How executives can kick-start their enterprise AI strategies, also, what is "agentic AI," with Adib Saikali

Episode 283 · Tue, 26 Nov 2024 10:54:01 +0100 · 50:54 · SRT · archive.org · all Tanzu Talk

All right. Well, Adib, you and I talk about AI stuff a lot nowadays. I don't know. Since the summer, our brains must have gotten warmed up. And now we're slowly getting chilled out some more. I'm not really sure. But I thought you, we, the last time we talked, you were like, I'm going to go figure out what agentic AI really means. And I was like, this is going to be great. So you told me you did a little, you know, hanging out, chilling out at home, cleaning up the yard, stuff like that. And then you figured out this weekend what agentic AI means. So I'm ready. What's, what's the, what's the big reveal? I didn't, I didn't know that I figured out exactly what it means. I kind of figured out what it's not and more where it fits historically. So I was a little bit kind of, because I'm old, I remember clearly the emergence of two application platforms, the web and mobile. I was doing a little bit of reflecting on that history. And what I noticed was that, you know, back in the nineties started building websites and people called everything websites. No one used the term web app, right? People started using the term web app after the emergence of this famous Ajax pattern. When people saw Google maps and were like, how's it doing this? And, you know, kind of work backwards. And there was this, if you remember mashups and web 2.0 and all of that, right. And, and fundamentally what it meant was there was a new UX pattern. So now if you follow the UX pattern for building code web apps, and then you use Ajax and single page applications, you could deliver a particular type of experience. And then businesses were like, build me a web app. They would go to a developer or a team or hire a web app team. And, um, it meant something. So I kind of, uh, then, you know, after that a little bit later, you know, I, I worked originally at Blackberry, which was like mobile stuff before mobile. I owned a Palm

pilot at one point in my life. So I'm sure, I'm sure the kids out there appreciate you explaining what Blackberry is. That's the Blackberry and Palm pilot. Like back in the stone age, right? You had to like buy this, like you could buy this little PDA for, I don't know what to probably be like more than a thousand dollars at this point. And it was like black and white screen more looked like a Kindle, like than a, than a modern device. And then they didn't have any network in it. You had to like plug it with a cable to a computer to transfer any kind of information. Um, but then eventually, you know, we had the mobile, everybody had a cell phone, but it was a flip phone. Sure. You could write apps on it, but there wasn't enough screen and keyboard to do anything really practical beyond, uh, calling, which was kind of the killer use case to begin with. And then eventually you got short text message, right? But really apps and app stores and app ecosystems didn't become a thing until the iPhone, because there was a UX pattern that came with the iPhone, like this whole touch thing. And people were like, I can pinch to zoom and all this type of stuff. And so I kind of, I kind of go like, then people start saying, build me a mobile app. What did they mean by build me a mobile app? They meant build me something that's optimized for touch and for a touch interface, which is a UX pattern, which can run on a handheld device, whether it was from Apple or Apple's competitors. And so my observation is that whenever we give a name to a type of like application, it's what we're doing is where that emerges when you have a new UX pattern and you have a type of, uh, capability that didn't exist before. Like for the web, that was that Ajax capability.

And in-place update, which birthed the single page app and the UX pattern that goes with it. Then you got on the mobile world, the touch interfaces with no keyboard, big screen, uh, relatively speaking, and, uh, uh, the whole UX pattern that goes with it. So when it comes to, uh, gen AI and the foundation models, well, that's definitely a new capability, a hundred percent. And it does enable a new type of UX. And so when you have UX plus new type capability, I guess I'll call that an agent or like it, it kind of gets its own name, right? So you'll see people talking about AI agents, people talking about agent, agentic apps. And so rather than trying to come up with a list of, you must do these 15 things, which I don't know what's on that list. I'm just saying, yes, it's a thing because it's a UX pattern plus a capability and it's just software that uses the foundation models. And it's trying to offer a user experience that is distinct from mobile and distinct from like web and maybe even requires less effort to use it. Mm. Yeah. No, I, I think, I think that's, that's great framing one, because I think about Ajax way too much, like, like in, in, in the, in the same, like, I think very similar to how you were thinking about it there as it's, it's, it, it, well, it, it was like, uh, uh, an example of what we're, you're just talking about that I lived through and actually did some work with and which is to say, I'm re-summarizing what you, you're saying, but like, you have some new type of, um, I'm going to call it like infrastructure, faceless technology. And then you have a new way of using it, new, new UX, whether it's a UI or delivery or whatever. Right. And it does, uh, and then that drives

