Subscribe
Share
Share
Embed
AI is changing how we work, but the real breakthroughs come when organizations rethink their entire foundation.
This is AI-First, where Box Chief Customer Officer Jon Herstein talks with the CIOs and tech leaders building smarter, faster, more adaptive organizations. These aren’t surface-level conversations, and AI-first isn’t just hype. This is where customer success meets IT leadership, and where experience, culture, and value converge.
If you’re leading digital strategy, IT, or transformation efforts, this show will help you take meaningful steps from AI-aware to AI-first.
Matt Lyteson (00:00):
There's some people that are all gung ho and some people that quite frankly have a lot of anxiety about what this is going to mean to them personally, their roles, maybe their careers. And I think the real answer is none of us really knows. And while chief executives can give some blanket statements, yes, it's going to be the people that work with AI who are going to remain in here. I think just like in other massive revolutions and change, there are roles out here that we haven't even began to discover that we need.
Jon Herstein (00:35):
This is the AI first podcast hosted by me, John Herstein, chief Customer Officer at Box. Join me for real conversations with CIOs and tech leaders about re-imagining work with the power of content and intelligence and putting AI at the core of enterprise transformation. My guest today is Matt Zen, CIO of technology platforms with IBM, and I'm your host, John Herstein, chief Customer Officer. At Box each episode we talk with CIOs, IT leaders, and AI pioneers to share real world insights that you can act on today. Let's dive into today's conversation with Matt. Welcome Matt.
Matt Lyteson (01:13):
Hi John. Thanks for having me.
Jon Herstein (01:14):
Great to be here with you and we appreciate the partnership with IBM, which we're going to dig into in a lot of detail here. But let's start a little bit with, I think everyone knows IBM, let's be honest, but let's talk a little bit about your scope and role at IBM.
Matt Lyteson (01:29):
Thanks for that question, first of all, and I think IBM is undergoing a massive transformation right now, of course. And one of the things that we realized a little less than two years ago with, as part of this transformation, we need to double down on what we were doing with technology. And so my remit in that is basically to transform all of our core technology platforms, whether that's our hybrid cloud, our network, all the way up to what we're going to talk about today and how do we make AI real for every single I IBM ER driving both productivity across our discreet domain specific workflows as well as what I like to refer to as everyday ai, which are those tools that can help us be productive on our day-to-day tasks, even as we're executing our jobs. So a lot of stuff going on right now, a lot of ways that we're trying to bring AI really into every part of our operation to accelerate our growth. And
Jon Herstein (02:25):
I'd love to. Well, we're definitely going to dig into all those topics. I'd love to get a little bit maybe personal with you and just around AI and your own why for ai, how do you think about it in your own life and what you bring to it from a kind of professional perspective?
Matt Lyteson (02:40):
Yeah. Well, I think of course we will start with the geek perspective. This is some great new technology and interestingly enough, I found myself, it's been a few years since I've coded it being an accelerant for doing sort of side projects and hobbies and even helping my young kids learn, it looks like to do things like develop websites for example, to, Hey, give me some suggestions on restaurants to go to with the family because these are the types of foods that we like to eat. This is where we're going to be, want to make sure that we're going to a great place where we can enjoy ourselves. It's also those sort of the personal life from a professional perspective. I think it gets into how do we do things fundamentally different? You've heard probably analogies with this AI revolution like some of the other revolutions in the past, maybe even going back to the industrial revolution where we saw new tools becoming available, new technologies and how that was impacting the workforce.
(03:40):
Part of my role is figuring out how we deploy this at enterprise scale and touching every single one of our 280,000 employees from what they have to do every day almost from the back office perspective. Our CIO organization is responsible for running all the internal systems for IBM so that our product people and the people that are in client facing roles can do that the best they can. So it's always top of my mind, even before ai, how is the new tools that we're going to deploy going to impact the IBM ER and impact what they can do? And now we've almost got to think about that even more intentionally because some of these tools are unlike anything we've ever seen before, and I think that's really exciting. So we get to play around with them a little bit and my team, try them out, understand what the implications are going to be, and then help the rest of the broader organization go through that change as we deploy them. And again, as I mentioned before, this kind of is touching every different aspect of our operation from domain specific workflows to the tasks that we've got to do every day, like take meeting miniature, create presentations.
Jon Herstein (04:52):
Yeah, I mean it's sort of a combination of doing things we've always done, but better, more efficiently, whatever, but also probably more importantly doing things we could never do before.
Matt Lyteson (05:01):
Yeah, I think that's a great way to put it. And I think what we find in this evolution is that your first inclination is yes, I'm going to find a way to do the same things I did before only better, faster. It's almost like if you probably have an electric screwdriver at home, it's like so great. We can now take care of those little house world repair items much more quickly and not strain ourselves so much. So we've got a little bit of that going on, but I think quickly then you start to realize, especially with this new generative AI tool set that we have, that there are things that we didn't even really consider to do as part of our roles. And these are really the, and that comes to this AI story that I think is really exciting.
