September 3, 2026
WHO IS THIS EPISODE FOR?
Federal technology leaders, military decision-makers, and government contractors interested in applying AI to cognitive operations, information warfare, and mission-critical decision-making.
WHAT WILL YOU LEARN?
How AI can help the federal government detect misinformation early, accelerate decision-making, operate in disconnected tactical environments, and measure the real-world impact of cognitive operations.
The explosion in sensors and remote devices has led to many discussions in the federal tech community. In fact, there is a podcast called Feds at the Edge. Today, we look at how an AI native startup can help the military make accurate and speedy decisions at the tactical edge.
Our guest is Carrick Longley, the CEO of ZenithFlow. He came to understand the problem of operating in a contested environment while serving in the U.S. Marine Corps. He begins the discussion by examining decision-making in a remote environment. Traditionally, one may use the well-known OODA loop. That is to say: Observe, Orient, Decide, Act.
This concept may have had application in earlier environments. However, today, a warfighter may lose communication in the first few minutes in a contested environment. They could be in a world where disinformation is present, and decisions are extremely difficult.
During the interview, Longley discusses circumstances in which a warfighter may be deluged with information yet lack a way to filter the most important data. Longley expands upon a concept called “cognitive domain.”
From his perspective, data is not the problem. He states that the way we process the information makes sense. That way, we can make better decisions off that data.
For Longley, speed in decision-making is accomplished by moving the human’s decision earlier. He emphasizes the importance of early detection and pre-bunking of adversarial narratives, as well as the need for synthetic audiences to test messaging effectiveness.
Listen to the interview to learn how a network must be able to operate in isolation and still inform military leaders to make decisions that are rapid, accurate, and effective.
Speaker 1 0:00
Hey, John Gilroy here. Today we’re going to answer the question: Can an AI-native startup help the federal government? Hit the music, Manny. Welcome to the Federal Tech Podcast, a podcast that connects you to federal technology leaders. My name is John Gilroy. I will be your moderator in the virtual studio today is Carrick Longley. He is the CEO of Zenith Flow, Z E N I T H F L O W. We’re going to talk about cognitive domain and the federal government and AIs and you better put your thinking hat on because Carrick, you’re going to dive in some heavy stuff, aren’t you?
Speaker 2 0:37
We’re going to get there, John, but we’ll try to power through in the next 20 or so minutes for sure. Thanks for having me on today.
Speaker 1 0:42
I fired up my little computer. I typed in zenithflow.com. I’ve got all these colors. I’ve got this thing called cognitive domain. So what the heck is cognitive? You trying to impress me or something?
Speaker 2 0:53
Well, well, well, no, John. You know, the cognitive domain has been coming up. I think fairly recently in in the lexicon within this space, but it is, in short, it is the contested space where perception, decision, and behavior are all shaped. This is not about content anymore; it’s about the effects in that space. So this is an area that we’ve been working in for for quite a long time, and we’ve been building software, specifically a platform called Cognitive OS to actually help the federal government do operations within the cognitive domain.
Speaker 1 1:25
Now, when I think of a cognitive domain, I did some research this morning, and I thought of my buddy Bob Goorley, and Bob Goorley is famous for these O O D A and popularizing it, and that’s something pilots used years ago. It’s called observe, orient, decide, and act. So, is that your your grandfather’s Chevrolet is that just you know five years ago, or where does the whole Ooda Oodaloop fit in with this cognitive decision making perception world?
Speaker 2 1:49
Yeah, I mean, so I think of of the cognitive domain as as a mission area. So it is terrain. It’s the it’s got a thinking adversary, and it’s got the entire targeting and assessment cycle. So, so Boyd’s Oodaloop, which you know Boyd popularized it, and and I think even though it didn’t get picked up as much in the Air Force, certainly as as a former Marine, we we spent a lot of time talking about the Oodaloop. There is this is no different in terms of the use of of the observe, orient, decide, and act cycle. What really is happening is is the use of artificial intelligence in these tools just makes that loop faster. So we have to compete against an adversary who is trying to compete with us in this domain, where where they are leveraging lots and lots of tools to move through this process faster. And it is incumbent upon us to develop, implement, and use the right tools to try to figure out how do we actually get within that same cycle a little bit faster than the adversary can.
