Techstrong TV February 20, 2026

Hey, everyone. Welcome back here to Text Strong tv. I'm really happy to introduce you to this next fellow.

He is the co-founder and CMO and a guest here on Techron with us. His name's Kevin Green. Kevin is with a company called Hap and their website, so, you know, off the bat is Ask aix, A-S-K-H-X ai.

So you can check that out. Um, but first let's talk to Kevin. Kevin, welcome to Text Drunk tv.

It's great to have you here. Thanks, Alan. Glad to be here.

So, I gotta be honest with you, I interview a lot of founders, a lot of co-founders. I don't often interview co-founder, CMO, and I don't, and I'm not, I'm not ragging on CMOs by any stretch, but when you look at the normal, you get your C-E-O-C-T-O pairing, right? So you get kind too techie guys sometimes.

And you know, the, the VCs always say, Hey, you gotta go find a sales marketing person, right? Mm-hmm. Um, or sometimes you'll get A-C-E-O-C-R-O, right?

So, and they're a salesperson. Sometimes the CEO's actually the sales or marketing person. So, but it's not often co-founder CMO.

How did, how did this come about? Tell us about your journey. Yeah, so, um, you know, the HAP journey started several years ago, and I've been part of the company since its founding.

So the original concept for the product and kind of what we're trying to do really started with me and a couple of other people. And the story behind why we're doing this is equally as important to the quality of the product itself. So, you know, there's so much happening in the space right now.

There's so much happening, there's so much news, and the hype cycle is so significant that controlling that narrative and making sure our message is getting through and cutting through all of the noise that's out there is equally as important as, you know, the features and capability of the product. So understanding that how we communicate and talk about HAP is, is equally important. That's why we've, we've made it such an important role here at the organization.

I love it. So give us a little of your journey though. What, what did, what did you do before?

Hap hap as you mentioned, it's a few years old already, but what'd you do before that? So, I've actually, uh, I've kind of started out in communications. Uh, I've always kind of found myself on the bleeding edge of tech, edge of technology.

com boom, and then kind of rolled the social media wave through that. Uh, and then had a bunch of, you know, SaaS startups, uh, then did some time at, you know, fortune 50 organizations leading digital innovation for, you know, global organizations. Uh, and just, you know, the passion that comes from working in a startup and, and building something new and really driving change based on how quickly technology is moving with something I couldn't resist.

And, and that's kind of how I find my way back here to hap is to, to kind of do that, that run all over again and, and, you know, bring something new to the market that's truly gonna make a difference. Very cool. Very cool.

Um, you know, I, I was, I actually just wrote an article yesterday. There was an article in, um, business Insider, in spite of all the talk about AI taking jobs and costing people their jobs and how marketing is one of the hardest hit areas. Communications is one of the most sought after positions.

Yep. From Fortune 50 on down with crazy salaries, right? Mm-hmm.

Mid six figures and and above for, you know, chief communication officers and, and people who have that real skill. Not that they're not using ai, they want communications people who u who can leverage AI but can still, they need to have that human spark, that human creativity, that human, that AI doesn't yet bring for sure. And, and they're in really, really high demand in spite of everything else we're hearing about about these things.

It's interesting. Yep. So every founder, co-founder I speak to, there's always a creation story around, you know, we saw this problem and it wasn't, it wasn't being solved and or it wasn't being solved well, and we think we had a better mouse job.

We could build a better mouse job, we could build a better solution. What was that for Hap Well, hap actually was born out of, um, the recognition that, you know, AI is going to create a lot of disruption for sure. Uh, but there was already a ton of disruption happening in our financial industry anyway.

You know, technology and the digital transformation around banking has been significant, you know, for, for years. Uh, but this new wave, this new cycle was a big risk. Uh, and, and we saw that we got out in front of it because we know that people are adopting ai.

It's not going away exactly what you said, you know, the fear of job loss, but the opportunity is significant if it's embraced correctly. And that's where we saw a huge gap, was in a regulated environment. You know, generic tools that are being built for consumers don't translate well into highly structured environments where governance and accountability and scalability are gonna be critical.

And so we felt like we needed to fill that gap because it's absolutely imperative that we maintain consumer choice in our banking structure. So technology is consistently putting smaller community banks and credit unions at risk. And when that happens, it actually dramatically impacts what happens in these communities.