Jon Herstein (05:47):
And I think we'll dig into a few examples of that as we go through, but maybe we'll start at a much higher altitude, which is what is your vision for embedding AI into IBM's business and culture? There's a lot there, but how do you think about it? At a high level,
Matt Lyteson (06:02):
We like to say AI and everything are really AI plus in everything. And there's a couple different dimensions of this as part of IBM's overall transformation, we've been on this mantra, this flywheel approach that we talk about, eliminate, simplify, and automate because there are a lot of processes that maybe we don't need to do anymore. It's an opportunity to rethink those. It's an opportunity then to simplify the core processes because some things still need to get done. And then looking at ways to automate those. But you find very, very quickly when you go down that path that maybe there is a new and better way to do it. And that's where this AI plus, or maybe to you can say aiy everything across what we do. And what I realized early on what this is not just generative ai, I think generative AI got our attention in some interesting ways because it's like this, oh, never seen a tool that could do this before, but really we're talking about traditional AI with machine learning, traditional business automation, the business transformation and fixing our processes and then bringing this generative ai.
(07:14):
All these things need to come together in order to really redefine the way that a business operates. And so you may start with some of your core processes and how can I speed up the execution time of those processes? How can I maybe reduce how much it costs me to perform a certain task? We call this our unit cost, for example. But then you find yourself very quickly reinventing, and that's how we like to think about it here with inside of IBM and applying the right tools, technologies to be able to do all of that.
Jon Herstein (07:44):
So it sounds like the approach that you're taking, Matt, is so different from what we're seeing with a lot of companies that we work with, which is piloting with individual projects. This is a much more comprehensive all in we're doing ai. How would you describe that sort of difference in philosophy?
Matt Lyteson (08:01):
Yeah, I think so. And I think you hear our CEO talk about aspiring for IBM to be the most productive company. And as part of that, we started off by examining all of our processes and really this mantra of eliminate, simplify, automate, even before anyone was really talking about AI in the new sense with generative ai and what you find with that, yeah, there are some maybe processes that you don't need to execute anymore. There are ways that you can simplify those and you look through that on a business process transformation lens and then you apply business automation to the ones that you need to keep around in this new approach. But I think what you also find then is that now you've got this new tool to be able to bring to this. And with our strategy being focused on hybrid cloud and ai, we realized that we had an opportunity to put all of our AI on top of hybrid cloud and do it on an enterprise scale approach from day one.
(08:59):
Quite frankly, I had a little bit of anxiety when I saw that this generative AI was going to be the new rage. And remembering back probably about two decades when cloud computing was the new rage in the IT organizations, we found very quickly that because development teams and technologists and even non technologists in an organization could easily gain access to these cloud computing resources, they did that in the IT department myself and a lot of CIOs spent years then trying to clean this up. And some of us are still in the middle of it. How do I govern this thing to make sure it's got the right security data privacy? How do I control my costs, make sure I'm maximizing the value for those? We didn't want to have that happen with the easy access to these generative AI tools. And then also because you and I both, regardless of what our organization say, can pretty much go and get a free demo of any number of generative AI tools, how is that safe to use at enterprise scale? We set up some early guardrails to be able to make sure we are using a common enterprise platform built on our hybrid cloud and then be able to inject not just use cases, but looking at the end-to-end workflow and then able to measure the efficacy of what we were doing against that workflow. And I think that's allowed us to accelerate both what we've been doing with AI as well as taking a hard look at what's working and what's not from an employee experience perspective.
Jon Herstein (10:34):
And I definitely want to touch back on the employee experience stuff when we get a little bit later here. Is there sort of a risk here that suddenly generative AI gives everyone essentially a new hammer and now everything looks like a nail and you've got to put some sort of governance and controls around that?
Matt Lyteson (10:49):
I think we've definitely seen that, and I think that goes back to our core principle of eliminate, simplify, automate. Ask yourself, do I need to do this process? Is there a different way to do that? I think if you start with that lens, what we developed was really this pipeline effect where yes, we want to get ideas from across the entirety of the IBM enterprise because the top down perspective, we don't know how work gets done down to the minutiae and what's could be most impactful to every single employee. You get that and then put some light governance around it, not just from is this the right workflow to focus on or the right piece of that workflow to focus on, but also how are we going to approach this problem? Is this an area where we can do more or less ai, more or less automation based on the trust and the ethical sides of it, which are extremely important when you're implementing these, but then getting to that point where then you can almost mechanize this and go through that evaluation very, very quickly and make sure that you're focusing on the right use cases at the right time to have the most impact on your business.
Jon Herstein (12:02):
And this is one of the challenges that we're seeing with a lot of our customers is essentially every vendor, certainly every SaaS vendor is adding AI capabilities to their products. And so the challenge becomes, well, which ones of those am I comfortable with? What's the governance process around that? How do I approve those? What are the use cases? What's the business value? You can't just say yes to everything all at once. I don't think.
Matt Lyteson (12:21):
Yeah, I think that's exactly right. I think going through some sort of AI ethics review, we've got an AI ethics review that we go through. How is that vendor handling my data? Are they using it to train? For example, can they see any of my data, whether that's my props going in or whether that's the results coming out, want to make sure that if we're going to use this in particular areas, I may not want the vendor seeing that and other areas that may be okay for them to see that. And then can I trust the results? And I think some of this is the vendor evaluation really a lot of similarities to the way that enterprises evaluate vendors from a cybersecurity risk perspective, but then applying some additional principles. And then there's also the aspect of how do we teach and instruct our employees on the appropriate use for this? And are you evaluating what's coming out of the AI or are you just kind of copying and pasting it and sending it to your boss? And then you can imagine the ramifications of that if two numbers, the wrong numbers are summed up in the wrong way or the wrong analysis he's done. So I think there's a multifaceted approach that we've taken that helps us to navigate the complexity.
Jon Herstein (13:34):
So is there a core, we'll get into principles a little bit, but is there a core principle that you have then that even if you're leveraging AI tools, that you as the human are still responsible for the output for the deliverable?
