Insight ON "Data Is My AI Strategy"
How UC Riverside Is Outrunning Organizations With Bigger Budgets

Audio transcript:

"Data Is My AI Strategy" — How UC Riverside Is Outrunning Organizations With Bigger Budgets

Matthew Gunkel (00:02):

We're considering right now we're running full day long design sessions where we bring someone in like HR, finance, a school, a college, whatever it is and we're like, "What do you want to build?" And then we're going to live vibe code it during the session, show them, build interfaces, have them iterate on the, "I don't want it to do that. Can you make this button here?" Whatever the small changes are, basically take it all, package it and then go, "This is what you approved as your spec." And then we go actually use AI to build it. And that would've taken months just to have the conversation and we're going to do it in a day.

Jillian Viner (00:38):

A day to go from problem statement to working application with the stakeholders in the room. I don't know about you, but that sentence made me sit up a little straighter. Welcome back to InsightOn. I'm Jillian Viner and we're wrapping up our series on citizen development today with a conversation that takes everything we've been exploring. Individuals building tools, organizations, turning internal solutions into products and scales it into an institution. My guest is Matthew Gunkel, Chief Information Officer at the University of California Riverside. Matt has spent four years centralizing and modernizing IT at UCR and what he's building now is genuinely exciting. AI agents for student wellness outreach, vibe coding sessions that replace months of requirements gathering and a data strategy that's letting a public university leapfrog infrastructure that private organizations are still trying to untangle. Oh, and he has a very specific piece of advice about folders that I promise is more interesting than it sounds.

All right, here's Matt. Well, Matt, it's really nice to have you here.

Matthew (01:45):

Thank you.

Jillian (01:47):

Talking about technology with someone in the education space is always fascinating because I feel like you face a lot of different barriers, levels of enthusiasm, levels of discomfort. Tell me a little bit about your organization and the role that you play. What's been the journey since you arrived?

Matthew (02:07):

Yeah. So I mean, the journey really has been one of IT empowerment and enablement where we've really been trying to take a strategic approach to bringing in the schools and colleges and really all of the people at UCR in how can they really be more thoughtful about how we use and leverage IT. And that's really in all aspects of the business from really sort of basic operations to really more strategic things where people are like, "Well, I want to do this new thing." Or they come to us with a problem first and we're like, "Great, how can we help you solve that problem?" We had a maybe historical culture where people would come and they would be like, "Here's the solution." And we were like, "Well, that doesn't help us." We're like, "We really need to start at the problem and then how can we aproach a problem, bring the right tools, put it in the hands of your users, put it in the hands of the people that need to do it every day in a way that again, is empowering with IT."

Jillian (03:01):

What was the biggest challenge that you faced when you started your role day one?

Matthew (03:06):

I mean, it's a challenge that we still face today and it's actually increasing with AI, which is really related to change management. And so it's the pace of change. So things are just moving so quickly and even for the people that are extremely knowledgeable in the technical space, it's still fast. I feel like I'm just relearning, retraining myself every day. I'm asking my team to learn new things every day and that's just to do their basic job and that's different. I mean, don't get me wrong, I mean, IT's always been on sort of a, you have to continuously learn path, but the pace at which you were having to learn was a little slower. And so that meant that as we were rolling out new tools, services, et cetera, they got to go a little slower. And with vibe coding and with all these other things that we have that are coming out, we're now able to release new applications overnight or we're now looking at doing ... So one of the things that we're considering right now are running full day long design sessions where we bring someone in like HR, finance, a school, a college, whatever it is, and we're like, "What do you want to build?" And then we're going to live vibe code it during the session, show them, like build interfaces, have them iterate on the, "I don't want it to do that.

Can you make this button here?" Whatever the small changes are, basically take it all, package it and then go, "This is what you approved as your spec." And then we go actually use AI to build it. And that would've taken months just to have the conversation and we're going to do it in a day.

Jillian (04:37):

That's amazing.

Matthew (04:37):

Yeah. And so we're really making people's heads are like, "You can do that today." We're like, "Yeah, we can do that in a day now."

Jillian (04:43):

What are some of the biggest pain points or frustrations that you've solved doing that method?

Matthew (04:48):

Yeah. So to be clear, new. It's a new ... I will say that the ability in the tools and services that we've seen to be able to run that kind of session really have only existed, I would say, in the last 90 days. Which is

Jillian (05:02):

Wild, but this is AI.

Matthew (05:03):

Yeah, it is. It's wild.