You know, the large global institutions are not going to extend your loan because they know that you had, you know, a bad agricultural season. You know, they're not gonna give a small business loan to somebody who's, you know, aspiring to open a restaurant that happens at the community level because community banks understand the impact and what that means. And so our standpoint was we have to give technology to these financial institutions that help them compete with less resources, you know, less technical capabilities.

So it had to be simple, it had to be easy, it had to be secure. And that was just a gap that, uh, you know, big tech was not thinking about. Uh, you know, they're looking at, you know, again, it's this, it's the old model of, you know, attract as many users as you can as fast as possible, worry about profitability later, we recognized, you know, we've gotta get, you know, governance systems in place.

We need to make sure that there's accuracy in place, uh, and that, you know, we're helping these institutions navigate through this, this wave of, of innovation and come out the other side healthy and strong and able to compete. I love it. I, I, and I don't disagree with you.

You know, my experience is, is when you get these waves right, like surfing waves, and you look at the surfers in the wave, the smaller guys, they kind of fall into two buckets. There's a group that are more nimble than the really big guys. And so they are able to outmaneuver out agile, if you will, and ride the real, the bleeding edge of the wave.

But, you know, when you ride the bleeding edge, it's, it, it could cut both ways, right? Mm-hmm. And, and you get that, and then you get others who say, look, we just aren't gonna have the resources to compete with those bigger guys, especially in an unknown wave, right?

And let's, let's see, let's see how that wave develops before we go all in, if you will. But they gotta do, you know, the thing about this AI is, it, it's different if, if you, I don't know if you could wait on ai. Agreed.

You know what I mean? You can't, I mean, there's absolutely no waiting on it. And, you know, if you go back to 1999 when everybody was, you know, rushing and had to have a website, you couldn't wait then.

But at, in 1999, we had no idea what was coming, you know? Yeah. I mean, if you think about where we were and everybody was standing up a website and we were measuring success based on eyeballs, you know, we, we couldn't tie that back to real revenue.

There was no, you know, it was just, it became a race for how much traffic could we get. And we're kind of in the same place with AI where, you know, we don't know what's coming yet, but in 1999, you knew you couldn't not have a website. And where we are today is, you know, you can't just ignore ai.

It's not going away, um, where it's going to go that, you know, anybody who can predict that, I, I mean, I'd love for them to tell me the secret, but, uh, you know, it's moved. They're probably not on the show talking. You and I, they're out, you know, dressed in on their island somewhere.

Exactly. But, um, You know, Kevin, it's interesting you bring it up. com and we did all of that good stuff.

And I, you know, I do see a lot of parallels. A lesson I learned back then was the bubble bursting wasn't because the technology didn't work. The technology was great.

The, the bubble bursting was a financial technology. com mm-hmm. Right?

com. com. I mean, they were attracting crazy money Now with, with ai.

I, so I, I tried to say, where are we in that cycle right now? And as near as I could tell, I think we're still in 96, 97. Yep.

Right now, we're probably a while out from, you know, that that financial bubble bursting, if you will. Um, that doesn't mean it won't, it doesn't necessarily mean it, it will in two years or three years either. But that, that's kind of how I, how I, you know, peg it when I look at that timeframe to, to, to this timeframe.

But, you know, the other things that we're seeing here, you know, and Kevin, the, the first, this first three years, three, four years of this AI boom has been financed on the hoards of cash that, you know, the, the hyperscalers, the big tech guys have been sitting on, just, I mean, they've been sitting on it since before COVID, they were building up war chest. And so all of these AI data centers and buying all these chips and, and a lot of the research and the, the circular investments has been, you know, off of that hoard. But now we're seeing that second wave of companies that are financing AI with debt, with VC money, with other people's money, as we say, right?

OPM. And, um, you know, it's a different, it's a kind of a different space. It's a different rhythm, right?

That money has to have returns, have dividends within a given amount of time. It's not free money. Certainly.

Um, and it's gonna be interesting because there will be a reckoning. I'm, I'm not saying the bubble's gonna burst. I'm not saying it's gonna be exactly like the dot coms or, or some of the other, I look, we had a real estate bubble in 2008, right?

