Dustin Hilgaertner

I am a software engineer, but I probably do a lot more than that. I've started companies before, I've been a CTO,

so I have a lot of experience on the software development lifecycle management end-to-end. But yeah, I pretty much would call myself a software engineer.

I work for a company, Radius Method, that services DoD clients.

So we mostly work in the DoD space. We do software factory work, so we help them build factories that output software assets that go into the hands of different war fighters.

We also do AI on the application layer, so we help the military build secure AI systems that can go through and automate some of the processes that they have internally.

We also are deep into security, so we're kind of forging ahead on some of the AI security with an AI gateway that we produced, as well as a zero trust networking, post quantum encrypted tunnel to get into networks,

which we sell to the DoD, as well as corporate clients in the price sector.

I actually did a podcast on this where I called OpenClaw, which I think is a great inspirational project of what you could do. I put it in the same category as like the Ralph Wiggum loop, where it's just, if you dig through the code base, it really kind of just added some capabilities that weren't there before, and then took a whole bunch of things and put them together that were there before. And so it's a great sort of inspirational project, but it's a security nightmare. I mean, it just leaves a port open on your machine. Anybody who finds that port,

it's worse than just an open port, because behind that port is an AI agent that's gonna answer the hacker's questions.

So the hacker doesn't even have to look through your folder directory to find what's going on. It could just go to the AI and say, "Hey, where are the passwords?" And it'll be like, "Oh yeah, sure, they're over here."

Exactly.

The Ralph Wiggum loop is just a while loop around a Claude code or other AI agent call at the CLI level. So it essentially is this idea of like,

"Well, what if I want it?" AI has this problem where you'll have it make a plan. That plan is like maybe 10 phases. And then you'll say, "Go implement the plan." And instead of going through and implementing the entire thing, it'll implement maybe like the first phase or the second phase and come back and say, "Look, what great job I did. What should I do next?" And you often have to say, "Just keep on going." Right?

So the Ralph Wiggum loop constructs a loop that basically stops it from stopping before the work is done. And so you construct your prompt in such a way where you can just repeat it over and over and over again. So instead of saying, "Implement this whole plan," you would say, "Implement this whole plan, but check things off the list when they're finished." And so you can just give it that prompt over and over and over again, and it will continue to make progress by checking things off the list and understanding where it left off.

The person who first put it on the web and kind of made it go viral, they called it a Ralph Wiggum loop

because I guess it's kind of similar to the character in "Simpsons." I think the character repeats himself a lot.

Well, I guess a lot of people are talking about it a little bit, which is AI-assisted development. A lot of people consider it vibe coding. So there's like this split in the software engineering market that's happening right now, where if you post something about AI-assisted development, you'll get a lot of people who have software engineers in their title that will come out and tell you that this is just complete garbage, it's producing bad code, et cetera.

So what I see is that there's a lot of people with software engineer in their title, but they're really identified as a coder or a programmer, and they see it as an attack on their identity, like what would happen if that were to disappear. But software engineering is a lot more than just coding. In fact, coding is probably the easiest part of it. The hardest part is the architecture, the vision. You find that when the coding is basically a non-issue and you can get something to the end with AI-assisted coding. Now, as a software engineer, you have to be able to put in the right plans and the architecture and the vision, and then on the tail end, you have to be able to review all the code and understand that it made a mistake or there's a security hole here, and you can patch those using AI as well. Now, if you're not a software engineer, you're not gonna be able to use AI-assisted coding at this point in time without producing some form of slop or like a heavily security issue.

With that being said, another thing I'm looking forward to is the idea of these AI gateways at organizations that are trying to secure their internal usage, get rid of what we call shadow AI, which is this idea that like, there's a lot of organizations out there who don't have a policy on AI. They don't have internal tools for AI, but all their employees are using their personal accounts because the capability of these things is too great. So that means all this company data is leaking out into personal accounts ready to be hacked one day,

as well as the organization's not getting all the productivity they could get. So their employees are copying and pasting in secret from their chat GPT accounts. Whereas if you truly integrate these things,

the tools can actually take actions on your behalf, as you've seen with OpenClaw and other things that do things on people's behalf. So I'm very interested in how that,

A, starts to get implemented within organizations and also flows through an AI gateway internally that secures and puts guardrails in place so things don't exit the network. And then at the layer of the AI gateway, if you think about it, if all your employees are using AI and they will be on an increasingly, increasingly basis, if you look at companies like Meta, you will essentially be put on a performance improvement plan if you're not using AI enough, they're tracking, they want you to be using it for everything. And I think that's gonna flow through to all the organizations at some point. So if you think about it, all these employees, their work in real time will be flowing through this AI gateway. That's a place where you can tap that information and get just basically,

it's like the artery of what's going on in your organization. So you can imagine that things that would traditionally be reported up like statuses and blockers and things like that can just automatically be generated for each level of the org chart going up.

Absolutely, I think it's great. I think that there's probably,

certain things you wouldn't wanna do, like you don't want to micromanage all your employees.

By looking over their shoulder and getting into their business, I think that there's potential for that stuff to be misused, as there always has been with all these tools that monitor usage, et cetera. Like you wouldn't want your manager coming in and saying, "Hey, you don't have enough commits today," which is already possible using Git and things like that. So I think there's uses for it that are important though,

for accelerating productivity, making sure engineers and other people within organizations that aren't even software companies can understand what other employees are doing, it can connect them with employees that are doing similar things. It can have an organizational memory that allows onboarding new employees to be much more effective. Imagine you had that senior person that you could shoulder tap anytime you wanted, and you don't have to feel guilty about bothering them again. And they answer all your questions very specifically, no matter how many ways you ask it.

It's funny you asked that because I said I consider myself a software engineer, but I hesitated in the beginning because I actually consider myself a problem solver. Software engineering just happens to be a highly effective way of solving problems that somehow clicked with me when I was younger. When people say they're afraid of AI taking their job, first off, I don't think that's happening in the short term. I think it's gonna take some time. You still need engineers at the beginning and the end, like I mentioned. But I think I would just find some other place that I could solve problems, and I'm excited about AI because it makes solving problems so much easier, and you can kind of elevate yourself to a level where instead of being in the weeds trying to solve some problem, you can sort of horizontally scale, where you can solve multiple problems at the same time, but don't have to get into the weeds of every individual problem. You can kind of make yourself as a single person, not have to be the individual contributor. You can be at the CTO level by yourself, which is nice.

I think in retrospect, the idea of people obsessing over the style or the efficacy of a single line of code, like, oh, I write a single line of code better than anybody else, is gonna seem like being the best horse rider. You know, it's cool, but it's not efficient anymore. You know, it's not the highest value thing you can learn.

Now, there's still gonna be people, I'm sure there's gonna be like, coteur coding, where people are doing it by hand still, even in the future, but I think it's gonna be, I think coding is kind of dead. And I think the only reason why it's not completely dead is because there's a lot of people who either haven't figured it out yet or are in denial.

Well, it's an interesting question because

mostly when we hire, we hire from like people we've worked with before. We have like a deep Rolodex of people that we've worked with in the past and other projects, so we go to that first. But if we ever got to the point where we didn't have anyone in our Rolodex that fit a profile, we would certainly be looking for people that understand AI and are using AI on a daily basis as opposed to someone who isn't.

With that being said, we also look for people who have DOD experience or security clearances, things like that.