Matt Lyteson (13:45):
That's absolutely one of our core principles and we see that as AI is complimenting roles in new ways and you need to understand how it operates. And even that was one of the principles that we adopted even with using traditional machine learning and getting advice on stuff that was core mathematics, like you can say, give me a distribution of what the s salary ranges are for people in my organization and based on their skills, give me some recommendations on who I might want to promote or who is the right person to give a raise to. We've been doing that for years and always the guidance was this is a recommendation engine and I think a lot of organizations have recommendation engines, but I think then it was easy to discern this is a recommendation engine, a human needs to take these recommendations, apply their expert judgment before making a final decision and submitting that same principles apply.
(14:42):
I think we need to be mindful of those because a little bit different scenario, and I think the way just as we humans think about things versus a decision that and a recommendation engine versus hey, generate a summary email of the conversation I just had with John so I can have a follow up with him. I think I don't know about you, but sometimes we're less inclined to, am I going to read through every bullet that came out of this meeting summarization email the conversation that you and I had in the same level of detail as I would be evaluating something that is clearly marked as here's a recommendation engine for some compensation decisions,
Jon Herstein (15:22):
Right? Yeah, yeah. We're all sort of figuring out what we can trust, what we can't and what we still need to do in our roles that can't be just replaced because AI summarized it for us. You can't just do that with everything. I want to pivot a little bit. You've mentioned the hybrid cloud a few times and obviously there's some technical components to that, but there's also some partnership components to that. And I'm just wondering if you could explain a bit more about how you think at IBM about hybrid hybrid cloud hybrid capabilities and what that means from a partnership perspective. And part of the reason I ask is obviously IBM is at its core, a deep deep technology company. You've got all sorts of technical solutions, but you don't do everything yourselves. So how do you think about hybrid partnerships and what makes a better together story for you?
Matt Lyteson (16:08):
Yeah, I think first of all, when we think about hybrid cloud, I think we've recognized for a long time now that most organizations are going to have workloads that operate their business and support their clients in a number of different physical locations. Some of that is public cloud on the hyperscalers, some of that is in a data center. And whether you call that private cloud or maybe you haven't modernized it enough to call it a private cloud, all these things need to come together using a common substrate and a common operating framework in order really to get the best value. Am I writing the right workloads and the right locations at the right cost profile to optimally have them perform? And that's what we think is hybrid by design and really being intentional about how you set up the substrate of your hosting environment. For us, that's obviously Red Hat OpenShift is a key component of that combined with our own IBM cloud.
(17:07):
And then on top of that, then you're able to layer the different types of technologies and application workloads to support your business and then you start to integrate things as your software as a service provider. So that's in a nutshell why Hybrid Cloud helps us to optimize workloads and run them again the right place at the right time with the right security profile, the right performance really in order to maximize the value that we're getting from that. Can't do that in a by default load. And then of course, as you highlight it out like to see is where does IBM have those core competencies ourselves versus where do we need to go out and partner with other organizations that expands into things that are running in public cloud on their own box, of course is a key partner for us for our unstructured data and how employees collaborate using that unstructured data. Also SAP, Adobe, Salesforce, ServiceNow, some of those other key partners who are helping with those core software as a service capabilities, our core business platforms that run the rest of the operation.
Jon Herstein (18:14):
And in the case of Box, we have this very interesting partnership where you are leveraging box and box AI around your unstructured content, which in turn is leveraging Watson. Right? So it'd be interesting to talk a little bit about that architecture and how that benefits IBM M and IBMers and then ultimately your customers.
Matt Lyteson (18:33):
Yeah, well I think it's a great use case of, I think there are certain areas within an organization where you want to leverage a core capability like box rather than building your own. I don't think anyone, most organizations aren't in the business of building their own unstructured data sharing platform. There may be a couple examples, and again, back to the statements about hybrid cloud. There may be use cases where, hey, this data is so special to my operation, I needed to operate and be in a tightly controlled area that can't be anywhere touching the cloud. Other than that though, I want someone like Box to be able to handle that. And increasingly we are expecting these AI capabilities as part of that. And so box embedding the Watson X platform underneath, so I understand a little bit better what's going on underneath the covers. That gives me a sense of trust and security, especially when you start using the trusted LLMs to perform the actions, but then similar to any other software as a service, you're going to provide me the core capabilities that I want so I can focus on the higher value things.
(19:40):
And so we look at how do we leverage box AI and plug that into the workflows that I talked about before, value add, what does that mean? That means I don't need to focus on where am I going to put that on structured data that's required for this specific use case that becomes part of our core architecture. And then plugging into that, you can imagine that I need to pull documents out of a folder inbox. I need to combine that with other information that's coming from the ERP system. I need to synthesize that together in order to perform some analysis or produce some other output for a specific workflow. Fits in great with cases like that.
Jon Herstein (20:20):
And how do you think all of this is translating into the experience that your end users actually have? When you think about AI enabled experiences being very different than what they had before, how do you see that today and evolving in the future?