Um, the mortgage bubble. I'm not saying it's gonna be exactly like that, but what I'm saying is that the financial side of this is so overheated sooner or later, the music stops and someone doesn't have a chair. Yeah.

And you're exactly right. And I, and I think even your timeline is right in terms of from the financial aspect. And the interesting is, you know, societally and, and where kind of the technology is, you know, it, it's very reminiscent, and I don't you probably remember, but you, I don't dunno if you're a football fan, but you know, it was too huge.

com Super Bowl, where 20% of average Yeah. Dot com companies. You saw it this year.

Did you watch this year? Yeah. Now I gotta tell you the, the, uh, what's her name?

Claude? Well, it wasn't the, the lady was from OpenAI. The, the old, you know, the, the, uh, I guess Android, whatever you wanna call it.

But the Claude, I thought the Claude commercial was great, but it got panned. A lot of people didn't like it. I I thought it was a great commercial.

I, and, and they all were. com. Dot com.

So this is now the AI bowl. This Is the AI bowl. Uh, But here's one thing I found out, the Budweiser commercial with the Eagle's wings flying over the horse, not ai.

Yeah. Not ai. It was real, all natural done.

They just caught it. Right? Yeah.

Hey, I love that. That was my favorite commercial, but yeah, Me too. Um, you know, but it's, you know, we're seeing kinda the same, the technology is moving so much faster, but even in that, you know, that correlation, to your point earlier, the internet didn't lose, you know, like the, the bubble may have burst financially, but the internet still won.

Yep. And when the music stops, there will be less chairs. And that's kind of what we think about from hap ex's point of view, is the ones who won were the infrastructure.

You know, the ones who transitioned to, uh, you know, not just thinking about new features and new functionality, uh, but they went directly into how do you start to get measurable, ROI, how do you, uh, you know, really build something that's scalable, uh, you know, and move just again, beyond the, the next feature. And, and we're even seeing it, I mean, if you think back to, again, that timeline, there was a race on features. You had, you know, the Yahoo versus Google, uh, you know, now we've got OpenAI versus Anthropic, uh, which we saw in there in the clawed commercials specifically going after Yeah.

Chat, GPT. So we're seeing these same parallels. And, you know, for us, we wanted to pivot to how do we make sure hap pix becomes infrastructure before, you know, instead of getting caught up in the, in the feature race, because Right.

Even, even at GPT, if you, I mean, how many people are actually using OpenAI's Atlas browser? You know, not a, not a ton that launched in October last year. Well, the, the, to be fair to them, how many people are using Perplexities or some of the other ones?

I don't know if, and, and I'll tell you why. So I'm a geek. Just probably like you are too.

That's why we do what we do. Right? You probably downloaded Atlas, you checked it, didn't you?

Yeah. So I Of course, yeah. Because that's what we do.

That's why you and I are on here. But you still using it? No, Me either.

Yeah. Why? Because it did, it wasn't, it wasn't a killer app.

There wasn't, I didn't see anything that said, oh my God, I'm gonna stop using Chrome for this, or Safari or whatever it is. Mm-hmm. Um, but that's, but that's tech, you know, it, it, you know, you gotta, you gotta kiss a lot of frogs until you find your princes here, or princesses or whatever you're looking for, Without a doubt.

And, and, you know, you could even make the correlation to the mobile movement, you know, remember, yeah. Everybody was mobile first and there was a new app on the number one on the app store that you had to have. And then, you know, now, I mean, was the average person uses less than 20 apps consistently on their phone.

Yeah. You Know, but I still have 150 on there, and I'm, and whenever I go to delete one, I say, but I, I might use it. I might use it.

Right? Oh, totally. I, six months later, I, yeah.

Discipline is the word there, Kevin. Yeah. Discipline.

I've got some that are like, older than my kids. So Yeah. I, I, you and me both let, let's talk, let's look at six, 12 months.

com, the rate of evolution here. Like, we used to talk about internet time, remember the crunching of internet time. Well, AI time crunches, internet time, right?

Mm-hmm. So the maturation is, is just crazy. Looking at over the next six to 12 months, what do you see?

What's in story? I mean, are we gonna have little banks, community banks? I mean, you know, from our standpoint, absolutely.