new opportunities and ways of doing stuff. That's my very abstract way of putting it. Yeah. Exactly. I mean, this is why I'm like agent is simply, I use, I built some software to talk to a language model or to a foundation model and I wrapped it in a UX. So users don't even know that there's a foundation model underneath. Right. They're not the ones, they're not the ones talking to the foundation model directly. Right. And, and I think, I think that highlights something that is, I don't know, uh, important about like what, what, what I like to call enterprise AI, I guess for consumer too, but like, I don't really care about consumer tech so much, but, but in, in a, in a business and an enterprise context, well, my, my first problem is always like, well, what are you going to do with this thing? Yeah. Right. Like, but I think, I think what you're getting at here is one, let's call it like strategic thinking tool to get closer to answering that question, which is, well, let's not think so much about like training and models and all this stuff. How let's start with how we're going to interact with it. And are there different like modes of interacting with this thing that we can kind of sort through kind of like, um, you know, those books where you've got like flip cards and you've got three layers of a monster you can make, like, let's kind of like flip through and figure out like what makes sense here. Right. Yeah. Because, because you know, the, uh, and, and, and I think, I think it's someone like Ben Thompson, who I heard first saying this, but like, you know, the, uh, the, the, like the chat part of chat GPT and all the other ones now is almost like an accidental innovation. Like, like when that, when that was released two years ago now, I think, I think the, the, the team over at open AI was just sort of like, well, here's something so that we can mess around with this. Right. And then, and then it became the thing, right. Like, like it became huge. And it's exactly what you're saying

is like, here, here's a combination of like an infrastructure technology and the UI finally, that has good user experience. Right. And, and like, we bring these two things together and you have something new essentially. Yeah, for sure. And I'm like, I'm, I'm convinced that UX designers will ultimately be the ones that define what an agentic slash agent is. So I'm calling AI agent equals to just software that uses foundation models though. I've seen hints of what that could look like. So I'm, I'm, I'm currently like really intrigued by this idea that a key difference between um, regular UI and, uh, AI powered generative UI is the following in a, in a regular kind of, whether it's a mobile app or a web app, I have to clickety click around to get to a widget to do something. Right. Like classic example, uh, say you're using an application with some sort of credit card stored in a, I don't know, Uber, right? Like as an example, you have to navigate to the screen and settings, payments, credit card, edit, ad, blah, blah, blah. Right. Or you go into the web app for Amazon and you go and you're like clickety clicks profile settings, payment, add, edit credit card. And you know what, if there was an AI in that was part of the app or an AI or a UX that's AI driven, I should just be able to say update my credit card. Hmm. Yeah. I just said that whether I type it on the keyboard or I say it, it doesn't really matter what the modality is for the input, but I would expect the widget to come to me on demand for the thing I'm doing right now. Yeah. Right. Right. Right. And then, and then, I mean, also if you, you know, in that very specific

example, if you had enough like security protections and et cetera, you could sort of like give it one time permission to access your wallet or whatever. I, you know, I don't even know what I'm talking about, but like it, you would almost have your agent talking with its agent to your credit card. Okay. So, so that, that touches on like a common theme running through various definitions of agent is is that the agent is autonomous. Yeah. And actually does stuff on your behalf. And I'm like, eh, don't really care. Like I don't need this thing to be, I don't need this thing to yet be a Star Trek level agent. Right. I needed to just do something stupid. Can you just free me from having to click through four different screens and remember, have my brain remember how to navigate to the payment settings page? Yeah. Yeah. Yeah. And like, can you just like figure out that like, like, can you imagine like an app where the only thing you see is just a blank screen and a chat bar at the bottom? And then as you try to do different things, like widgets start showing up and allowing you to do the things you need to do. It's a very different, it's a different potential way to actually just design a user experience. And that's what I'm excited about. So for me, I'm going to call that agent, an AI agent. Yeah. Yeah. You know, I think, I think maybe to, to maybe generalize it to be too general, you know, essentially maybe an agent, an AI agent is something that's figures out, not figures out as the wrong word in this context, that it's kind of like mediating you using an AI thing. Like it sits between you and the model. And like, it doesn't have to be, so, you know, you could say like a chat thing is an agent, which, you know, once, once you understand how much goes on with those chat

