Agentic AI Is Coming: What State Governments Need to Know Now

Amy and Alex discuss how AI is moving beyond generating text to executing real tasks and workflows. In this episode, we unpack the rise of agentic AI, what it means for state governments and how technology leaders can prepare for this next phase of AI innovation. 

Find the report here: https://www.nascio.org/resource-center/resources/beyond-generation-the-rise-of-agentic-ai-in-state-government/

Transcript

Amy Glasscock 0:05

Hi, and welcome to NASCIO Voices, where we talk all things state IT. I'm Amy Glasscock in Lexington, Kentucky.

Alex Whitaker 0:11

And I'm Alex Whitaker in Washington, D.C. Today we're talking about what comes next after generative AI, and Amy is here to tell us all about her new report, Beyond Generation: The Rise of Agentic AI in State Government.

Amy Glasscock 0:23

The report was released last week, and I thought it would be fun to give our listeners an audio overview.

Alex Whitaker 0:27

Absolutely. So let's get into it. Your report is called Beyond Generation. So, what does that phrase mean in the context of AI and why is the conversation moving beyond generative AI right now? I feel like we just got into the AI conversation and we're already moving on.

Amy Glasscock 0:41

I know, that's how it feels, but it has been several years actually, if you can believe it. So for the past three plus years, state governments have been focused a lot on generative AI tools that generate text summaries, write code, they analyze. So, you know, pretty much everyone's using it now. Um, all the states are using it, companies, individuals. And a lot of states now have AI governance policies in place. Most of them have been running pilot projects or proofs of concept with generative AI. And a lot of employees are already using generative AI in their daily work. Um, you know, we ask states about that, and many more than were the year before. So we are always at NASIO thinking about what comes next. So the technology is evolving, where are we going? And we're seeing a shift now. It's not super strong yet, but we're seeing a shift to agentic AI. And in our research, several states already reported early agentic tools in production. So the shift is beginning now.

Alex Whitaker 1:42

Mm-hmm. Yeah. So let's talk a little bit about that actual difference though, because I know many of us are familiar with tools like Chat GPT or Copilot or whatever else you may use. But how do you explain the difference between generative AI and agentic AI in plain terms that someone like me would understand?

Amy Glasscock 1:57

Yeah. So I think a simple way to explain it would be like generative AI is a very smart assistant that writes things for you. Um, also can summarize and analyze, and you know, really smart ones can write code. But agentic AI is an assistant that can actually carry out tasks on your behalf. So, you know, generative AI is generating text and images, code, but you know, it requires human prompts. It has limited memory and context, though that's getting better. Um, but it doesn't execute workflows necessarily. So agentic AI, it can plan and execute multi-step tasks, it can maintain context over time, it integrates with systems, databases, workflows, and it can act with minimal human input. So a simple example would be hey, generative AI, ChatGPT, Gemini Claude, write a response to this citizen inquiry. Agentic AI can read the inquiry, look up relevant records, draft the response, open a case ticket, and then route the issue to the correct agency.

Alex Whitaker 3:00

Wow. So that's that's really fascinating. And I'm sort of seeing like all sorts of I've I've we you know we see things like that in movies, and the fact that it's actually here is pretty crazy for us. Yeah. So you outlined five phases in the evolution towards agentic AI, and where are most states on that journey today?

Amy Glasscock 3:15

Yeah, so I'll just kind of give you a quick overview of these kind of five conceptual phases that we talk about in the report. And and they're not like you must go through these steps. It's just sort of where I I feel like states might read these and kind of see themselves somewhere in this spectrum. So uh phase one is assistive generative AI. So AI is drafting documents, summarizing policies, writing code. We'd be hard-pressed to find a state that's not already using AI in this way today. Phase two is more context-aware generative AI, so chat bots that remember conversation context, AI tools trained on agency knowledge bases. And I'd say most states are already here as well. Even personally, who among us hasn't noticed our own chat GPT or Gemini bots understanding us pretty well at this point? Or maybe you've created a custom GPT or something based on your work or a project that you have going on. I'm not sure.

Alex Whitaker 4:08

Just quickly, Amy has done that for NASCIO, so check it out. I wanna I wanna make sure she gets her due here because she's done that and it's great.

Amy Glasscock 4:14

And personally, too, I used ones based on local building codes to help me with the renovation recently.

Alex Whitaker 4:19

So also a fun process to watch.

Amy Glasscock 4:21

So uh I did learn that I really need to go read the code after and just have it point me in the right direction, though. So uh that's an aside. But so phase three is task level automation. So this is where we're seeing early agentic AI starting to happen. So it's filing help tickets, pre-filling forms, routing documents for approval. And some states are probably seeing themselves here. Phase four, multi-step workflow agents. So it's able to manage processes across systems, and then phase five is the proactive AI agents identifying problems and initiating work automatically. So nobody's really stuck at phase one at this point, and nobody is really gonna be at phase five yet. So I'd say states are somewhere between two and four, and a fair number of states are probably starting to enter and work through phase three task automation level.