Um, you know, we think that, you know, this technology allows smaller organizations to compete. And that community banks and credit unions are well positioned. Because one of the biggest challenges that comes with AI is the operationalizing of it.

You know, it's making sure people are educated. And, you know, we've done surveys, you know, 57% of banks say that they are AI mature. Um, a ma AI maturity is not, well, everybody has a license to chat GPT or copilot.

That's not maturity. So, no, what we're trying to do is help people operationalize AI and truly kind of reimagine some of the workflows, uh, that currently exist to do them better. And to our earlier conversation, smaller, more nimble organizations are going to find success faster, as long as they operate with discipline.

And that's the thing. I mean, Amazon survived all of that. I mean, remember, they were losing billions of dollars a year and everybody was criticizing everything they were doing, but in the background, they're building logistics infrastructure.

You know, everybody thinks they're an online bookstore, and that's not what they were doing, right? And so, and I'm not saying we are the next Amazon by any stretch of the means, but that mentality, it's, you know, thinking about it that way, as organizations think about how this is infrastructure and how it changes their operations versus it being assistance and chatbots, those are the ones that are gonna win. Uh, and I think smaller organizations are gonna be able to, to pivot, um, you know, pivot to that type of approach faster.

Love it. I love it. ai.

Uh, what advice would you give people? How, what's the best way to engage with y'all? Yeah, so I mean, certainly, you know, people can reach out.

Uh, we've obviously can multiple contact methods on the website. Um, but one of the things that we did do is recognizing that enablement component. Uh, we did build a free AI certification specifically for bankers that's also accessible on the website.

You know, our mission really is AI only works when people truly understand it and embrace it from the top down. Um, so, and, and it has to be, you have to understand it through your own unique context. So your role, your industry, your business, you know, general education and training, while helpful doesn't translate into execution.

Um, so, you know, obviously through the website is the best way to reach out to us. Uh, you know, taking advantage of the tools and the resources that we put up there, uh, hopefully will help people start to pivot away from what we have a chat chat bot. So we're, we're AI native to, uh, we're organizationally structuring ourselves to make sure that we can operationalize around ai.

I love it, man. 15 minutes goes so quick. I'm sorry, but we're outta time, Kevin, come back, keep us posted what's going on here with ha with, uh, HAP X, and I wish you all the best.

I, I appreciate it. Great conversation and definitely look forward to coming back. ai.

Check it out. We're gonna take a break here on text on tv. We'll come back, we'll be right back.

Hey guys, thanks for the, we're here with Eric Fourrier, who is CEO of gi. Uh, let me, it's GI guardian. Guardian GI guardian.

Yeah. GI Guardian. All right.

Try that one more time. Here we go. Hey guys, thanks for the, we're here with Eric Fourrier is the CEO of Get Guardian, and we're having a little chat about keeping secrets in application development in the age of ai in the wake of them picking up an additional 50 million in funding.

Eric, welcome to the show. Hey, thanks. Bother me, Mike.

How we buddy? Alright, So, um, you guys have been at this for a while and I think it's, the landscape is fundamentally changing, though with the rise of these AI agents. So what's your perspective on what's going on here?

Because a lot of people are not, well, they're not sure. Is this a new type of identity that we need to keep secrets for or Trump? Or is this just a non-human identity that will manage much like any other non-human identity?

Yeah, it's, it's, it's a tough question, Mike. I think, uh, when you think about it depends on the workflow, but it's, it's a bit of both. So, uh, when you, when you think about it, like you take an AI agent right now, it's the, the reality is like every company right now is trying to deploy AI agents and an AI agent to be efficient.

It needs to have access to data and need to have access to different data sources. Could be your, uh, CRM could be your product data and, and, and to get access to the data it needs credential, or it needs API keys, it needs tokens. And because what's lacking right now, it's, it's very hard to identify, uh, an AI agent and get access to safely to different resources.

So people are doing it the old way, which is giving them secrets, which, which is a huge problem when you think about it because it's, it's accelerating the, uh, the secrets post. The secrets end up like leaked everywhere. And that's what we've been doing at Guardian for the last six or seven years, is trying to look for these secrets that are leaked in every system and to make sure attackers cannot choose them and to protect our customers.