applications that has nothing to do with AI, you're like, oh yeah, I see. Right. I mean, there's like all sorts of stuff that they're, they're doing, you know, pre and post processing and all of that, that, so you need that kind of, and you know, words sometimes matter, but like, the whole point of the word agent is that you have something acting on your behalf, right. Like doing something for you. And so to that extent, like the chat thing's doing it for you. And just like you're saying, if, if the agent is, uh, figuring out the right, the right, the good UI for you to have to interact with stuff, then, then, then that makes sense as well. So, yeah, I mean the, the, the, the crazy like part of agents that I just not convinced is ever going to really work. Um, well, I could, I, I'm probably could be, I hope I'm wrong about it, but it's like, there's kind of this obsession with replacing people. Like if you go to certain companies, websites, especially, uh, AI company startups, they're going to make some very bold claims about how you're going to, we're going to replace your human, your human with an agent, right? Uh, some of them make less of a bold claim, which is we're going to create an agent that can help your employee, like be better. So maybe that's 50% of their job, but they're like much more productive. And unlike my ambition is so much lower than that. I just there's so much lower hanging fruit. Like, why don't we just focus on that? So like, this is probably going to go through the standard kind of what they call it, the hype cycle, right? Like, like, I don't know if we're yet at what do they call the top of the cycle? Uh, the, the, the, it's like the, uh, that is it the apex of expectations? No, it's something like that. Yes. Yeah. Yeah. All I know is that there's the, the, the through of disillusionment and then the slope of enlightenment. That's right. The trough of disillusionment. Yes. Exactly. So it's like, it's, uh, Oh, and then another good phrase, the plateau of productivity,

right? Yes. That's, that's, that's both judgmental and descriptive. Right. Exactly. Exactly. Well, there was, there was an interesting announcement, I think maybe a couple of weeks ago from cloud with their computer use. I don't know if you heard about that. Yeah. Yeah. So it's, it's kind of like, okay, are you going to adapt the AI to understand what every screen looks like? Maybe do image recognition or maybe tap into the assistive features of the OS. So it doesn't have to like, you know, kind of use computer vision models to understand what's happening on the screen. But either way, you're trying to teach the AI how to use an existing tool. I don't know. Maybe you should change the tool. Maybe we should change the apps and build them more on the AI's capability than the other way around. Yeah. Yeah. Yeah. Yeah. You know, like, so, so one that, that, that makes sense. And two, I think, I think as, as you were saying earlier, you're, you're older, as am I, we remember XML, HTTP request, and the magic of untangling how Gmail worked and all of that stuff. But, you know, I think when it comes to automation, uh, I think I've lived through maybe at least three, at least three giant waves of like automation. And the story is always the same, that it's going to replace people. And there is some moving around of people like, like, you know, I, the, the, the startup I started on was like an online banking one. And there's not really like as many like bank tellers, maybe as, as there used to be. I don't know. I still feel like I talked to bank tellers, whatever. That's, that's a, that's a weird example, but most recently in kind of like the, the platform cloud native cloud DevOps space, like in theory, there's a lot of replacing people and what they do. But I think in practice in it, and I think this is probably true of a lot of

knowledge work or white collar work is one, it's hard to just get rid of people. Like as it, you know, it doesn't, it's not quite as simple as like, oh, a month later, now we've got this AI thing and we got rid of all the people, right? Like it's, things are a little different than that. But two, I think what ends up happening more is now you just get the people to do more work, right? Not, not more work, but they, they have something that's automating what they're doing, freeing up time and they do more work. Because if you were to go ask management at any company, like, is there more you'd like to be doing? I'm pretty sure they'd be like, where's the line I can stand in to tell you about that. Like I always want to do more stuff. So I don't know. The number of humans was limiting, was limiting their ability to grow because the invention of whatever time-saving technology was there. And so if it was those companies management were already probably planning, Hey, we'd like to expand. So we probably need to get more customers. So we probably need to hire more humans. Yeah. That's, that's, that's a, that's a clever way of putting it is that the humans are the bottlenecks. So you don't get rid of the humans. You want to make each human more productive. Exactly. I mean, I'm a, I'm a, I'm a techno optimist for the most part, uh, except in recent years when, when tech really became, uh, uh, usable by, uh, um, um, in situations to polarize people and divide them, then I'm like, maybe I should revoke my membership, be more skeptical of my membership in the tech optimist club, but. Indeed. Well, that, that, that, that brings back to the main thing I wanted to talk about. Like I was saying earlier, I don't really care about consumer tech very much. Right. And I think, I think maybe these proclamations of optimism in the consumer tech space are a little, it's not all bad, but, but it's a different, you have that conversation differently because it has, it has, it's just different materials you're working with. But in the business context, I think, I think equally, there's kind of like different ways of, of, uh, thinking about how you use tech. And I don't know,

