augmentation, not substitution.
Ask your team to critique what it gives back. Ask them to deliberately work without AI on some days. We need that thinking muscle, that thing that you have to protect. Now, before you go, if you liked today's episode, can I ask you a personal favor and share it with your friends and colleagues? Let us know what resonated, what you want us to explore next. What is interesting to you?
You can email us at podcast@alteryx.com. And you can subscribe to alter everything on YouTube, Spotify, Apple podcasts, or wherever you listen. I really appreciate you listening and we'll see you next time.