And yeah, AI agents have drastically increased, uh, the number of secrets used and the number of secrets leads, and there is currently no way to, um, to, to safely secure them. And, and, and that's really like the big problem we, we seeing right now. Mm-hmm.

Um, those AI agents, um, how good are they at keeping secrets? 'cause you know, some of the people I talk to say, you know, the guardrails aren't so good and they're about as effective as keeping a secret as your average five-year-old. Uh, that, that's a good question.

So, on on that part, it is really interesting because at the end of the day, especially with, if we look at a bit at coding agents, so it, it has evolved a lot over the past like two or three years. So when, when, you know, the first model of, uh, OpenAI started two or three years ago, there were actually, you could actually, um, with the right prompt, it was super easy to get like actual secret from the training set. So because, you know, these models have been trained on, on lot of open source code, and inside this open source code, you have a lot of secrets and credential.

You could actually get like credential from somewhere else on, uh, from, from a request to the model. But hopefully it has evolved a lot. And now the coding agent are, you have more, a little bit more wild ways.

They are more powerful, but the real problems come from the person using the LLM. Its, and, and that's, I think, the world danger of white coding that a lot of people are underestimating right now. It's, it's really, you have people that are non-developers, could be marketing, could people, could be salespeople, could be ops, like, could be people that have no training and software engineering that have no idea what, what it's credentialed and how they should be managed that now are able to vibe code application.

And, and that's usually them, uh, making the worst mistake. Even if the, the agent or the LLM is not doing them, it's the human that will introduce the vulnerability. So you have this combination of both, like an experienced human with like this coding agent that push you to be extremely fast.

And like you have this, you know, the hype of, uh, of the, of the trend that makes, that forces us to say, okay, we need to adopt fast, all these new tools. We need to release this AI agent to our employees, but we, we need to do it fast. So we, we just, you know, we don't do, we don't put any security guard rates.

So the combination of these three things, which is untrain people, uh, pushed to, to, to, uh, the velocity on, on, on the code and, and the, the absence of an insecurity galleries, uh, make it a critical, critical issues, uh, for, for secrets and for, for, for companies right now. So I think 50 million is still a lot of money even in the age of ai. But, um, what is the plans or what do you think guys thinking about investing then in Yeah, it's, it's great question, Mike.

Um, I think for, for us, it's on, on two side. Um, so first is, um, expand our current GoTo market, uh, operations. So, uh, GaN was created in France, but you know, our, we have a very unique, uh, story for our French company.

Like our first customer was US based. Uh, we now do 70% of our business in the us. Uh, north America is our fastest growing region, but we, we really started like having reps in the US only two years ago.

So, uh, we have 25 people right now. We see a tremendous traction here. So, uh, we definitely want to double down, uh, in the us And at the same time, uh, over, over the next two or three years, uh, there is tremendous opportunity.

And we have already customers in Asia, so Australia, India, Japan. So, uh, we also want to open new region, uh, both in, in Asia Pacific and, and also on the Middle East. Um, and, um, the other side is product, like, as you've seen, like AI agents, uh, the pace is, is so fast.

There is new, new innovation, new type of agent and new type of ities almost every three months. Uh, so we need to keep the pace and build more and more product, uh, so we can help, uh, our customers safely, uh, deploy AI agent, uh, while securing like the credential. And we need really try to, to, to make them, uh, them more safe and, and, and really give them the ability, uh, to, um, to deploy them and enjoy the, the velocity and the a OI while, while being also safe at the same time.

Is this whole shift to AI gonna exacerbate an existing problem? 'cause I think a lot of developers aren't very good at keeping secrets themselves. They might be slightly better than the AI agent, but it seems like one's just gonna compound the other.

So what do we need to think about doing here to kind of protect our secrets? Yeah, it's, it's a great question. I think definitely every coding agent, uh, is making developers like more productive.

So, uh, give them the ability to write small lines of code. So if that already, which is the case, you're right, already leaking secret, it's just accelerating the problem. So it's why you need, uh, you know, shiftless have been, has been so important, uh, has been really important in the, in the past, the past five years, but it's even more important right now is like we should equip developers, uh, as soon as possible.