I don't know exactly how it is in the consumer space, but you know, a couple of months ago, we were talking about like, um, kind of that question I threw out earlier, what, what are companies going to do with this AI stuff? And, and I think you brought up an excellent point. I really hadn't thought about that, that we, I wanted to discuss here a little bit is like, all right, so sure at the, the, the, the, let's say the operations, the developer, the UX, the people making the product, that's a whole interesting thing. But what's not really talked about very much is like, what does management or the, or an executive do like, what is their role in like this, this AI future that we're going to be in? And the way that you put it that I thought was interesting is, is like, uh, you know, it's not, you didn't put it this way. This is my rewording is like, they're not the ones who are going to be doing the technical work. Like they only have a limited amount of things in their toolbox that they can do, which, which I thought was a good point. And so that's, I thought, you know, I thought we'd see how far we go just, uh, figuring out kind of brainstorming what it is you think a manager and executives role is and sorting out how they're going to use AI. Yeah. Yeah. I mean, I look at it, let's, let's take it like use a, as a mental model, a bag, right? We have a large bag, you know, a really rich bank, you know, a bank that has, um, a very well-paid CEO and think of the direct people that report to the CEO, right? Maybe you'll have like an EVP head of personal banking and EVP head of commercial banking, EVP head of capital markets, EVP head of something. And so each of those, what, one of the models I've seen in a bunch of the banks is they might also have an EVP of slash CIO head of for the bank, right? Or they might, uh, move the, that function underneath each of the EVPs. So like you'll have your capital market CIO, your personal banking CIO, right? And, and I've seen

organizations kind of like shuffle between the two, but let's say now you're, uh, let's just stick with the, the, just to keep the model simple that you have the one CIO that reports to the CEO. Now, if you're the personal banking, uh, leader in the bank, your success is based on innovation and trying to make more money or save money within personal banking. You're not really necessarily that dependent on what's happening in the commercial banking or capital markets are kind of like kind of isolated. Your, your objectives are going to be independent and you'll be successful or not within your silo, uh, without too much perhaps dependency on that. Right. But the one person that's, that's head CIO, right. That reports to the CEO, like everybody else needs something from them. And personal banking, I'd like, wouldn't it be great if people could be qualified for a mortgage in 60 seconds or less so we can like, you know, sell more mortgages. Right. And, and so they become kind of like, uh, uh, the bottleneck and on, on the rest of the business. And they also have the potential to become unpopular because as that head leader, right. Of technology, again, you don't do any of the work you have all these teams. So what I've observed in my interactions with these executives over the years is that they get into that position because they're really good at managing their team and getting the most out of their team. And I've noticed a common operating pattern, like the new technology comes along. That executive basically says, if my team agrees that something is a good idea, I'm going to do it because they all agree. If they, if you have, let's say out of your direct team, uh, a couple of people, two or three, you know, uh, uh, uh, directs that are themselves executives

are just not reporting to the CEO. Right. Then, and they, they completely disagree about a technology or an approach. Uh, what I've seen these execs do is they'll just do nothing. They'll be like, there's too much disagreement. Clearly the market is not mature for the most part. They'll wait. Now, I forgot to say that this particular bank, let's put the model that they're more of a conservative institution. They're not trying to be the first, like with, with, with something. And so they're, they're wanting like, you know, to do something that a few other people have tried. They don't want to be the first. They're in line to be the second user of something. Right. Right. Right. Uh, and then, uh, uh, if your management, if your team kind of like has some disagreement, which is healthy, but they're like, I disagree and commit, that's easy. You can still make the chase, uh, the, the choice forward. This is my really long winded way of saying that, um, what the executives are doing is they're managing the conflict and the energy between their team members. And when it comes to AI, like, that's what they got to do. They got to figure out how to get the most out of their existing team. And to do that, um, it's really hard in the case of AI, because you need to experiment. Right. And experimentation means that you're probably going to fail, but everything is going to work. And a lot of these executives are like, oh, I made an impact. That's why I was promoted. I made good decisions. So it's like, Hmm. So if you promote people based on them making good decisions and hitting the plans and the objectives, how could you measure them on something that's probably going to have ideally a 50% like failure rate? Right. Right. Especially. Yeah.

Yeah. Right. And, and so, uh, what's the number one thing you should do is figure out how to create that situation where it's okay to experiment. And that is something you badly need to figure out, Hey, what's going to work in the bank if we add AI. Now you don't want to put those people completely outside of the no accountability. You can do whatever you want to use whatever technology you want. You can break all the rules debate because then they do science projects that have no impact to the business. Right. And, and, you know, I've, I've, at least in discussing not an AI, but in other areas, the other problem with that is sometimes if, if those experiments are successful and you try to go apply them to the main business, there's an easy amount of chaff excuse that, you know, your, your rivals can say, which is like, Oh, well that was in the labs where they could do anything. That's not going to work here. Like it's almost like you're not in real enough world situations to, to expand it to your actual business. Yeah. I would go like, why don't you just get the people that are actually doing the actual real thing. Don't need to create a lab. Like don't go out and create an AI lab. Right. Yeah. Like that's just a recipe for wasting money. I saw way too many digital factories created in the last 10 years where it's like everybody else in the bank gets to use a windows machine, but the digital factory has max. Everybody has black coffee, but the digital factory has a really nice coffee machine that can make you all sorts of coffee and they have max and they kind of look more like a Silicon Valley office and all of that. And, uh, you know, it's a little bit of cargo culting, um, uh, software companies, but it's really, really just like, as an exact, what do you need to do? You need to make sure you create that environment that allows your, your, your reports or reporting to you to actually

