Interview transcript

Terry Gerton The Information Technology Industry Council, or ITI, has a new paper out, Understanding Agentic AI. Help us with the basics there. What is the difference between agentic AI and some of the most common AI tools? We’re all familiar with ChatGPT and its partners.

Jason Oxman Well, agentic AI is the next generation of AI, if you will. Agentic refers to the use of an agent. And an agent is someone who does something on your behalf. So when we talk about agentic, AI, essentially what we’re saying is giving the AI some autonomy to help you as the human user of AI to accomplish a task. So for example, you can set your agentic AI to perform a task for you that you otherwise might have to do yourself. Telling the agentic AI, for example, I want to go travel somewhere, the agentic AI can go search out different travel options, book some tickets for you. You can set the level of autonomy that it has. But the general idea of agentic AI is it is out there using the AI tools to perform a task on your behalf.

Terry Gerton And your paper argues that AI can boost productivity, streamline processes and enhance security. So beyond me using it to put my Christmas shopping list together or book my airline tickets, tell us more about how it can be used in a more particular business sense, I guess.

Jason Oxman I think that’s a great way to think about it. The business sense uses of AI are ones that we think of as enhancing productivity for the human beings that are involved in a particular task. The best example I like, and this applies to both the private sector and certainly to government, is in customer service use cases. If you’re, for example, filing a claim with the VA because of an insurance event or because of the passing of a veteran. You’re going to need to call somebody to get some help. And a lot of that help that you’re going to needs is in very basic tasks. The VA has a lot information about service records, but it doesn’t necessarily have all the information you need to process the claim. Agenetic AI can help by telling the agent, here’s what I’m doing. Here’s what I’m trying to accomplish on behalf of someone. Can you go out and find this information for me? I need to compile together service records, insurance claims. I need find a death certificate. I need to find the copy of the policy. Can you go out and find all those things for me? A human being would have a lot of work to do to do that. AI can help. And then the human being can focus on the more complex and more important tasks, not those kind of rudimentary tasks that AI can help with. So that’s a good example of a business use case for AI.

Terry Gerton So that sounds like it’s a bit more extensive and involved than a chat bot that we might all sort of be familiar with. And you gave there a VA claim example. Talk us through some more cases where government, maybe the federal government or state government should be thinking about agentic AI and what it’s gonna take to put it out there.

Jason Oxman Yeah, and you’re right to highlight that this is a little different than the usual use cases that we think of with chat GPT or Gemini or Claude or any of the other GPT-oriented AIs that we use for various tasks. What the government can do with AI is very different and I think very exciting and I think about this in the public sector arena in particular. Think about how public sector procurement works. The United States government is the single largest purchaser of information technology in the world. And those technology purchases need to happen across agencies in many cases for different use cases. And those agencies need to talk to one another and coordinate what they’re doing. So that agencies can buy technology that interacts effectively across agencies. AI can help with that, agentic AI in particular, by looking at procurement cases across different agencies in a way that a human being wouldn’t have access to because human beings work for one agency at a time and can help with those procurement coordination efforts. We also see the government using AI across use cases where consumers need to access information. Tax information is a great example. You usually have to call a human being and sometimes wait a very long time to get access to that human being to get information on prior tax filings or your current tax filings or to ask questions. Agentic AI can help with that because you can assign it a task and let it know what you’re trying to do. It can also help with very simple things like setting up appointments. If you need to talk to somebody in government, set up an appointment to do it, whether it’s a healthcare appointment at the VA hospital, or whether it is an appointment with somebody at the IRS to talk about your tax situation. A human being shouldn’t have to spend time with that. They should spend time on actually helping with what the customer needs. The agentic AI can do the things that are pretty basic, like setting up those appointments. These are all the kind of use cases that we see. They are really about improving efficiency of employees.

Terry Gerton I’m speaking with Jason Oxman. He’s president and CEO of ITI. Jason, you gave a couple of different examples there, some that are customer-facing or citizen-facing, some that back office, internal work. But when the government takes on agentic AI, does it face any particular risks or challenges compared to, say, private sector?