And we, uh, have tool that are, uh, really strongly suited for the agent, uh, workflow, for coding agent, uh, on the laptop of the developer. So it's, it's basically developing a series of, of, of tooling and product that are very close to the developers. So you can, uh, you can help him write, uh, more, more secure code and avoid leaking credential, but also build like awareness, uh, and, and educate them on how really they can safely manage secrets.

Will we maybe have an AI agent that will go look for all our secrets at some point, and that will just be part of the DevOps workflow? Yeah, it's, it's a great question. So the, the, I think on the, on that, that's a question like we, we have asked ourself at the guy.

So we, we have our own detection engine to, to detect secrets and like, can, can we use an AI to do it? So for now, the economics is, is still not there because, you know, especially for our customers, we need to scan like terabytes and sometimes petabytes of data when, meaning when we actually look, uh, for, for secrets in new data sources like SharePoint, Google Drive, you know, all the, uh, JIRA tickets, all the docker images, just so, so much data to scan that at the end of the day, you, you cannot send everything to an agent or otherwise you, you will pay like open eye or on tropic, like tens or 20 millions of dollars to scan your data. So it's, it's why where you need like products like GI Gallian and be smart about it.

We call it hybrid detection. So it's a combination of a first step that's, uh, deterministic engine. And the second step that's really like, we use also AI in our product to do prioritization and to help our customers better remediate the secrets.

But at the end of the day, the first stage is, uh, needs to be, uh, deterministic, fast and able to process a lot of, of data that far now, uh, lms, uh, are unable to do. Do you think the bad guys are gonna get better at using AI to find our secrets? And this may be a big part of the problem going forward?

Yeah, it's, it's, it's a great question. I think like the, the big issue, um, and it's, it's, it's a, it's an issue that has already existed in cybersecurity for a long time. It's, which, which exists in life.

It's, it's always easier to, to destroy stuff than, than build stuff. And, you know, it's, it's exactly the same for, for cybersecurity. It's the job is, is is unfairly easy, uh, for an attacker to, to do damage compared to the defender.

And, and yeah, AI has accelerated that a lot is like, unfortunately now with ai, that accounts have the ability to, uh, they will scan the, the company data for, if they enter into a system, they will scan the, uh, the system for secrets. And after they will probably use an AI to ask like, okay, what are the most important secrets? What are the secrets I can use to move laterally and do more damage?

And unfortunately, LLMs are very good at that. So, uh, that's why it's, it's so important. Like, and it's, I always tell like, customers and prospect is like, now we need to move into a world on, on zero trust and your data.

So you need to consider, if you have like secrets giving, you know, I always say giving the keys, keys to the kingdom and giving the lot of access, uh, you need to consider that they are gonna be compromised because you will get breached, an attacker will get in. And with the AI can be what would take like, you know, 10 days of, you know, scanning secrets, uh, evaluating them, trying to find the right, uh, the right way to enter into a system can now take like, you know, minutes, tens of minutes. So yeah, it's, attackers are now incredibly powerful with ai.

What is the relationship between the application development teams these days and the security folks that are nominally responsible for all these secrets? Is that getting better? Because, you know, historically they've always kind of been two ships passing in the night a little bit, but, um, you know, are you seeing more collaboration and what's driving it?

Yeah, again, very good question. I, I think it's, it's, um, the relationship you write has always been like, it's almost like you had a wall between, between teams. But I think, and it's really, um, my vision for GYN in the next year, at the end of the day, like the problem of of secret security and non identity security has become so important that you need to, uh, first you need to, to have a product and a platform to address it.

But at the end of the day, like for us, like we need to have like the application security folks, the identity players, the developers, the DevOps, like teaming together to solve a problem. It's not like, it's not the responsibility of one team or the other. At the end of the day, it's like you need, and that's, that's why it's so difficult to build, and that's why it's so valuable is like, you need to build a tool for, for every, uh, persona and for every type of, of, of people.

And, but at the same time, give them the ability to collaborate together. So I think they have no choice. Uh, the problem is so important that, uh, it needs to be tackled from multiple angles.