do something meaningful in their current area. And that means that you and your org have to start learning about AI and doing a little bit of experiments and stuff like this. So, so what do you think is, let's see, where should we start with this? Well, what do you think is the, the, the foundation for doing that? And, and, and let me extemporaneous here a little bit to, to figure, to tell you like what I'm asking is, you know, one answer would be like, well, you need to set up like a, a common AI service, just the technology for people to use. So they don't have to worry about that. And that's fine. There's another answer that's sort of like, well, as the executive, you have to figure out how to prove to people that experimenting is worthwhile. Right. So, you know, we've got, we've got the classic big company change vectors, the technology and the culture, uh, like, like coming in. So like, and I don't know, maybe if there's a third or a fourth one you're thinking of, but like, what, what do you see the executive doing on those two vectors to get people experimenting with, with, with AI? And, and I want to point out before I stopped talking that like, you know, what you said just there is probably the most important thing that an executive can do at this point with AI is realize that they just need to experiment. They need to learn what it's even useful for. Right. And, and I, you know, I would almost say, you know, that one, what would you call it? One test is like, if you're sure, you know, what's going to pay off for your AI stuff, you're probably not thinking about it correctly. Right. Like instead you should be like, I have no idea what we're going to do, but let's experiment and track those experimentations and let's learn, learn might be a better word than I was told by one of our customers in the military. Once I should never talk about failing to learn. I should just talk about learning. Yeah. But you need a way to learn, uh, what, what's, what's useful. I think that

there's two, there's two, maybe three types of learning that I would, I would advocate for. Number one is the basic one, which is you should be personally using whatever AI products are out there that have emerged, no matter how crappy they are, just see how you respond to it personally. Right. Um, and, uh, that's, that'll give you a sense of the impact too, is obviously talk to a lot of people, talk to people, talk to normal people, not people in technology, see what they're using. Ask them, Hey, is there an AI app or feature? Has it been great for you? Uh, in my case, uh, I, I have a very distinct, uh, situation with my electrician who language is not English. And, uh, he told me, uh, um, a while ago, Hey, I started using chat GPT for a copy and paste all my emails. It makes it better. And all my colleagues were like, did you take an English class? And he's like, yes, I did. Because he didn't want to tell them that he was like using chat GPT. Uh, and I'm like, uh, did you know you could actually, rather than doing that, you could actually ask chat GPT to explain to you what was wrong with your English. And it can like, like, don't tell it to fix it, tell it to tell you what's wrong with it. Then you fix it yourself. Yeah. And then you'll learn better. And he looked at me, he's like, can it really do that? I'm like, yes, I show him how to do it. And his head exploded. Like, it's like, this is like a free English teacher, right? Like, I'm like, yeah. So, so this is the part where, uh, it's like, it is worth it to talk to normal humans who aren't in tech, aren't in your industry, see how people are using it. And then the second learning layer of learning is to actually understand fundamentally what this technology is. Like you need to do a little bit of deeper diving than you might normally need to about established technologies. Like let's say using the bank executives,

you probably don't need to know the ins and outs of how your ATM machine works or how the phone system works or something. But because this is a new technology that can be very disruptive to your business and to your processes, you really should learn a bit more than you normally would in situations where you would delegate something away. So what's the right level of understanding what AI is for an executive would be the next question. How much should I learn about it? Should I go and like crack open my math books and, and you know, get into that? No, absolutely not. What you should really be looking at is understanding its limitations and understanding how those limitations may disappear over time. Like what's going to get faster, cheaper, better, and what's unlikely to get better. And once you understand its constraints, then ask yourself the question of how can I do something useful within the constraints that exist today? Right, exactly. And, and maybe, maybe that's a good, uh, uh, a good additional way of thinking about it is, is you're, you're not only learning how it works by using it, but as you say, you're discovering the constraints, like constraint seems like, like a negative word in this context, but really it's just like a more colloquial way of putting it out. It is just like, you're figuring out what it can do, right? Like, like what it's actually capable of and, uh, versus not like, you know, I was reading something over the weekend that, uh, what was the metaphor? It was like a lot of people think about using AI as an, as an intern, but it's the metaphor that's better is that they're like a brand new employee that can't remember anything. Right. And so like they're capable of doing something, but you have to constantly be reminding them and coaching them and doing things like that, kind of giving them context. Yeah. I, I, I, I, I find that analogy to be, uh, like it triggers me to think like,