Alex Whitaker 5:16

Yeah. So if some of them are getting to that phase, I think we could probably look at, and and you maybe you could suggest what are some practical examples of how agentic AI might be used in state governments in the coming years since we have this base? Are we are we really trying to think, you know, how could we actually use it in state government?

Amy Glasscock 5:31

Yeah, so I think a big way that states are going to be looking at using it is citizen services. So, you know, AI agents helping residents navigate government services. So it can automatically fill forms, retrieve documents, check eligibility, complete multi-step transactions. And in fact, you know, we've seen um in the press that Alaska is exploring agentic capabilities in their My Alaska portal for citizen services. Another way would be with internal government operations. So, you know, automating some of those administrative workflows, um using it for employee onboarding, compliance monitoring, procurement processes, knowledge management updates, and also for operational efficiency. So detecting anomalies or fraud, um important, you know, cybersecurity partner, identifying procurement bottlenecks, monitoring compliance risk. So uh we also have seen that Tennessee is exploring agentic capabilities in their next generation ERP systems. So looking into that. Wow. Also for service management, so AI agents helping manage citizen complaints and service requests. Uh, we have seen that Texas DOT Department of Transportation is piloting AI agents to help assign and resolve citizen complaints faster. So I think a lot more states will be looking at that.

Alex Whitaker 6:52

Okay, wow. So so this isn't just in the theoretical. I mean, we're really looking at some ways that can this can be incorporated and improved services. That's that's really fascinating. But um, you know, for me, one thing that stood out in the report is that Agenic AI, and you just talked about this, um, can actually take actions rather than just generate content, which is, you know, a huge shift. Um, but tell us why that's such a big shift for government technology teams.

Amy Glasscock 7:14

Yeah. So, you know, with generative AI, humans are still performing the actions. Uh with Agenic AI, AI can begin executing parts of the workflow itself. So, you know, with generative AI, it's creating stuff, but humans are still doing the work, and with Igentic, the AI can start doing some of the work. So this matters because, you know, we all know that government work is filled with complex processes, multi-step workflows, and agentic AI can coordinate, you know, across systems, track progress across multiple steps, and then of course escalate issues to humans when necessary, which is an important aspect of all of this. Um, there is a possibility for a huge productivity impact. So some of the research that uh I saw when I was writing this report was you know 30 to 60 percent time savings and significant acceleration of processes. So this is important because you know, we have talked about the workforce shortages that states are facing. You have this, you know, growing retiring workforce and also increasing service demands. So agentic could help employees become a lot more productive. And it's interesting, uh, the policy team was sharing an article here at NASIO a few days ago about how as it becomes easier and smoother and sort of a better, more streamlined experience for citizens when they interact with state government, then actually there might be more demand for services from state government because you're sort of reducing some of that friction. So that's an interesting aspect to think about as well.

Alex Whitaker 8:48

Yeah, absolutely. You know, I I know I, for one, and I I try to work on this, but I tend to be a little bit more pessimistic on some of these things. You know, when I mentioned the movies earlier, I tend to think of the movies where it's you know, agentic AI has has gone terribly wrong.

Amy Glasscock 9:01

Yeah.

Alex Whitaker 9:02

Um and I try not to do that. So and I know you you just mentioned a little bit of the you know, the the pluses, the opportunities, but any other kind of big opportunities for states that you might want to mention about if they do um adopt agentic AI thoughtfully, they can help my uh put some of my anxiety at bay.

Amy Glasscock 9:17

Yeah. So, you know, like I talked about supporting the workforce, which has a shortage, helping citizens, you know, there's they still in many cases are struggling to navigate government services. So just kind of helping guide residents through services from start to finish, clearing out backlogs, automating routine tasks and doing things more efficiently, reducing administrative bottlenecks, you know, just sort of those kind of basic things to boost up the with the workforce in states.

Alex Whitaker 9:47

Yeah, absolutely. But um, so well, those are those are great, and I'm glad there are some some obvious positives, but it it's not in my nature to not look at the risks too. So um, you know, you do talk about risks as well. Um so what are the biggest risks for governance or security concerns that states should also be thinking about as they they start to go down this road?