Uh, developers needs, uh, tools inside that development workflow to avoid leaking secrets, application security teams need, uh, a holistic, uh, workflow and, and dashboard to be able to map all the, the issues and focus on what's the most important and the most critical, uh, piece. Um, and identity folks need to provide a way to, uh, app on the remediation, making sure we are like secrets manager in place, making sure we are short lived, uh, tokens and identity. And, and it's a teamwork at the end of the day, uh, you, you need to work in teams and, and that's how we, we see the future with get done.

And we were gonna orchestrate, uh, all this workflow from, you know, um, detection, prevention, remediation, and, uh, governance and hygiene. And, and it's gonna be, um, it's gonna be out, it's gonna be long. But yeah, that's, that's where we see the value, and that's, that's our mission.

So what's that one thing you see teams doing that just makes you shake your head a little bit and go, folks, we gotta be a little bit smarter than that? Yeah, I think a, a great question. Um, I think like the, the, the big shift I'm, I'm seeing, and I, I've seen it, um, you know, from the early days of GI Gian when, when I started GI Guardian.

So we started with, with public monitoring Sogar style. The first part of GI Gian was like, we, and we were, well still very known for that, is like, we find like secret exposed on public data, like, which is today probably one of the most critical vulnerabilities you have, your AWS secret, you have a database secret sitting like on public a so everybody in the world can see it. So very dangerous.

And, you know, in the early days of good gum, people were like, yeah, it's, I understand the vulnerability. And I was trying to educate people on like, yeah, but, and you also probably have secrets in your internal code base. And that's, that's also a big vulnerability because if an attacker can find it, it can make damage.

But yeah, people were arguing like five or six years, years ago. Okay, it's, but it's in my internal code base, so it's safe, it's behind my, my internal firewalls and systems. And, you know, the world has shifted because yeah, code has been designed to be shipped, uh, distributed in, into binaries, docker images.

So, uh, people realize, okay, like definitely if I have secrets in code, it it, it's gonna be super easy for an attacker to find it. So it becomes a strong vulnerability. And I think now we are at the stage where even like where places where we sold it was safe, like, uh, especially on the laptop of dead drops in our environment, viable, you have, like, especially on all this, all the vibe, colors, and even devers have so many secrets sitting on, on the laptop and, and on places where you stay.

It's, it's, um, I think I'm safe. It's not that bad. But right now with like all the AI agent and all the, the open doors and the, all the circulation attacks we have, even these places where we sold like, your, your identities, your secrets were safe, are not safe anymore.

And I think for me, it's one of, uh, the biggest, uh, you know, attack vectors. I see right now it's, you know, the laptop of the developers, uh, because you know, you have so many secrets. You have so many privileges.

You have so many AI agents getting access to, to data sources. It's, it's, uh, it's a crazy, it's a treasure of our attackers to, to just see them. And, and the issue right now, it's also what I was saying at the beginning is like, now you have people that are un trained, that are non-developers, that are beginning to code, that are beginning to get secret.

They're beginning to launch applications. So it's just expanding, like the threats for company at, at the pace, um, which is, which is very hard. And that, that they don't have the current security tools to secure right now, because nobody has thought that, like, yeah, the, the, the person at the marketing was going to be, uh, now installing software, building software, having secrets, having a database, and, and yeah, it's, it's a tall new, um, new domain and new attack surface to, uh, to defend for our companies right now.

All right. Well folks, you heard it here. If you think your secrets are safe, well, thank again.

Hey, thanks for being on the show. Thanks, Mike. All right, and back to you guys in the studio Control.

This is agent dev. I'm in position. Copy that.

Dev, Stand by for go Standing by. Hey, welcome everybody. You have joined another great episode of Agents of Dev, the podcast where myself, Mitch Ashley, and my co-host Bradman, both of us from the future room group analysts in our respective areas.

Al always talking about software development and what's happening, what we're working on, but also mostly what the vendor community is working on. Brad, good to see you again. How's it going today?

It's good to see you, Mitch, enjoying this glorious spring oh eight. We're a little ahead of that. The presumed coming of spring in a month or two.

Yep. Just doing good. Spring wasn't very springy yet, so we'll have to wait for some more spring to arrive.

It's a bit an arctic blast as That is for sure. That is for sure. I think we're getting some snow this weekend or something.

They're talking about it. We'll see what happens. Nothing like what you've seen, of course.