I have to spend a lot of effort to train and coach and monitor. I'm like, can I think of them more like a default, like, you know, those little power drills? Yeah, yeah, yeah, exactly. I don't need to coach it. I don't need to tell it. Like if you use the AI as a single purpose tool for a very specific task, right. Then you don't have to do all that coaching. You just get the benefit. Exactly. And see, see that, that's the kind of thing that you would learn from experience, right. Is that, is that it's maybe, let's see, maybe there's faster ROI, so to speak, if you just try these kind of like one shot or highly specialized uses, right. Instead of having just like a general help you out with whatever, which might be cool as well, but maybe you'll get more early successes by just figuring out like, what's, what's like a simple thing that it can do really well so that we don't have to like load it up with context and have it understand this over and over again. Yeah. Like the, the, I think though, like one of the issues we might all be collectively dealing with is we've had a lot of really good movies and sci-fi where AI is featured and AI is implicitly assumed to be like faster, better than humans. Like the benevolent AI vision or the AI is like, you know, kills everybody, turns us into human batteries, like matrix or terminator or whatnot. Sure. And really it's the AI as like the power drill, the power, the power tool. That is the harder one, I think for everybody to imagine because it's not exciting. Right. But you know, what if it could do that and you don't need to change a lot of stuff. So I have a different relationship with constraints. I think constraints are the only way humans make progress and that every set of constraints contains within it, the thing that will allow you to violate the constraint and not die. A good example

of that is gravity. Like we have airplanes, right? Right, right. So within the constraints, within the constraints of the laws of physics is that some combination that allows you to fly, right? So it doesn't matter that the AIs hallucinate, that the AIs have certain problems. The question is, can you engineer something on top of this technology with its constraints to allow you to accomplish something better than you could before? And the answer in 2024 is absolutely, yes, you can. Like it's may not, it's not Star Trek level AI. It's not going to live up to all the hype, but it may be just like a 5% improvement over something. And hey, a lot of 5% improvements adds up pretty quickly. Yeah. So let's close this little, well, first of all, for our listeners, Adiv and I have been talking about a little book. I always call these books the something for the busy little executive. We'll see what we actually call it, but you know, like Enterprise AI for the busy little executive. And I think, I think it's, it's good to kind of talk through these, these ideas so we can transform into text. And so, you know, we've got a little section here and, and I think maybe to finish this section out, you know, like the dog is very enthusiastic about our project here. Yes. But I think, you know, you alluded to this earlier, or we talked about it directly earlier, where you and I both having worked at Pivotal and VMware and Tanzu and et cetera, and wherever we are now, so forth and so on, like, we've kind of been through a similar thing of like, I don't know, this, this discovery of rapidly being able to do software and enterprises. And I remember we spent several years, as you kind of said, talking and witnessing executives who were changing over how they thought about software. So there's kind of that past experience to draw from. And I don't know,

you know, the technology, as with your agent thing, the technology determines a lot of the behavior and vice versa. But maybe drawing on that and also like new thinking you have, like what, what kind of structure or expectations or timelines? I don't, I don't know, like, what is, what's the program look like that, that an executive kicks off for this kind of experimenting? Like what, what do they do? I would say step zero, use the technology. Step one, understand its constraints. Step three, get deep on the processes and places where you think it would apply in your company. And then design an experiment in there and, and run it. So it's, it's, it's like, for real, not, not, not, not like, yeah, we're not going, we're, we'll put it in the innovation lab and see what they come up with. I don't know, do it with the people that actually own that business and own the MBOs on that. So yeah, it's not just in the lab. Yeah. And, and, and I mean, tell me if you think it would work, but I think I liked the, uh, the pivotal labs and the Tanzu labs kind of structural answer to that, which was like, pick a, I don't know, pick a, uh, a two to three month project, uh, that's actual business facing and do something right. Like I'm, I'm, I'm sure if we're, if we're kind of pitching it in a formal way, it's a little more structured than that, but it's sort of like, and, and the experiment that you're establishing, you know, I, I really want to call it out because I think it's important is it does have to be business facing. It doesn't necessarily have to be customer facing, but it has to be something real, right. Uh, that, that will have that involves the business in a real way and we'll have some kind of tangible business result, right? Like one of my, one of my favorite examples, even though it's ancient is, uh, you know, Home Depot did a couple of these, uh, back, back when it was pivotal where they,