So the fact that they're out there now, I mean, we're just starting to run those. So I would say we're at the planning stages internally. So we just ran a session for our own internal team two weeks ago and we already have people that are now coming out and going, "So I took this session, that was amazing." And now they're like, and now I'm doing this thing and they're now using Gemini or Claude or whatever it is. And they're like, "Hey, I built this thing." And we're like, "Wow, okay, that's fantastic. And then now can we pick it up?" So we're just kicking off some of the initial design sessions. Let's see, I have three in the next 30 days that we'll be working with. So people had use cases, they had ideas, they had problems that they brought to us and now we're like, great, this is what we want to do.

We want to go sit down, have this design session, show them what the iteration looks like and go from

Jillian (05:54):

There.What are those problem statements? What are some of the biggest, I guess, technology challenges that education spaces do run into that have taken months or maybe years to try to solve?

Matthew (06:05):

Yeah. So I think one of the things that's important to understand is that a university is a conglomerate of small businesses. And so we pretty much have every function when you think about ... So you'll be out in the business world and you're like, "Well, do you do this? Do you build buildings?" We do. Do you have facilities? Do you have finance? Do you do procurement? I was talking to somebody else the other day and they were like, "Oh, do you do point of sale?" And we were like, "Yeah, we run lots of point of sale." And so for us right now, some of the main challenges that we're looking at, one is related to procurement. So we're really looking at time to procure, how do we do security analysis and assessments as part of procurement and/or how do we do thoughtful risk mitigation and what does that look like?

Because security is becoming an increasing challenge in the world today and it kind of starts at the point of procurement. And so we're looking at ways in which we can do that analysis, do contract work and really manage the volume that comes through and using agents to augment sort of our human knowledge in that space, but also to enhance our speed on the security side.

Another use case, our student health and ... I'm going to get their title wrong. So student health services, it's wellness, but we're really working with them. They really want to do proactive outreach and so they want to help students and it can be with basic needs services, it can be with healthcare, different things that we provide for just all of our students that are attending UCR and how can we actually give them better insights and engagements with an agent to begin what's available. Previously, a lot of that was done with websites. So people had to go out, they had to read, they had to find, they had to navigate, they might have to go four or five click layers deep to get somewhere- So

Jillian (08:03):

Much work.

Matthew (08:03):

So much work. And now they're just like, "I want this thing or how do I get it? " And then they were just like, "Here are the people you call or here's the list of things." And in some cases it will actually help us with then doing proactive outreach where we actually can be more supportive of something that a student may need where they're just not really aware that the level of depth of service might be available to them depending on what kind of thing that- That's beautiful. Yeah, kind of challenge they have.

Jillian (08:28):

So those are problem statements, use cases that they're coming to you to help build.

Matthew (08:32):

Yeah.

Jillian (08:33):

Have you unleashed AI in other areas of the university where it's helped other departments be, I don't know, more innovative or unlocked in new capabilities for them?

Matthew (08:43):

Yeah. So the answer is yes in the sense that we have enterprise licensing available for, I'll call it broad AI. So we're letting our researchers and faculty go and play and use and get accounts. And so they can get Copilot, they can get OpenAI, they can get Gemini, they can get Anthropic, whatever it is and they can go and basically use those things in their daily work. And in that case, we're doing, I would really just say basic education on like what is AI and how do you structure things like whether it's gems or skills or I don't know, like micro agents that are now personalized and specific to your context. And so that's sort of like, I would say like 101 is just like, how do we help people do things like write better? How do they organize We're also a big user of Notebook LM.

And so Notebook LM providing a basic RAG model for like a set of closed data and information is incredibly useful in the education space. It works really well in the classroom space because the instructors then know that they're grounded in the information, which then makes them feel much more comfortable with whatever they might be doing. And so we're doing that a fair amount. We've had that on for, geez, we're at like two years now on Notebook LM. So we had it in Alpha before it was even released to the public getting started on work there.

Jillian (10:14):

I have to ask the obvious question on behalf of educators, do they come to you asking you like, how do I prevent my students from cheating? How do I make sure that they're not just using AI to do their work?

Matthew (10:24):

I mean, they do. They want to know. So yes, I mean, there's definitely this concern. How do we know that the student is learning the thing or the outcome that we want them to effectively learn? They're

Jillian (10:39):

Not outsourcing their thinking.

