Terry Garton The Aerospace Industries Association has released a new white paper about agent AI, saying it is this solution that will speed up decision-making across the defense industrial base. When you think about all the problems that the Department of Defense has with procurement and industrial infrastructure and understanding the supply chain, how confident are you that AI is going to solve the fundamental problem, and that it’s not just another organizational or cultural or just traditional process?
Tim White I think we’re highlighting one of the most important features of what we’re actually doing here. It’s that there’s a complex system that we’re all dealing with. And what I believe, and what we’re seeing now, is the fact that AI is an enabler. I’m an assistant. But it doesn’t solve everything that already exists. We need to address our culture, our processes, and our workforce. All of them are with us now. But I can predict that AI will bring significant changes to the way we address each of these problems. So while it is neither the final nor the only solution, it is certainly a powerful enabler.
Terry Garton Let me talk about real world testing that is being done in public as we speak. The Pentagon is openly clashing with Anthropic over the extent to which military use of AI models should go. The Department of Defense is promoting access to Claude for all lawful purposes. Anthropic takes a hard line against the use of fully autonomous weapons and mass surveillance. I don’t want you to comment specifically on this disagreement, but given the fact that this is playing out in public, how do you think it will shape the path of agent AI in defense?
Tim White In general, I think all companies that do business in the public sphere, and most certainly companies that work with the federal government or the Department of Defense, need to be aware of what those uses are and need to be clear about that. I think a lot of people working in this space now are very used to using it in the right way and are acutely aware of the importance of governance in AI. And in fact, that governance is on both the procuring and delivering organizations’ sides. It’s, are we all on the same page? And that’s one of the areas that we’re really focusing on in the whitepaper is making sure we’re on the right page. Are we on the same page about this? Do we agree on how this is used and what the guardrails are?
Terry Garton Governance is easy to say, but it can be difficult to implement. What would the appropriate governance structure look like here?
Tim White In fact, about a year and a half ago, we published a paper on governance in aerospace, AI governance. It then explains all the different elements needed to achieve effective governance. Because keep in mind that what you’re trying to do is make sure that whatever tool it is, you can use it properly and safely. So you’ll want to consider including things like use case libraries and policies. Over time, you’ll incorporate the ability to revise the documentation and modify expectations as new features or new challenges arise. All of this has real implications, but how do you manage this? And remember that governance is truly a living process and a living capability. You know, agents are new, I would argue. It has changed the way people view governance. Just a year ago, we were looking at large-scale language models. And now we have an agent that actually performs actions on these outputs from our large language model. It changes the equation and therefore the governance.
Terry Garton As we mentioned earlier, agent AI is a system that actually performs actions, often without human guidance. What safeguards are needed to ensure that governance structures are in place to prevent AI from running out of control?
Tim White First of all, you need to make sure you have a process in place, but above all think about policy. People want to know what the guardrails are as they’re introducing all kinds of AI and new capabilities. And it’s incumbent on leaders and experts to come together and say, “Let’s understand what the risks are, let’s understand what the possibilities are here, and let’s provide our community with answers about how to use it appropriately.” You need to have monitoring processes in place, you need to get training, and you understand that the workforce is going to be a big part of deploying AI, and most certainly deploying agent AI. You need to make sure that all of these things to consider when acquiring new tools are in place during deployment. When you start discovering things, things don’t necessarily come to mind. It has to be intentional. And I want to drive home the expectations for leadership engagement. This is the core of a leader’s job: to unite, empower, and provide employees with new tools, but also guidelines and boundaries.
Terry Garton I’m talking to Tim White. He is vice president of engineering and technology for the Aerospace Industries Association. Tim, let’s go a little further because we’re talking about governance. Talk about guardrails. But if agent AI is indeed integrated into supply chain, contracting, and production decisions, where does the responsibility lie if the AI makes the wrong decision?
Tim White My sense is that the current and future accountability lies with the leaders because they are the ones actually sourcing the models. They provide the model, they own the policies, and there is an expectation that employees are using these things appropriately as well, not going outside the guardrails, and raising potential concerns. There’s definitely a partnership there. However, when talking about agent AI, keep in mind that there are many different ways in which it can operate. One way is to involve humans. That’s where agent AI comes in, taking your prompts, providing suggested actions, and asking, “Do you want to go and do this?” So it’s a very clear responsibility for the user to have to say, “Yes, in fact, I’m putting my signature, my own signature, on this action.” Another way it works is to just tell it to go. There are risks. And if you have good governance, you’ll consider those risks and say, “Do we really want to do this?” And depending on your use case, I would argue that for non-critical, non-guaranteed processes, that’s probably fine. But in many cases, and I’m adamant about this, when it comes to aviation, aerospace, defense, we provide warranties, we provide warranties. So we almost always want to involve a human to provide the accountability that you’re looking for.
Terry Garton Now let’s talk about those humans. I mentioned the workforce and the role they play in this process. But the Department of Defense and its contractors always struggle with training. The Department of Defense’s systems themselves also suffer from problems with data hygiene and uneven digital infrastructure. What is the practical path to implementing something so advanced into an organization that is not yet ready?
Tim White I would answer this in two ways. And I think you’re right that there are always ongoing challenges in training. But I would flip that around and say that one of the big benefits of AI is its pattern recognition capabilities. And that leads to the opportunity to use it as a tool to help people when they need help. So think about someone who has just joined a company and is working on a new process. You probably got the training three months ago and now you have to go through the process and you don’t remember everything. The AI recognizes that this is effectively a new user and allows me to participate in the process, provide feedback, and provide immediate and appropriate help. If you think about the traditional way of thinking about agents, I think the idea is absolutely to have the computer do something. There’s a risk there, but I would argue that with these challenges that we have, there’s actually a huge opportunity in AI to solve those challenges, to address the workforce training issues, to address the workforce competency issues, and to be able to use the workforce that we have to accelerate and deliver higher quality solutions.
Terry Garton So you’re making a strong, positive case for the deployment of agent AI, which may be out of our control. It will be here whether we necessarily want it or not. But what does a responsible deployment strategy look like over the next three to five years? How do you see agent AI actually being integrated into, say, Department of Defense workflows?
Tim White I would argue that agent AI is a tool in a toolset. And that means you have to start with the end in mind. What is the end result we are looking for here? Faster procurement? Is it higher quality? All of these can be addressed in different ways, right? Six Sigma, technology, agent AI, they’re all options. So I would never say that agent AI is the beginning of an improvement process. Rather, let’s think about what challenges we face. What tools are there that can address those challenges? Agent AI is probably one of the AIs that can help you deliver faster, more accurate, and higher quality, for example. If it’s appropriate, we’ll move on to the next stage of design and say, what are the policies that we need to implement? How can we properly use agent AI to address this problem that we have? And then we’ll move to training and then eventually to steady state and making sure that we’re doing what we promised, monitoring and making sure that we’re doing what we promised. And it actually plays a key role in the entire improvement lifecycle.
Copyright © 2026 Federal News Network. Unauthorized reproduction is prohibited. This website is not directed to users within the European Economic Area.