the, uh, the software they had that was running the custom paint desk was not great. And so their, their kind of position was like, that's where we could learn to do software differently because it's already not great. And so any improvements we do will be valuable and we can't really make it worse. Right. And so it was kind of this area where they could go in and discover things to do. And I don't know. I mean, do you think that same approach would work in kind of experimenting with AI stuff? Oh, absolutely. I mean, like, again, going back to, um, the banking example, right? Like you can think of, um, your, I don't know, let's, let's call them, whatever the people that you go to in the branch that you would say, I need a business loan or I need something, right? They got to learn about your business. They got to manage, like, uh, talk to lots of different companies and all of this stuff. They're clearly in charge and the final owner of the process shepherded along to say, hand out a million dollar loan, start a business or something. Think of all of the things where, uh, you could make them save them an hour or save them half an hour and just talk to them and find out like, what is something you spend a lot of time on this week, every week that you wish you could do faster. Get that list of, I don't know, 10 items, go through that list and ask the question, what could I do? If I, what could an AI do here? Like, can it do anything? Maybe it can't, right? Maybe the number one frustration is playing phone tag with, uh, right, right. And, uh, uh, maybe it's that customers, they, they talk to the customer, the customer shows up and they tell them, bring these five things and they don't bring, they bring four of the five documents they need. And then they need to have a follow-up appointment, right? Where the fifth document is, is brought in. So I don't know, can, can AI help in,

you know, maybe the AI simply takes the five documents that's needed. And, uh, the, that, the person in the bank is able to, uh, get a text message out like 20 minutes before the meeting, reminding them of the five things, right. That are, they need to bring with them. Yeah. Yeah. Yeah. And like, I mean, I mean, you bring up a good, a good, um, mechanic. I don't know what else to call it, which is, which is like, uh, user interviews, otherwise known as ask the people doing the work, how you could help them. Right. Right. And, and like, I think, I think the first part of what you're saying, um, well, there's a couple of things. One, you know, of course you want to go talk with, with users, another word for people, you want to talk with whether it's an employee or a customer and just kind of ask them an open-ended question of like, what would be helpful. But I think that's where your constraints point comes in as if, as if whoever's doing this interviewing, they know the constraints of an AI thing. They can kind of fish around for like, would this be useful or would that be useful? Exactly. And like kind of observing and kind of like mapping out what the, what these people do, uh, what these, the, the users do, and then applying some, and then coming up with the theory of what to do. Right. And what I want to call out about that is I think it would be easy. And I bet a lot of people do this is you, you, you go through, uh, the Adib steps, you, you know what it is you understand. And then without going to talk to the employees, you just come up with like, I think this would be a good idea. And you might go out and implement something that, that automates or that helps out some part of what a user is doing. And then you go give it to the user and they're like, yeah, I really just need people to bring all their documents, right? Like you've spent a lot of time solving a problem that may have existed or from a technologist point of view was an exciting thing to, to make up a word

technologize, but, but wasn't as valuable as other things that you could do. And going back to that Home Depot example, right? Like in that example and a few other ones, the first thing that the teams did was like go to the store and just watch and kind of like see what people are doing and interview them. And that's where they started with that constraint of knowing what they could do, if they could deliver new features and software on a weekly basis. And then, and then actually seeing what people, you know, the users are doing, they were able to kind of come up with some experiments. Um, and I guess, I guess that comes, that brings the, the, the, the final thing that I think would be interesting to bring up is so an experiment, if I remember my science is you have, let's see, you have a theory, uh, hypothesis, uh, a theory or hypothesis, and then you run an experiment which generates data. And then you've got to analyze that data to see if our hypothesis was true or not. Uh, as I think that's the scientific method. I'm not, I'm not up on my Sir Francis Bacon. Uh, but like, how do you think that applies to like AI experiments? If, if it's even possible to it's really no different than the UX studies. Like I remember back in the day at pivotal, we'd have like some local banks that'd be building some app or something. And literally they would have like two meeting rooms, one where they had a user that they recruited literally off the street. Like somebody would stand on the street from the, from the labs team and be like, Hey, we're, uh, working with so-and-so bank and we're experimenting with some stuff. Do you have a bank account? Do you do this kind of thing? Okay. Would you like some, uh, you know, gift card for a restaurant or something, uh, come in for 10 minutes to our office. We're just going to ask you to do a task. Right. Yeah. And like, they would literally have them. They'd have like a couple of webcams pointed at them and figure it like the user study. I mean, this is why I was saying earlier, I think the UX