Speaker 1 2:45
Way back in time in 2015, if you can remember that far back, there was a term that was popularized called big data. You know, I love talking about it because didn’t say accurate data, didn’t say actionable data. It’s it’s just big. It’s just big, and I think the federal government has been bombarded with big data, and and what’s good, what’s bad, what do you keep, what do you throw out, all kinds of things. So, so how can you help someone make a decision when they’re getting bombarded with I don’t know what every possible sensors everywhere. Maybe how do you know which what to act on? So how how does the cognitive process work and filter that information. How do you handle that?
Speaker 2 3:22
Yeah, for sure. So you know, I think the big. You’re absolutely right. You know, the the the volume and velocity of data that is now being produced is almost difficult to even fathom how how big it is. And and the problem or the bottleneck has never actually been collection. You know, we’ve never sat around and said, well, we don’t have an update on this, right? Right. Problem now is how do we how do we shrink that distance between observation and some decision quality that’s out there, and this is where I think AI in particular really starts to earn its keep. And there’s there’s a lot of snake oil about artificial intelligence, but I think the part that actually is very very relevant is is helping to do a few things. And and I’ll and I’ll tie this into the the cognitive domain here specifically. It helps with the volume and velocity of data that exists today. Artificial intelligence can help us detect narrative structure at scale. So, in other words, which narratives have momentum, who is carrying them, and how do they connect together? It can help us forecast rather than everything being retrospective reporting. And more importantly, and I think the part that that you’re getting at here is it helps compress that collect to decide cycle from days to minutes. Right, a human being cannot possibly process all of the signals and information that are that are that are coming.
Speaker 1 4:33
Impossible. So this is
Speaker 2 4:35
where we think these types of tools actually help compress that. And the the question that you should be asking is is you know when I think about do these tools or how can these tools help, it’s does the output of whatever process you’re doing end with a recommendation and a confidence level, or does it end with a chart? And if it ends with a chart dashboard, we just move that analytic burden downstream and we. It automation. We call it big data processing. So this is all about how do we reduce this massive volume into a decision that I can do something with as as a decision maker, rather than just well, we need more more data. Data is not the problem. It’s how do we actually process it in a way that makes sense, so that we can make decisions off of that data, not just we’re just producing it for the sake of doing so.
Speaker 1 5:23
So, Carrick, you mentioned your military background. You probably heard of false flag operations. So, a lot of lot of information coming in, not maybe all accurate. So, so how can you help an organization or warfighter, for example, respond to this rapidly changing narrative? I mean, you know, there’s there’s good information coming in. Too much information. Some disinformation. I mean, you have disaster response situation where communications are cut off, coming on, coming off. Not a good situation to make these life and death decisions. I mean, it happens all the time.
Speaker 2 5:58
Yeah, it certainly does. And I think the you’re going to continue to see more and more of the adversarial narratives, you know, in our in our space. And I think there’s a couple of things that that are worth calling out. One, you know, in adversarial narratives, it’s about detection timing, and detection timing beats detection speed. What do what do I mean by that? So the value is in catching a narrative at its seed stage. So there’s all kinds of research that that shows that you know inoculation or prebunking predicted adversarial narratives not only is cheap but it’s fairly effective. If all we’re doing, however, is detecting and responding, and many places talk about countering narratives. Now the problem with countering narratives is once you once it’s gotten to a place where there is real momentum and speed behind the narrative, whether it’s true or not. Once it starts to scale, we’re doing damage control, and in many cases, we’re at a permanent disadvantage. So the the if it bleeds, it leads, and the and the and the lead story is you know if I if if you get the first word out, even if it’s not correct, but you can do it quickly and you can do it at scale, and it starts to really to to to spread within a particular population. Even if we are trying to counter it, we are at a distinct disadvantage. So this is where I think these types of approaches can can help. So you know, to your question of how can an AI startup help the federal government? Well, one, it’s about how do we process the right data at speed, and more importantly, at the earliest part of when these narratives start to to be disseminated, and can we catch it at that seed stage? And if we can, then we start to be able to come up with techniques and approaches to not only counter but to pre-bunk what we think the adversary is going to do. Then it’s about how do we take this and how do we start to identify who are the local entrusted intermediaries to be able to push out our messaging. Official broadcasting in almost every case does not work, so we have to look at how can we not only detect early, but we can identify who are the right conduits, who are the right intermediaries that we can then push our messaging out, so that we get, so that we become competitive in this space. That’s really, I think, that’s really the gist of of understanding, you know, everything from mis and disinformation, et cetera. It’s about early detection, understanding the right audiences, and understanding the right conduits, and then being able to get our messaging, our messaging, our narratives out, so that we can get the speed and scale that we need to be competitive and start getting inside of their Udaloo, rather than us always being in a reactive mode. And that’s frankly where a lot of our a lot of our challenges are today is we are very very reactive. We are not proactive in in many regards.