Matt Lyteson (20:35):
I think there's a couple things, and I think we've started to have this realization over time and it started to become very intentional about it. Let me rewind A couple years back, even five six ago when we talked about design and when we talked about the user experience, everything was about let's do the journey map of how someone is interacting with a mobile application or with a website and what are the best practices around the way that that application site is laid out or the mobile experience is laid out. A lot of effort was going into that. Now if you think about it, what do our AI powered experiences look like? It's basically a prompt,
(21:19):
It's a natural language prompt. And so you've got a couple aspects around this where I've got this interesting dimension of not everything is going to be generative ai. When you're operating in a business context, you need to get deterministic information managing the prompt. We've got a tool and a digital assistant called Ask hr, which has generative AI capabilities. It also has workflow capabilities. So more of that business process automation say, how many vacation days do I have left? It's got to know who I am. It maybe is going to synthesize that into a summary answer for me based on my unique circumstances, but it's got to go and look me up. It's got to look at the deterministic vacation policy. It's got to look in our HR system to find out how many days I've taken already for this year. Very deterministic. If you go out and ask a general LLM, how many vacation days do I have left? It's probably going to try to answer you
Jon Herstein (22:20):
Right, and it will answer you confidently and incorrectly.
Matt Lyteson (22:25):
Yeah. Yes, exactly. And you can imagine I'm not an HR professional, but you can imagine the potential ramifications for that. So understanding first this deterministic aspects versus these non deterministic aspects of it. So you are going to require humans to input very precise information in some of these workflows as they're interfacing with this natural language prop. How do you do that in a way that is not going to be overly burdensome? And we had an interesting example early on where we are trying to automate one of our workflows, and I was talking to the manager of the team in the hallway. He's like, Matt, we are finally able to simplify this because when we put this workflow that was asking for 40 fields of data entry for the user in this prompt window, we realized that that was a bad idea. Seems obvious retrospectively, but that's what the team would do.
(23:21):
Let's take what was in our webpage, let's make sure that information could be inputted into our digital assistant, which again is just a single prompt interface. It's silly with that. So now you need to rethink about what that experience is going to look like. And I think it forces two things that we found. Number one, it forces us to re-examine the overall process even if we haven't done that in advance. Perfect world, yes, would examine the process and kind of the stepwise way that I articulated earlier, but if you don't, you're going to catch it before you go into production. Like this use case that I just described. Of course it's silly to have 40 data inputs in there, but now you're forced to go even further. How can I preempt even having to ask the user about a particular data field? Is there something else that I can access in the environment maybe from another system?
(24:14):
Great example is how many times you go and interface with a digital assistant and it says, what's your email address? Well, who knows who I am? You'd think that it could go out to the employee directory and get my email address and then not even ask me that question. So it's a fundamental rethinking of this, and this becomes extremely important for us because that's really where people are going to find the value and how they use these tools both from a workflow perspective and then from this every DI perspective. So that's kind of what I would call bucket one and how we think about the user experience. Bucket two that we've been really intentional about is understanding that this is a new mode of interaction. Similarly, I had ANT experience early on in my career and we were modernizing systems off of green screen into web-based system.
(25:07):
This was early days of web. I was at a different company and a client facing role modernization was all the rave. And we did all this work to develop what we thought was this awesome beautiful website. And I remember going, we were doing some user training, I think this was some sort of procurement system, and we were getting feedback from the users and one of the women came up to me and she's like, this is awful. She's like, this is slowing me down so much. I'm like, what do you mean? This is a great web-based interface that we made. You don't need to remember all these fancy key combinations on your keyboard in order to input. And she said, it is so slow to have to use the mouse on this, on all the fields. And I'm like, okay, that makes sense. So just the time and the speed of entry.
(25:56):
And so then there was this acclimation period, there's this acclimation period in bucket A that I described. How do you design these experiences? But if that's the way the world was going, she and her colleagues eventually learned how to use these tools. We learned how to, okay, you can tab through the interface rather than clicking on the mouse. My point is we need to make sure that employees are acclimated to this technology. And so one of the approaches we took early on, this was back in 2023 right after our Watson X products were announced, we said, we need a generative AI experience for all IBMers to do and get them acclimated to this generative AI technology. So we took watsonx, we created something that was called Ask IBM. We trained it on our intranet data so people could go in there and ask it. I am getting ready to have a meeting with this important client.
(26:51):
His name is John. Draft me a summary agenda that I can use about the latest in IBM's AI strategy, and it would pull the most recent information from our intranet draft. A nice little summary presentation again generative and we made sure it would not answer HR questions. Like I mentioned a few minutes ago, we did this very intentionally in order to get people the experience and engaging with these tools in a trusted way in our environment that was very specific to our organization. I can't get that from going out to chat GPT, right? It's trained on the intranet and world data. I can't get that from copilot in a way that I can guide it and say, Hey, here's the corpus of information I want you to look at and bring back the latest on our strategy. Or I use it personally every week to help me write my blog to my team because it can go, it sees on the internet. Here are the last blogs Matt wrote to his team. So it's starting to understand about my strategy for my team specifically how that relates to IBM strategy. I of course procrastinate on writing left blocks.
Jon Herstein (27:59):
It helps you get it done faster.
Matt Lyteson (28:00):
It helps me get it done faster so I can go in Monday morning when the person on my team who's helping me to do this, Brie is pinging you, Matt, where's your draft? Because she wants to get eyes on it. And what used to be just procrastination and I would have a couple sentences, it writes it, I go and tweak it to make it sound a little bit more like me. This is getting better. Send it over to her and away we go. But it is all about this experience and getting people used to using these tools. And then the secondary aspect of it, which we're only starting now to talk about, is that we realized that there were going to be hundreds and hundreds of these digital assistants. So where do I go to get an answer to a question, which digital assistant do I engage with?