Jason Oxman I think the government’s primary challenge is with the customer data that governments have. Governments have a lot of information about us, and a lot of that information is sensitive information, and we want to make sure that information is protected. So I think that’s a challenge, certainly in the private sector, but it’s highlighted particularly in government because government just has a lot more information about us. And that’s not to say that that information isn’t useful and shouldn’t be deployed on our behalf, but we want make sure when agentic AI is accessing information that it’s protecting the personal information that we all have. You’re a former government employee, I’m a former governor employee. They have a lot of information about us. We wanna make sure it’s protected, that’s one thing. A second thing is cybersecurity. There are a lot bad actors out there that really wanna access information. And in some cases, AI can help the criminals in that use case. We’ve seen recently, for example, a lot fraud that retailers are facing, AI being used to trick customer service representatives into providing refunds to AI, not to actual human beings. So cybersecurity is the second big area. And the third area is the usual challenge of government, which is making sure that government agencies talk to each other. IT modernization is a big theme that we talk about at ITI, and it’s because government systems are old, they’re antiquated, they don’t do a good job of interacting with one another. So we really wanna make sure that the government updates systems to get the best value out of IT investment.

Terry Gerton One of the other things that your paper brings out is the need to train the workforce for this revolution in operations. What are some of the most critical recommendations that you have for government workforce training?

Jason Oxman I’m really glad you raised that because there is this tendency to think of AI and agentic AI in particular as posing a threat to people’s jobs. I think it actually enhances and increases opportunity from people’s jobs. But as you noted, the only way that works is if people are trained in AI. So the way I like to frame it is, AI is not going to take anybody’s job. It’s only going to take a job of somebody that doesn’t know how to use AI from somebody else that does know how to use AI. And that’s what we need to make sure we focus on, training the workforce on how to use it. So in the same way that government workers have, over the last decades and even centuries, had to adjust to new technologies, learning how to the internet, learning how use email, you go farther back, learning how to use typewriters and phones. And all of those technologies have improved the efficiency and the effectiveness of the government workforce, but we need do training. So we need make sure that the tools are available to government workers so that they know how to use them they can make use of them And they can improve the way in which they provide value to their employers the American people by using those tools.

Terry Gerton AI policy has been a bit of a flash point lately as the federal government and state governments debate who should be in charge of it and how centralized we should make it. Your paper recommends developing a national AI strategy and updating government IT infrastructure to prepare for agentic AI. What would you put at the top of the list for policymakers and legislators as they move into this agentic time frame?

Jason Oxman Top of the list is absolutely a national strategy. And the reason I say that is technologies do not recognize state borders. We wanna make sure that there aren’t 50 different regimes governing the adoption of AI. And the threat of that is that technology will not work as effectively in some states as it will in others if different regulatory regimes are adopted. So the reason we really emphasize the importance of a national AI strategy is because we want one strategy, everyone knows what the rules are, everyone knows what protections are in place, and that is of primary importance to ensure the success of AI. And then within that national strategy, protection of data, a national privacy law, which is something we also don’t have in the United States, is really important to ensure that everyone is protected in the way that they want their data to be safeguarded. We need cybersecurity measures to make sure that AI is protected from foreign intrusion and from criminal intrusion. All of this really needs to happen at the federal level. And as you noted, it hasn’t happened at the federal level yet, so we’re really urging Congress to make this a priority, the Trump administration to make this a priority, so we get that national roadmap in place, and then federal agencies and operations can adopt AI knowing what the rules of the road are, and deploy them on behalf of citizens knowing that they’re protected.

Terry Gerton The president has issued an AI policy or strategy. What needs to be added to that to complete the picture that you’re describing?

Jason Oxman Yeah, the president has adopted that national AI strategy and we’re strongly supportive of it. It’s a great looking document. It’s great strategy. It has a lot of things to do and accomplish and a lot of different federal agencies are looking at different pieces of implementation. You know, the things that we think are really important to focus on certainly is having Congress adopt a national AI law that replaces the possibility of 50 different state laws. But also, there’s a lot of implementation work to be done within government. NIST within the Department of Commerce, for example, is working very hard on adopting AI standards so that there are voluntary consensus driven industry standards in place for the adoption of AI so, again, we know what the rules of the road are. That’s really important. We’re also seeing a lot of work being done on driving energy policy that will power data centers that will make AI even more productive. That’s in the AI strategy that the president adopted that’s really important. And then the other thing that we think is really, really important is to make sure that the U.S. Is globally competitive and is making the technology exporting it to the world. That’s a part of the strategy that we think needs to move forward as well.

 

The post From gift lists to government systems, agentic AI is changing how we plan and prepare first appeared on Federal News Network.