Speaker 1 8:47
Talk about speed here, timing and speed. I used to work for a guy who said that you know when you race a car, if you go 167 miles an hour, you’ll lose the race. If you go 168, you turn over. Right, right. You know what’s the right speed? Well, depends. I think this is the situation in now. We’re in a situation where someone has to make a life and death decision, and it has to be quick. But it’s got to be responsible, and there must be oversight. But it’s got. But so so, character over, or do you you know, how fast do you go? I mean, that’s that’s what you have to decide.
Speaker 2 9:25
Yeah, you know, there’s there’s there’s no doubt that that artificial intelligence is helping us move at a much much faster clip than we could have done manually. I mean, that’s that’s that’s true in nearly in nearly every regard. But I think the design principle we’ve got to center around is that the the AI does sensing, helps you structure, and it helps you generate options. But the human is the one that decides and authorizes anything that quote unquote leaves the building. The real unlock is is how we start to pre-delegate decision rights, and and this gets to the you know if we can tier effects by risk. If we have playbooks that we can pre-approve or get approved in advance, it’s all about that operator that is executing inside of a commander’s intent instead of asking permission in the moment. So that’s where I think we really get we really get speed. And you know, and said another way, you you don’t buy speed by removing a human being; you buy speed by moving the human’s decision earlier, so when I talked about a lot of our, you know, a lot of our operations are very reactive in nature. It’s because we have not built in the process where we’ve either pre-approved or where we have adapted our processes to this new reality of how much faster things can move. But there are some mechanics that that that we can put in place. You know, human on the loop for monitoring, human in the loop for release, having pre-approved message libraries. Some things we do today. Some of these are are are in practice, but but having being able to have it’s much easier to have oversight to defend when it’s logged rather than than being reconstructed after the fact. So, you know, to your analogy of of the car faster for going faster for going faster’s sake is is not really the the desired output. It’s can we make the human more competitive, but do so where they are still very much in control of of not only what is being produced, but but but what what’s being released as well.
Speaker 1 11:19
Okay, now we go back in time. Put on your main uniform. You’re in the mud somewhere at night, a disconnected environment. Maybe if it’s connected, slow bandwidth, and is what they call a contested environment, which is a nice way. Oh, I believe I see an adversary over there. I’m some people talk like that, but it’s a contested environment, and people know what that is. So, so can AI, can that help our fighter at the tactical edge? I mean, that’s a question I would ask. I was sitting in the mud somewhere with a rifle, so how can you help them?