(28:43):
So this is starting to be the funnel, the single point where you can go to where it will direct me to the ASK HR digital assistant. It will direct me to the Ask Sales digital assistant. And even we're starting to connect in box AI for that for content that is hosted in box so that you don't need to go to these different tools or different digital assistance to get access to the information and the answers you need, whether that's generative or deterministic. But I think these examples and then being very intentional about how we're going and promoting what's working well for people have been important aspects. Back to your original point of how are we thinking about the user experience through all of this,
Jon Herstein (29:24):
Do you see things evolving into the point where, I mean I think this is where it's going, but where in a lot of cases a user will never go into an app, there won't be an app, there'll just be an interface that's chat or whatever makes the most sense, that goes and figures out where the data pieces it together and brings it back and you won't be thinking about things like which field goes where in a web-based screen because that's just not the way people will interact. Do you think it goes that far?
Matt Lyteson (29:51):
I think it definitely does. We've already done that from an HR perspective where 90 some percent of our employees never go into the actual HR system,
(30:00):
And we were able to swap that out behind the scenes and basically tell no employees and launch at the beginning of this year because we've got Ask HR that handles some 70 different HR related workflows behind the scenes. So you go and ask hr, I need an employment verification letter. Boom, here we will download that. So of course it's going to the backend systems, it's looking that up, it's applying it to what's the latest IBM template for that, et cetera. Also, I need to transfer John to Mike John from Susie to Mike, transfer employee to a different manager doing that and interfacing with the assistant. So the use cases, and again, some of that even started before the generative AI range. Now what we're finding that it's easier to get them done. If you know the data and the systems that you need to interface with and you understand what that workflow needs to be, you can develop some of these very rapidly. So I definitely think that's where it's going. And I think you can imagine all of us just interfacing with some chat agent on our phone the same way that you and I may text to get information from each other to say, prepare for this call interfacing with an assistant. Then it's going to be tied into your organization to perform any number of these types of workflows.
Jon Herstein (31:18):
These are what we've been talking about are mostly user-driven reactive workflows, like someone's asking for something and it gets an answer back. But then there's also the idea of agents starting to work autonomously to just go get things done that need to get done. And IBM is obviously going to play a very important role in the orchestration of all that. So can you talk a little bit about how you all are thinking about orchestration of these sort of workflows and agents, and maybe not just for yourselves, but also for your customers.
Matt Lyteson (31:45):
Yeah, I think the orchestration is absolutely key to us, and I think a lot of CIOs are looking at it from this perspective because the reality is, as we were mentioning earlier, there are some ais that I'm going to get from my strategic business platforms, my ERP, my HRMS, my IT operation platform, my marketing platform, all that's going to be baked in. Then I've got this other ai, like we talked box ai that's going to have some corpus of data here. That's very important to my orchestration. Those inherently aren't the end-to-end workflows because you imagine even a seller having to go out and prepare for a client conversation, they will kind of want to know what are the open opportunities they want to know. So I can get those from my CRM system. They want to know what we did and do we have a two-way relationship with that particular client?
(32:37):
This needs to come from my ERP system now I want to know what's the latest on any interviews or podcasts that the specific person is that I'm speaking to on what's the health of the relationship coming from the client interaction system and the client support system. If you want to be able to bring all that together and synthesize it very, very quickly, and the only way you can do that is through an orchestration layer because everyone's going to have a little bit different ERP and A CRM. They're going to get whatever is best for their organization. And so this is where our Watson X orchestrate really starts to enable us to bring those together in how we define the workflows for our organization or how you might define the workflows for your organization where, yeah, maybe we're using the same CRM, are we using the same ERP? We got to plug into these. You got to get data in a little bit different way and contextualize it for the enterprise. I think this is where orchestration is really going to take this to the next level. It takes this sort of general purpose tool and says, okay, here's something very specific I need to do with that tool and how it plugs into the way that my business actually needs to operate.
Jon Herstein (33:53):
And do you think that, we'll, I mean put your CIO hat on really explicitly here. Do you think that we evolve away from API based integrations and connectivity to agent-based where you could argue it'd be a lot more flexible, less brittle, less subject to change to connect things together? Or is it too soon to sort of imagine that?
Matt Lyteson (34:14):
I think there's probably a number of different schools. I don't think APIs are going away.
(34:19):
Just to be quite bluntly. I think where we've always struggled with APIs, you imagine we're also caught up in this rave of service oriented architecture back in the day, that was again the latest tool that was meant to solve all the problems, and it just became too complex and unwilling, and I don't think a lot of organizations really got to the full adoption. I think now we've got technologies that make that a lot easier. I still need to go and know what is the definitive source of truth for my client data? What is a definitive source of truth for my employee data? And that's where APIs become more important. I think what we find then is we've got this trusted data source. So we mentioned hybrid cloud before I mentioned an AI and automation platform. You can almost think it's parallel to that. You've got to have your enterprise data platform at the very least. Do you know what the core sources of truth are for key data that's important to your organization? We've got this set up with our chief data office, so I know if I want a customer record, where am I going to go that's going to be accessed through an A p? I think what happens then is the agents overlay this in the way that we're thinking about it. So I have atomic units, so the agent that give me the latest customer data,
(35:39):
It's almost like an API. But if I've got these atomic agents that I can plug into an orchestration layer, yeah, I can have my top level LLM figure out a step of task to execute, and now it's got these set of tools in its toolbox. One of these may be go and get the latest customer information and ract from that, who's the customer contact? The customer contact is John. Okay, I've got another agent that says, give me the latest public available information on John and some of his latest thoughts on ai. Now we can combine those together. So I think we're going to have these lower level agents that are almost that abstraction layer to those APIs, if you will, and able to assimilate that, that then get attached to higher level agents that can then start to get orchestrated together in new and different ways,
Jon Herstein (36:30):
Right? Yeah, so you can think of lower level agents, just go get this piece of information that in a way that maybe you would've done through an API or through a user interface before, but then that aggregates up to a higher level orchestrated agent that's doing a whole workflow for you.