Matthew (10:40):

Yeah, exactly. They're not outsourcing their thinking. And there are a couple of different ways in which we can do that, but one is really leaning into a concept that we call authentic assessment, which is really taking this idea of we know the 10 things we want you to learn and we know them kind of at a very deep level from a knowledge perspective and how do we then assess whether or not you've learned those things? And interestingly enough, if you're doing authentic assessment correctly, then AI actually doesn't get in the way. I actually can tell in how you get responses. So if you think about AI as an augmenter of your existing knowledge, right for instance, if I went and started conversing with AI on neurosurgery, I'm now not a neurosurgeon, right? It's not going to make me a neurosurgeon overnight. Now, sure, you can teach me a lot of concepts.

I can learn a lot of things. I might get more than like a cursory level just from reading a website. Again, still not a neurosurgeon. Put a neurosurgeon in, they're going to get a level of depth that I'm incapable of getting because they understand the actual knowledge and that's what we're talking about is how can we actually go and assess that? And so we are working on new ways in which you can go and assess that and then effectively leverage AI. Now this does create disruption in historical testing. So you've probably taken lots of multiple guest tests and in those tests, yes. I mean, AI is very good at taking a snapshot and then going, oh, it's A, C, D, E, A again, and those are the answers. And then you're just like, "I didn't need to read anything." And so you may or may not actually know that knowledge.

And so again, it's different strategies for how we actually do assessment, how you tie things back out to learning outcomes. And this does require some additional maturity over some of what education may have been doing in the past, but that's okay. It's a good evolution.

Jillian (12:37):

Yeah. I actually, I'm going to take the optimist approach because I feel like in the past you were just memorizing things for a test and the framework that you just outlined actually measures how well I'm understanding the material and can kind of relay it back. So maybe there's hope that AI is actually going to make us better learners.

Matthew (12:56):

It's definitely going to bring depth, which I think to that regard is definitely going to accelerate what we need to learn and/or your pace at which you can learn it.

Jillian (13:05):

Yeah. Everything you've described so far is very exciting, optimistic, you're solving problems, you're changing things in good ways. What's been the unlock to make all of that happen? It's not just that you bought a bunch of AI licenses. How did you, I don't know, just get the infrastructure or the data or the capabilities to start doing this work?

Matthew (13:28):

Yeah, it's a great question. And I want to be clear, it's one that we're still actively pursuing. So when you look at like ... So I've been asked several times, what's my AI strategy? Data. Data is my AI strategy. It's not about the agents, it's not about the models, it's not about the things that they're working. If I can't get my data organized and contextualized in a way that an agent can effectively use, it won't matter how good the agent is because it'll still be garbage in, garbage out and it won't work well for us. And so we've been putting a lot more effort into good data maturity practices and really stepping forward in how we're actually leveraging and using our data and our information. UCR was in a position really where we've been able to do some leapfrogging where maybe they were previously behind in data.

And so that's actually allowed us to jump to new data warehousing concepts. So like when you look at vector databases, you look at graph knowledge, some of those kinds of things, we're able to start using those now because we didn't have some of the entrenchment in historical data warehousing, data lakehouses, some things where people were heavily focused on high structure and required a lot of work to get to high structure and now people are trying to understand how do you use organic information and much more unstructured information. So in a lot of ways, the fact that we didn't have high structure is now actually a little advantageous

Jillian (15:05):

Interesting. From your experience, because I feel like you're right about the data, that needs to be part of the strategy. What is a lesson that you would give to others who probably don't have the advantageous stance that you have, but what did you learned along the way that might help another CIO of another education group or even a business leader?

Matthew (15:23):

I think it's really sort of taking a step back and looking at what it is you're wanting to accomplish and then thoughtfully organizing around what AI then is good at and how do you step in? And I guess what I mean by step in is I hear a lot of people they'll provide these like, "Well, it can do, I don't know, whatever it is, everything." And it's this all seeing, all knowing entity. And at the end of the day, the use cases that are most successful are actually just to start with some very simple ones and again, get the data right around the very simple use cases and then you can extrapolate to larger execution.

Jillian (16:11):

For a CIO or an IT leader right now who's trying to make the case for major infrastructure improvements, maybe they've got a skeptical leadership team, what's an argument that you've seen work there?