Amy Glasscock 10:05

Yeah, so I would say this is probably like the the most we've written about like security and risk in an in the AI report at NASIO. Um because, you know, when you start thinking about AI autonomy, agentic AI can access systems, it can move data around, it can trigger actions across platforms. So if it's misused or manipulated, or if it just kind of goes rogue, you know, obviously it can cause large-scale issues quickly. So some of the key risk areas I think that states should consider are security vulnerabilities. So, you know, AI agents uh again can be manipulated to access sensitive data, escalate privileges, and trigger unauthorized actions, which would all be very scary. There are data and privacy risks. So if you have an AI that's interacting with multiple systems, obviously that increases exposure. And then, of course, trust and workforce concerns. So some government workers still remain, you know, cautious or skeptical about AI adoption. So that's something that, you know, has to be considered as well. So we did put in some kind of recommended safeguards in the report. And so some of those are that states should focus on, you know, making sure you have human approval for any high-risk actions, of course, strict limits on AI permissions, and then continuous monitoring and logging. So, you know, you really have an handle on what it's allowed to do and you're keeping a close watch. So the overall approach should be kind of, you know, agenic AI should be treated as powerful, but untrusted by default for states.

Alex Whitaker 11:40

Awesome. Well, those all sound really um reassuring that there are some safeguards that we can put out there too. So I really appreciate you talking about the report. Um, so we will put a link in the show notes, but everyone can also find it in the NASIO Resource Center on our website, which I really encourage everyone to do. Um, because it's a really fascinating report, Amy.

Amy Glasscock 11:58

Thanks. It was fun.

Alex Whitaker 11:59

Um but of yeah, good, good. But of course, um, we have to have a little fun before we end this episode with a lightning round. Are you ready?

Amy Glasscock 12:09

I'm ready.

Alex Whitaker 12:10

Okay, so let's keep it going with AI. Um, and if you could have an AI agent handle one everyday task in your life right now, and that can be work or personal, what would it be?

Amy Glasscock 12:20

Well, I wish I could say doing dishes, but that's physical AI, which we're not really there yet. It's probably probably a relief. Um, so I would say, you know, we are um kind of in the home stretch of this upstairs edition on our home. And so I feel like I am like trying to figure out all these details lately. So if I just had an agent that was like really smart about all the things I care about that could just go find products for me to buy, like, you know, a light fixture or trim for my windows, like that would be that would be really useful, which it kind of is, but like not perfect yet.

Alex Whitaker 12:54

Yeah, no, that makes sense. You can get some weird things showing up at your doorstep.

Amy Glasscock 12:58

Or like, oh, this is a thousand dollars more than I was really hoping to spend.

Alex Whitaker 13:02

Well, maybe it would stick to a budget. That would be a good thing.

Amy Glasscock 13:04

Yeah, I think that would be a key uh risk uh permission.

Alex Whitaker 13:08

Yeah. All right. So spring is finally arriving, and it certainly is here in DC, and I think it is in Lexington too. Uh, but what is one thing you look forward to every year when the weather warms up?

Amy Glasscock 13:19

Um, for sure, all of the beautiful flowers that start blooming and seeing what my landscaping will do as spring and summer arrive, it's always a little bit like better every year. We invested in some landscaping in like December of 2023. And you know, the first summer it was okay, but then the next summer I was like, oh wow, like this is awesome. So I can't wait to see what it looks like this year. And of course you got to get through the pollen season, but you know, I'll take it.

Alex Whitaker 13:45

Yeah. All right. And finally, what is one AI tool or technology that still feels a little bit like magic to you?

Amy Glasscock 13:53

Um for me, it is like how good it has gotten at creating images. And uh, you know, I think we've mentioned this a couple times on the podcast, but I do some interior design work on the side. And so I used to think like, oh, one of these days I'm gonna sit down and figure out how to use Photoshop, you know, but now like I don't think I'll ever have to because I can take a picture of someone's kitchen and just say, show me what this would look like with, you know, this backsplash and this countertop. And it's like so good now. Um, it's really been a huge jump in like the last six months or so.

Alex Whitaker 14:27

No, for sure. And you know, I feel bad for the graphic design industry. I I really do. It's tough. But yeah, I just I also agree. I always wanted to learn Photoshop, but don't think I need to anymore.

Amy Glasscock 14:36

Yeah, yeah. You still get to be the director, you know. So but yeah, huge time savings too.

Alex Whitaker 14:43

Sure. All right. Well, Amy, I think that's it for today, but thank you so much. I know this is really fascinating. And again, everybody should check out that report.

Amy Glasscock 14:50

Great. Thanks, Alex.

Alex Whitaker 14:51

Yeah, thank you.

Amy Glasscock 14:52

Thanks again for listening to NASCIO Voices. NASCIO Voices is a production of the National Association of State CIOs. Learn more at NASCIO.org.

Alex Whitaker 15:01

We'll be back in two weeks with this to be State CIO, Craig Orgeron. Subscribe to the podcast so you don't miss it.

Amy Glasscock 15:06

Talk to you soon.