Matt Lyteson (36:45):
I think so, and that's certainly the way that with my CIO hat on that we're thinking about it because then it becomes a very tractable problem to solve, and then I can focus on these if I can get access to all the information and I can do that in a tool-based gentech approach, and then I can use the common architectures to plug those in. I can have an agent, maybe you expose an agent from box where I give it some context and search information that says, Hey, bring me back my biggest problem, find the latest IBM template for PowerPoint presentations that's acceptable to use externally and combine it with this other information. I mean, that's probably the number one impediment when I'm going to create a client facing presentation that's new because everyone that everyone else gives me is a little bit different than the last version that I use. But if I've got that repository box, I can call box agent that says, go and get this,
(37:39):
And then I can orchestrate that workflow together with maybe some of my team strategy. We talk a lot about client zero. There's their conversations that I have frequently with our clients. Then it can be an accelerant, but it's got to kind of be at that lower level because then we've almost got this composable architecture. And I think then as that gets smarter over time and we as humans learn how to better orchestrate that, that also gives you the adaptability to the way that your digital operating model can actually function because of maybe this today, but maybe tomorrow it needs to operate slightly or I can skip a step based on certain conditions. If it's at the atomic unit level and we're thinking about the agents starting to orchestrate that at a higher level, then I'm not changing these low level building blocks all the time or what I do. It's much simpler to change that, and then it just plugs into a higher level orchestrative flow.
Jon Herstein (38:31):
And I think the real power here is when you start to combine the information proprietary to your enterprise, like you said, your PowerPoint template, right? That's been blessed by your marketing team with all of the other knowledge and information that's available publicly, bring those things together. That's where I think it gets super, super interesting and exciting.
Matt Lyteson (38:49):
Correct, and I mean you and I both do that. We're going in to talk to a client or a peer and you do research on them, and what do you do? You go into your favorite search engine and you pull that up and you're like, I think even when you met, Hey Matt, here's what I saw you say in some of the previous podcasts or did some blog posts. We all do that, so why not make that easier? And so I think that's exactly right. You got to combine the different sources.
Jon Herstein (39:16):
Yeah. Can we talk a little bit about metrics and benefits that you're seeing? There's obviously huge amounts of investments going into AI in IT operations and everywhere else. We talked about Ask IBM is one specific example. You mentioned the reduction of, I think 90% of people having to go to an app versus being able to just go ask What other benefits are you already seeing in the business today with ai?
Matt Lyteson (39:45):
I think there's a number of these in the way that we look at first. I think we've progressed on this journey. I think everyone has talked about hours saved and then you try to have a conversation with C ffo about hours saved. It's like, what did you really do for me today type situation. So we've tried to move past that. Look, don't get me wrong, I think hours saved are extremely important and I think all of us need to learn how to work and interact with AI on a day-to-day basis. But I think really when we're articulating the value, it is got to come to the bottom line. And one of the ways that we found an effective way to communicate this is either per my unit cost of the output of my workflow, think of what does it cost me to produce an invoice?
(40:30):
What does it cost me to spin up a virtual machine? We've been very good, I think over the past probably decade in talking about this in terms of it sense, I should be able to tell you how much it costs me per virtual machine per terabyte of storage. That way, if our business is growing, we can have a strategic conversation. This is what it's going to cause versus Hey, Matt, just cut, cut, cut. Am I able to optimize within that if I'm looking at through that lens? Yes, absolutely. We use, for example, AI with our devices and getting telemetry all by devices. I mean the laptops that we give to every I, ibm, er seeing significant reduction in unit costs for that, almost to the point where we think we're by the end of this year, we'll be below some of the external, the bottom point of external benchmarks that we can find publicly available while still giving high capability machines to the people that need them.
(41:30):
This is a little bit different approach where I might be issued a device based on the specific job role that we're in, and then beyond sort of a 3, 4, 5 year cycle. Of course finance team always wants you to expand that one more year. We're like, we can do better. I'm going to give you a device when you need it, and I'm going to see what you're using it for, irrespective of what your job role is, but what are you doing on a day-to-day basis? So if all of a sudden, John, you start doing hardcore AI coding, instead of maybe doing presentations that you were before, we're going to notice that and then preempt that and say, Hey John, we're going to send you a new more high powered capable device because we notice you're doing different things. AI helps us do that. That's less generative ai, but that is still AI and learning from that and adapting.
(42:18):
So that's one of the use cases. Another one, we start to think not just about the unit cost, but the flow velocity is the other metric that we're using from our digital capabilities perspective. She can think of how long does it take someone in our procurement operation to prepare for a meeting with a supplier? So they've got to again, do that research kind of like we described earlier. And if you're talking to a client, similar supplier, I want to get all the summary of the recent transactions, what that looks like. I want to risk score, I want some external information. Now we can produce a supplier brief in about five minutes. So tracking that type of value that then ultimately will translate into something that A CFO can recognize in terms of what is this doing to our aggregate cost structure in the particular functions that we're interested in.