Speaker 2 11:48
Yeah, I mean, in fact, that’s so. So all of the work that we do actually starts with the tactical edge first, and this came from from a lived experience on a on a deployment on the USS Iwo Jima off the coast of Yemen, and we supported the embassy evacuation that happened. This was in the 20 early 2015 time frame, and we were not originally going there, and that was not originally part of our plan. And I was tasked with coming up with all series of of different effects and to support the the maneuver force that was there. The problem was is we’re on a navy ship, which has very very limited bandwidth, and we have very limited connectivity back to our quote unquote reach back, and so it became really problematic to understand the environment that we were trying to operate in. Fast forward to 2023 when I created this company, and I was particularly fascinated with the application of artificial intelligence on edge at the tactical edge, and and so I started building with the idea that every single operation we do, every model that we run, every process has to be able to run completely disconnected, because as soon as the balloon goes up in a in a large scale combat operation. The first thing to be targeted are our communications, our communication channels. So if we build all of our infrastructure around really, really large cloud compute models around these massive data centers that require me to be able to be connected, the moment that that radio link goes down or that satellite link goes down, I’m in a real problem. So there are a number of things that you know that that that we we’ve designed around that, but not not just us. There are others that are building for this tactical edge. But being able to be able to run all of your operations in a standalone environment is sort of table stakes for us. The other knock-on effect or benefit when by doing that is data sovereignty. So we do a lot of work not only with the U.S. but with our Five Eyes partners. And one of the big questions they always ask is, can this run within my own environment? Can we pick up this entire stack or this entire box of tools, and can we deploy this on our own infrastructure? Because we need to know where the models are running, where the data is being stored, how we’re actually processing, and that that is one of the side benefits of this tactical edge approach that we that we took is that the entire footprint can be in this local first architecture, which then gives us flexibility. Yes, we have a cloud connected version. Yes, we can do this on prem, but more importantly, we can run all the way down to that tactical laptop, so that that marine sitting in the fighting hole who is tasked with trying to come up with effects or trying to come up with messaging that they’re doing local dissemination has some ability to actually use that right there on device, independent of any kind of connectivity whatsoever. A capability that only works from the network up isn’t a tactical capability. It’s it’s a garrison capability.
Speaker 1 14:43
Kirk, I spent some time in the classroom, and I like using fancy words. I even press people with fancy words. You know, my wife’s a Latin teacher. She’s all kinds of fancy words. And when I first started research, and when I start writing cognitive domain, that seems like kind of a fuzzy kind of well, you know, make sure it looks good. So so, what happens when I mean you’re gonna you’re gonna run into some military guys? They’re gonna say, “Hey, give me the numbers. I yeah, I want to. How do I measure this? Oh, Ree, that sounds good. Well, how do you how do I know that? And so how? So so what do you say when a general points you and goes, okay, big guy, let’s put the you know pencil the paper here. I mean, how do you measure the outcomes of an improved cognitive campaign?
Speaker 2 15:24
Yeah, great question. I think this is the the biggest challenge in this space, and it’s you know frankly not a solved problem. But you know where we have been historically is information operations, which is the the typical label, at least in the U.S. cognitive warfare, and in in the NATO lexicon, there are a whole series of what we actually call this, but for for decades now, I think we’ve we’ve led with what I call vanity metrics: reach, impressions, engagement numbers. Those are all measures of performance. They they tell you that the machine ran, not that anything actually changed. What we’re talking about is how do we actually get to the effects? Can I measure the effects, and and and more importantly, what are those behavioral indicators that that exist? So so can I measure that the adversary posture or messaging spend changed, that their recruitment and defection rates dropped, that we that we are seeing an increase in tips line volume, which is directly tied to recruitment of of you know individuals within a military organization, evacuation compliance. If I’ve got messaging that’s about evacuating a particular area, up uptake in rates in health context, narrative displacement, so on and so forth. So it’s really about we need to be very explicit about one what we are measuring, and two, that we write objectives that can be measured. You know, part of the problem has been we’ll get these very, you know, these these very high level things. I want to counter the adversary narrative. Well, I don’t even know what that means, but if I can, if I can take that and decompose it into, I want to. We want to see a reduction of X percent of people that support the regime in this area. Well, that’s something that’s measurable that I can actually start to try to build out. What are all the ways and indicators that might allow me to measure over time where where that that behavior change is happening? There is there’s also a real problem that we have, and this has more to do with our our episodic nature versus being in a continuous state of operations. You know, our adversary does not view cognitive domain operations as a thing they spin up or spin down, and as result of this being steady state, they have a baseline from which they can measure change. We often will show up to, let’s say, an exercise. We haven’t thought about any of our messaging strategy or posture strategy till the absolute last minute. And more importantly, we have no baseline in the environment from which we can actually measure change. So, if you said, “I want to increase the awareness of U.S. support to this population in this particular exercise. If I don’t have a baseline, how in the world am I able to say that we had a change, positive, negative, neutral, or anything? And that’s that’s a lot of the the problem is that we’ve never done, or we don’t consistently. I shouldn’t say we never. We do not consistently do baseline baseline the information environment from which we can actually start to see that change over time.