Matthew (16:24):

So interestingly, I mean, our faculty are continuously asking for more GPU, CPU computing power. I mean, primarily we're trying to argue for time to science. So we're trying to really help everyone at the university understand ways in which we can accelerate the time to research, that time to science metric and how do you go about making a thoughtful investment in infrastructure or cloud or whatever it might be and then what is that return? We then usually look at that return in output with grants and dollars awarded and/or discovery. So were we able to go, like the big one would be, can you go win a Nobel Prize? In that case, I will say it's sort of a self-propagating thing. We're more constrained by physical constraints than we really are by demand. So I'm really not convincing people so much that we need more infrastructure. It's really more that we're tactically planning how and really working through prioritization.

Who gets to use it? Why do they get to use it? Where do we actually think leveraging it first is going to be more aligned to our mission, to our outcomes, to things that we want to do as an institution. We also do smart ag, we do quantum work and so it's like what are those areas and then which ones are going to have the sort of quickest or best return when we're putting it in our infrastructure. I mean, we have high performance computing and it pretty much sits at 90 to 95% every single day.

Jillian (18:15):

Since you are in the education space, I have to ask you, what is on thing that you think listeners should go educate themselves on right now tomorrow?

Matthew (18:25):

I think it's a great question and I'm going to answer with like the lamest of answers.

Jillian (18:31):

Blame

Matthew (18:31):

It on

Jillian (18:31):

Me.

Matthew (18:32):

Folders.

Jillian (18:33):

Folders.

Matthew (18:34):

Folders. What

Jillian (18:34):

Do you mean folders?

Matthew (18:35):

Yeah. So right now we hear everybody, they're like, "How do I build an agent?" Everybody's like, "Can I have an agent? I want an agent." And so when you look at the tools, when you look at the CLI tools, the sort of-

Jillian (18:49):

CLI meaning?

Matthew (18:50):

The command line interface tools. When you look at Gemini CLI or CloudCowork or I don't remember OpenAI's name, but when you look at these tools, one of the sort of first things that they do is the agent will actually look at how you organize your folders into a process or steps. And so this is just workflow and when you talk about it on the business side, it'll be like, "Well, I do step one and then I do step two. And then step 12, this is now the outcome. This is the thing that occurs." Previously, we would've built a lot of interfaces for this. So in ERP, you go and you build all these complicated web forms and then it's step A, step B, and then you get to approve something and they get to approve something and then who gets to move it next and so on.

It's the same with an agent, but we haven't yet explained to people how to go do this and you do this in folders. So folders actually break this apart into steps. You can create the markdown files, which are just readable text files and then that's how you actually start to craft an agent. And so this intermediate step where people need to take a process that they know well, they're not technical, but they need to describe it, helping them understand how to start to break it down into folders and text files that then essentially make an agent is the biggest thing that I think people can go work on and learn today because then when you're working with a CLI tool, it'll actually work well for you.

Jillian (20:23):

I'm going to go learn folders. Does it matter if I have more folders or less folders? I just need to understand how folders work.

Matthew (20:30):

No, not really. Again, it's more about sort of the thoughtful process and writing it down. And I think this is where people get hung up. They're like, "Well, I don't write code. I don't understand a CLI tool. I don't know why I would be in Claude code or whatever it is. " And at the end of the day, it's really, but if you go document what it is you want it to do in a thoughtful way, you can ask it to then translate it into its own narrative, its own language that basically like, "Hey, translate this for AI, write an AI markdown file and then let me read it or explain it to me. " But this is where it's a big opportunity for people to now be able to really build agents that are going to bring power. And I mean, there are verbose frameworks like Vertex where like, "Yeah, sure, you can go train your own data and do all these things." But the average layperson doesn't need to go train their own data set.

That's not really what we're talking about. I mean, in some cases maybe, but for the most part, most of us are just outlining process that we want to follow. And if you can thoughtfully lay it out in folders and markdown files, you can build a very impressive agent very fast.

Jillian (21:37):

Love it. I'm going to go take that away. Matt, thank you so much for the time today. This was lovely.

Matthew (21:41):

Thank you. That's been great.

Speaker 3 (21:43):

Thanks for listening to this episode of Insight On. If today's conversation sparked an idea or raised a challenge you're facing, head to insight.com. You'll find the resources, case studies, and real world solutions to help you lead with clarity. If you've found this episode to be helpful, be sure to follow Insight on, leave a review and share it with a colleague. It's how we grow the conversation and help more leaders make better tech decisions. Discover more at insight.com. The views and opinions expressed in this podcast are of those of the host and the guests and do not necessarily reflect on the official policy or position of Insight or its affiliates. This content is for informational purposes only, should not be considered as professional or legal advice.