(43:12):
So those are just a couple examples that we've got deployed and we're seeing I think huge benefits, but we try to be very intentional when we're looking at these use cases when they're coming in, is this just going to be an hour saved? Okay, if it is, let's acknowledge that and not say that there's going to be huge financial savings taking out of the bottom line. If it's not, let's try to pinpoint what that impact is going to be. I think that's again important. So we don't do the wrong behaviors. Obviously I could reduce my aggregate cost of devices if I gave everyone the cheapest device having a technical population, you and I both know that's a silly idea even without knowing the details of my operation. So how do we do that in a much more nuanced, intelligent way?
Jon Herstein (44:00):
And all of these things have a direct impact on employees and employee behavior, but also end users that are outside the enterprise and so forth. So one thing I would love to chat about a little bit is how have you thought about all of this from a culture perspective and a change perspective, and how has your office addressed that? What kind of support do you have from the rest of the business in driving that kind of change? And has there been resistance? And anything that you can give the audience in terms of practical tips on what to do and maybe what not to do around change management would be super, super helpful. We all know the technology is almost the easy part of this in some ways
Matt Lyteson (44:35):
I think that's spot on John, and we learned some surprising things. I'll share some personal experiences first of what we learned early on, we were, and I'll start with the traditional and then go to the generative ai. We were getting ready to implement one of our solutions that basically looks at our OpenShift containerized environment. This is Omic and it will fine tune the allocations for virtual CPU. And virtual memory team was like, we're really tight on this. Here's our excess capacity. And I'm like, let's turn this on in the production environment. They're like, no, we went on for weeks and weeks and finally I just made a call. But their reticence to that, they would be more willing to trust the junior engineer coming off the street to make the allocations versus trusting the AI to go and do that as it's learning to monitor performance.
(45:27):
So that was a little bit of a surprise for me, and I think kind of coaching the team along that journey and understanding what they were really concerned with. And the concern wasn't, we weren't going to see in results, but it was really we're going to have issues with some of the applications if we constrained them too much because, and what we learned is that we had to work with the product team to fine tune the AI model. So it was taking into account the startup costs. We had some applications spiked memory in CPU when they started up and then went down to a flat level. If you're only learning on that flat level, you're going to have a problem when you try to start up.
(46:05):
So this is kind of dig, where's the resistance coming from? Another great case is we've got to learn from each other on this. We were early adopters of Watson Code assistant for Ansible, also called Lightspeed with one of my SRE teams before we even released it as a product. And the team was like, this isn't really showing us any benefit. It's helping some of our junior developers but not really helping our senior developers. We are in the midst of a significant automation push and using Ansible as part of that for all of our hybrid cloud infrastructures. So you can think of how I set up, how I configure and maintain the state of my containerized environment and any virtual machines that I may want to have. I'm like, this was really quite frankly disappointing. They came back to me like a month later and they're like, Matt, we figured something out.
(46:58):
So we created a library of props that people have found are the most useful for them, and then the productivity kind of went through the roof. So just sharing those knowledges and experiences with that. So the other big push is we've done as part of our Client Zero initiative Watson X challenge days every year for the past couple of years where we get the entire employee population playing around with the tools from a raw product perspective like watsonx, orchestrate watsonx ai, as well as the things that we build on top of them, like Ask IBM and developing their own workflows for their teams. So getting employees exposed to the technology, as I mentioned, sort of one of the primary criteria for developing Ask IBM to begin with, but then also giving them the ands to play around with these and have the ideas that they can come and get into our, let's get this on a path to production pipeline so that we can realize enterprise value for that.
(48:02):
That's been extremely important. And then something that we've started recently because quite frankly, we looked at some of our adoption statistics for our everyday AI tools, and you can usually see when we send out a blog, okay, the week after, there's a pretty big spike, whether that's box ai or when we turn on copilot chat or implemented some new features with Ask IBM see a big spike, and then it kind of dwindles off. People kind of get tired of it unless they have to continue to use it. We've now been on a push to really show people what's working in new and innovative ways. And I think you find back to some of or earlier conversation people start, well, how can I do things faster than I do today? I got to send out a recap of this meeting that I was in. So I'm going to take my notes and I'm going to ask AI to summarize it for me to, okay, how do I ask it and question.
(48:58):
And I was showing some people this. I think one of the things that your team showed us was that number one use case for box AI was Excel spreadsheets. Interestingly enough, I'm not a spreadsheet fan myself, but I was synthesizing some data. I had a spreadsheet in there and I said, give me a summary of this data Did a great job. I said, okay, now I want you to group these into categories. I briefly explained the categories and then was able to group it in a different way. I went back and validated it back to earlier points of the conversation because I don't want to send it to my boss or outside of my laptop to someone without having validated that it was synthesizing it correctly. And then I ask it, well, what else can you do for me?
(49:43):
Well, I can give you a way that you can export this and bring it into an actual database, or I can generate an interactive web HTML webpage for you. Interesting things that as we get better at this, and I shared these with the leadership team and some others and light bulbs are going off. So just those experiences that we are happening, I think that is all part of the change management that we as CIOs need to lead our organizations on and get the proponents of the technology for this. Because John, I think you're seeing with a lot of the clients and maybe in your own organization that there's some people that are all gung ho and some people that quite frankly have a lot of anxiety about what this is going to mean to them personally, their roles, maybe their careers. And I think the real answer is none of us really knows.