Speaker 1 18:20
Carrick, I’ll state the obvious here: AI seems to be impacting all kinds of different levels in the federal government. So let’s take a look at your specialty, this cognitive domain stuff. So trying to look at three or four years down the road, how do you think AI will transform cognitive operations? And will some things become standard three or four that we’re just maybe investigating today, what do you think is going to be real popularized after AI is applied to this problem?
Speaker 2 18:46
Yeah, there’s a couple of areas that I think are going to that that are really going to change, and where we’ve spent a considerable amount of our time. You know, for many many years we had no real ability to measure what the effect of our campaigns were going to be outside of small focus groups, and in many cases we didn’t even have those, and we would just do what I would call what’s traditionally called spaghetti marketing. We’re just going to throw something in the wall and see what sticks. The where this space is now, though, is we’re now starting to be able to generate synthetic audiences that are backed by real science, by real models, and really start to test what we think the effect of our operation, our messaging, our whatever the case may be, our actions are going to be, and see how those synthetic audiences will react. That gives us something we’ve never had, which is the ability to rehearse our actions in the cognitive domain before we actually release, you know the the joke is often it’s easier to drop a bomb than it is to send a message, and and that has that is largely true, and that is also largely true because we know we have a joint munition effects manual, a JMAM. We know exactly what a kinetic action will do. In terms of physical destruction, we have very little. We have very little ability to say here’s what the effect will be in a message or some other effect. We’re trying to change that, and I see that’s where the artificial intelligence is really, really helping. The other is that it’s dramatically improving our the quality of our that assessment loop we just talked about the measurement trap or the measurement challenge, and and I do see you know one of the ways to be able to instrument all of this measurement, the the the result is you get a lot of data, which brings back to the beginning of our conversation, the big data problem. Models, artificial intelligence can help you make sense of that data, and help you start to make actions. So I see it continuously improving in this loop. So within cognitive domain operations, yes, you can generate lots and lots of volume, but that’s not really the thing that’s that’s hard. It’s can I generate the right message or the right effect? Can I test that message and that effect? Can I then quickly iterate on what what right looks like when I actually move it operationally and and do that effect or take that that message or whatever we’re doing externally. Can I then collect all of the indicators that I need to reinforce and measure whether or not this thing is actually working or not, and be able to do that very very quickly? So that that’s where I see all this going. Is that is all of those pieces moving in concert together to to make that oodle loop much much faster and much much more accurate in the space?
Speaker 1 21:30
Good, good. So, Carrick, if this conversation maybe provokes some the audience to learn more, where can they find out more about you? Maybe meet you face to face.
Speaker 2 21:39
Yeah, absolutely. So we’ll we’ll be at Apex 2026. That’s put on by the Information Professionals Association, September 8th and 9th. That’s in the Kiz Harrisoff Center in Reston. You know, if you’re a team that’s working narrative assessment or or in cognitive operations, if you book 30 minutes with us or come chat with us, we’ll show you what we’re doing with Cognitive OS and how that platform can help agencies and organizations be competitive within the cognitive domain. Otherwise, you can visit us on the web. That’s zenithflow.com, z e n i t h f l o w.com, or you can find me on LinkedIn, Carrick Longley. It’s not hard to find. I’ve got a fairly unique first name, so so you won’t find too many of us on on LinkedIn. But that’s that’s how you can find us.
Speaker 1 22:19
Yeah, and what I’ll do, Carrick, is I’ll put a link in the show notes to to that event. They could register and go down because at the restaurants, it’s right at Carisoft. It’s so easy to find. Easy look. You can come and go. You can go in there, meet Carrick, and go back to work. I mean, it’s just a it’s a wonderful meeting for soil. I love it there. Absolutely,
Speaker 3 22:36
absolutely.
Speaker 1 22:37
Times great. Well, unfortunately, Carrick, we are running out of time here. You have been listening to the Federal Tech Podcast with John Gilroy. Like to thank my guest, Carrick Longley, CEO of Zenith Flow.
Speaker 2 22:49
Thanks, John. Very much appreciate you having on today.
Transcribed by https://otter.ai