designers are going to be the ones that define what an AI agent is because everything we just described right now is like, watch how the human is doing it and make their life better. This is classic UX work. And, and, and, and so I don't think it's all that different. Like the parts that make me cringe when people talk about AI, especially executives, especially from the, like the companies that should know better. Um, I, AKA, you know, the big giant, uh, hyperscalers, uh, when they make claims about how the AI is going to continuously do something for you, like detect that your customer is unhappy or handle it. And I'm like, you don't understand. I'm a human. I want to talk to another human. Right. Right. Like it's not, it's not like, that's the person that can make me feel better if I'm upset, not your AI. Yeah. You know, I, I like, like, like this is, this is a simplification that, that, and therefore makes it inaccurate as far as a hundred percent statement. I should know more concise way of putting that. But I think one, one, one helpful way to think about this question of how to figure out what to do with AI and enterprise is the interesting stuff is going to be in the, in the user experience. Right. Like, like that, sure. There could be stuff like, well, I don't know, like, like fraud detection is an interesting case, but even in that case, you probably need a user experience, right? Like, and so there's stuff on the back end that maybe is, maybe not so UX driven, but that could also just be the case of like, well, it is very UX driven. You're just not the person doing it, obviously. But, but I think probably, probably the higher priority, uh, the higher, uh, if you'll pardon the phrase impact that you'll get with, with AI stuff will have a tremendous, you'll, you'll end up doing a lot of user experience work because that's where,

where interesting things are going to happen. Right. And like, don't start with a Star Trek level assumption of what AI can do. Start with the more mundane. These are the constraints. Like I'll tell you, I have two favorite AIs that I interact with. Um, I think the first one pre-chat GPT that was like, oh, wow, this is really cool was in PowerPoint. Like PowerPoint has this little button like that you can press to spiffy up your slide. And I didn't know that this like AI assistant designers, I didn't even know it was an AI or what algorithm it used. But one day I had like the slide deck that was off of the white template in, in PowerPoint. Yeah. And I'm like, I pressed the button and it gives me some really good suggestions. I'm like, oh, that's clever. I, I, I select it and I go to the next slide and it gives me a different one. And I'm like an hour later, I'm like, oh my God, this presentation, like I would have never, ever been able to put this together. I like, I like, I like, you know, one thing I think that's happening there is an hour later. I bet a lot of that because you were very entertained. Yes. Like I'm going to keep doing this. This is wonderful. But because, because, because normally in my imagination, like I'd seen other people with slides that were not bullet points and not on a white background. I'm like, how'd they do it? They must be like talented in the arts and crafts, but more artistic than I am. That's one of my limitations. I'm not very artistic. Or maybe they paid somebody right to, to help them with it. But I'm like, like that was an amazing use of AI. Because it didn't take me out. It didn't go and make the presentation for me. And now when I'm going to have to make, I use it because it's kind of fun. And yeah, now I can do it not in an hour. Like I can do it in like five, 10 minutes, but like that is, it's not taking me out. And the second thing, the other one is chat GPT and that

kind of chat interface. And that certainly doesn't take me out as a human. Right. Right. No, it relies on you in fact, right? Like it can't talk to itself. Yeah. Yeah. Well, great. You know, I think, I think there's a, there's a, there's a key sort of principle here is look, look for the, the interesting AI applications at the use, the, the UX, the user experience, the UI layer, right? Like that's where some interesting stuff, a lot of interesting stuff will be happening. Not all of it, but like always kind of be questing for that, right? Like what's the interaction with a human that you're doing and, uh, how, how does that work out? Well, great. Well, we, uh, we, we have, we have plenty of upcoming items in our, our rough outline to go through. I think, I think we really, uh, we figured this one out pretty well and, um, you know, so, uh, hopefully over the next couple of episodes, uh, you'll, uh, we'll, we'll kind of get the verbal version of, of this, this booklet we have in mind written out and, uh, I think it'll be fun. Well, thanks, thanks for, uh, thanks for doing this today. Do you, you know, so, uh, I'll put a link to your, your, uh, your LinkedIn thing where you had a post on this, but do you have any, anywhere else that you want to point people at? No, at the moment, no, thanks. That's like, like, you know, in the past five years, well, three to three years, everyone is pretty much like, no, I just have LinkedIn. There used to be all sorts of other stuff, but now, uh, now that's it. What? Yeah. It's, it's like the, the shadow corporate world social network that I think is actually out there. All right. Well, uh, if you want to get those show notes, you can go to tomzutalk.com and, uh, find the episode and, uh, check them out. And with that, we'll see everyone next time. Bye-bye.

Tanzu Talk · 🍕 Get more pizza.