(50:37):
And while chief executives can give some blanket statements, yes, it's going to be the people that work with AI who are going to remaining here. I think just like another massive revolutions and change, there are roles out here that we haven't even begun to discover that we need. I mean, think what we talked about earlier on the AI ethics and prompt engineering was all the rage. Now we're all prompt engineers. There's going to be new roles and new skills that we need. And so I think it's important, those of us in leadership positions to help lead our organizations through these, understand what's working, what's not, understand where there is some extra anxiety and what we've got to do to really dig a couple layers deeper on that. The anxiety my team was having could be pretty big issue if we have AI tuning the CPU in memory and then it takes down an important production system at the wrong time. So understanding and talking through that so then we can move further on the journeys. All I think critical elements of our change management approach,
Jon Herstein (51:43):
There's a lot to consider here. And I think until the light bulb goes off for each individual person, like you said, you see a spike in usage. It's like, oh, here's this new thing. Great. Let me try it out. And if it doesn't really click for you, the benefit to you, your job, your career, et cetera, you sort of move on. And I think people seeing those unexpected insights that are coming, because the AI has the context, you're prompting it, but it knows a bit more and is able to bring that value, suddenly you go, oh, there's a lot more here to it than just answering a direct question from me. And I think that that's what we're going to start to see in our world in particular, with all this unstructured content customers have, suddenly you can get insight from it that you would've had to go read the document right word for word to really get that you could do that a much easier way. And when that clicks for people, I think adoption starts to really trend up
Matt Lyteson (52:32):
Well. The clicks and I would say is back to the user experience when we are as IT professionals, either developing or develop or just deploying, making sure that it is good enough. Because in some of these things, you kind of only got one shot and it's got to be good enough and people have got to see the value because if they don't, yes, we can continuously iterate on that, but if it's giving the wrong answer or spurious answers and we're just kind of laughing at it, it's going to be a long time before we go back to that. And so I think that's where understanding what that experience is from the get go, again, good enough. And I think that good enough is looking a little bit different from an enterprise perspective than it has before because everyone's using other solutions in their at home world. We started off with where they're expecting the enterprise things to be just the same, if not better.
Jon Herstein (53:26):
So I know we've had a great conversation, and I'm probably coming up on time here, but I want to wrap with three things that I think about a lot in my role running customer success at Box and their value, culture and experience. And we've touched on aspects of all these in here, but I just wonder if I can get a few final words on these, but what do you see as the critical path to value realization on using ai? So how do we deliver value around these things, not just something cool and new, the important criteria for influencing culture and then the end user or employee experience around this. So there's a lot there, but when you think about this from a practical application of these things, for all the folks audience, what are maybe one thing they should think about for each of those categories? Culture, value, and experience. I
Matt Lyteson (54:16):
Think value, I would start with being clear and articulate on this. And look, if you're just going to save someone hours, I think our friends at Gartner call this productivity loss. You're going to go to the water cooler as probably am I, if I'm saving a couple hours writing my blog, that is different. Be honest about that. That's different from am I impacting my unit costs or my flow velocity for my process that I can translate into an actual dollar that I can show my CFO. So being clear on how you're articulating that value and what you're going after with that is the point on value.
(54:54):
I think on the culture piece, this is really, we need to learn together and really promote what is working, what's the tip that you can give me that maybe I haven't thought about and vice versa. I think people are being creative about this and getting them more comfortable. And then again, understanding their anxiety at the same time and what's holding them back, even from using the tools critical element of the culture that I would recommend. And then the last element, it's got to be easier, and I think this is where we're really having heart to hearts within my organization because it took us, I would say, a long time to get to the point where we are even focusing on the user experience with how we're doing web design and mobile design and things like that, unless we were doing externally facing. But I'm kind of thinking about the internal IT departments and now we're in another fundamental shift where I am literally talking about it as what is the human experience?
(55:58):
Because back to your point on agents, agents are going to be interacting with these things and there's going to be an agent experience. There's also got to be a human experience. Understanding what that looks like, using that as an opportunity to simplify and automate as much as you can so the sellers can sell, so the builders can build, so the customer support people can focus on customer support and get out of all these tasks that add friction and make their lives more difficult. That's what we need to be thinking about from the human experience perspective.
Jon Herstein (56:30):
That is such a great message to end on, right? This idea of sellers need to sell, builders need to build, and what our role is to provide the technologies and tools and platforms that just allow 'em to go do that, right? They don't want to get bogged down in the details of apps or interfaces. They just want to do their job and do it well and do it with quality and efficiently. And that's, I think, the core reason why we do any of this stuff. Right.
Matt Lyteson (56:56):
Well said. Well,
Jon Herstein (56:59):
Well, Matt, I really, really appreciate a lot of things. One, the time that you spent with us, I think we could easily spend another couple hours talking about these topics, but the long partnership that we've had with IBM, both as a partner and a customer, and we know that we'll go many, many years, many, many years into the future and the way that our technologies are working together for the benefit of IBMers, but also our joint customers is something we're really excited about at Box. So thank you again. We really appreciate Matt IBM for contributing to the conversation. And if you find value in this conversation, please subscribe, share it with your professional network, and we've got additional resources and companion materials available to you on box.com. And please join us for our next episode. We're going to continue exploring artificial intelligence, content innovation, and the evolving landscape of enterprise technology. Thanks all. Thanks for tuning into the AI first podcast, where we go beyond the buzz and into the real conversations shaping the future of work. If today's discussion helped you rethink how your organization can lead with ai, be sure to subscribe and share this episode with fellow tech leaders. Until next time, keep challenging assumptions, stay curious and lead boldly into the AI first era.