So that's crazy. You know what this is, I think we should do a podcast about every other day, maybe going to every day, because it is just that freaking insane. I mean, there's that much stuff happening and, uh, you had reason, reason why I kind of set it up that way is you have a new segment you wanna introduce, um, which kind of Oh, I do speaks to this.

So tell us about what you wanna do. Yeah, we, we want to try Tora all of what's happening that we think is important and useful, uh, in the marketplace. And to do that, as Mitch says, we would be doing this every hour on the hour.

Uh, but because we, we don't have the ability to do that, and we wouldn't test your patients with that. We instead have decided to go with, with a new section we call the callouts. And this, this is a segment that will be at the beginning of our podcast where we'll just take a moment to talk about something that we see a vendor, uh, in the markets that we cover, uh, doing something we, we think is useful.

And, uh, I'll, I'll, uh, I'll start us off, if you don't mind, Mitch. Great. Absolutely.

Please do. Okay. So, um, as some of you who follow the, the research practice that I, that I'm working in, uh, we're in, I, for example, am talking a lot about the semantic layer right now and the importance of that across the board, not just for BI tool, you know, tooling and data scientists, but for everybody.

Um, we have been following this idea of, you know, the semantic layer as being needing to, and it did start within BI solutions as a means of saying, what does this business term mean and what does it mean to different people? And that's important because you kind of have to have consensus. And as we talk about on our podcast, you know, having that ability to have that ground truth of meaning, uh, is critical for AgTech development.

So, uh, I I say that as a starting point to give a bit of a clap for, for Microsoft. They just released two ga, uh, the, uh, uh, a sort of evolution of an existing tool they built a while back called Semantic Link. This was built basically to, you know, sit inside of Jupyter Notebooks and allow them to, again, my, my favorite place, uh, and, and allow them to, to basically, you know, build and work with, uh, these semantic models or semantic layers, I should say, uh, and coordinate with Power bi.

So you didn't have to like, treat them as two separate environments. Well, with this update, they've actually now sort of con, not sort of, they have conjoined this within their entire fabric, uh, platform. So whether I'm a data scientist, a bi professional, or a data engineer, or a software engineer or an agent, uh, I, I will be able to, you know, work with these semantic models and change them to make, you know, uh, create new models to create new data products, uh, if you will.

And to have those available to and reflected across, you know, the, the Power BI group, the fabric group, and my own group that I'm working in if I'm building out software. So, uh, I think that's terrific. I'm, I'm glad that Microsoft is evolving the technology that they have and not just trying to reinvent Kind of what's old is new again.

Hey, there's nothing wrong with that. You know, Hey, take what, what's working and extend it into this new environment. Yeah, We're gonna talk about that a lot today, actually.

We're, Yeah, the theme coming on, um, I, I have a couple, one tiny one run a tiny, but one short one, um, Atlassian, uh, ga their, I think it was into the last month or first part of this month, their n CCP server for robo agents. Basically, they've opened up through robo, whether it's a queryable interface or for the agents, all the information that's in Atlassian's applications, primarily Confluence. A lot of it is what we use.

So it can be for operations, it can be for development. Um, and of course I think that would benefit from a semantic layer, something they, that they've talked a little bit about. Um, but I think the, the thing of it is, it kind of reminds me of this, I was on this project called Project Alexandria, and it was all about the, the library of Alexandria.

Did you have a library? Yes. No, no.

We were gonna assemble it. Well, it didn't quite work out. But anyway, um, you know, it, it's kind of what, what we're doing across the board.

You know, Google has made available, uh, all their documentation to agents to be able to use for technical documentation, user information, data for the business. So, huge trend, I'm sure, you know, not news to you. Uh, but that seems to be more and more sort of a cost of doing businesses.

Opening that up. Yeah. The second one, just to not take up too much air oxygen out of the room is, uh, kind of, kind of hit the airwaves pretty big, uh, is entire this new company that, uh, was formed by the CEO, former CEO of GitHub, uh, Thomas Domkey.

And everybody was kind of wondering what did, what, what did he lead to, to do? Said were A little worried when he left. Yeah.

We know. Like, I, well, what's gonna happen? Well, guess what, again, something you'll love, and I think all of us appreciate, is you think about software development and all this interaction we do with models to create it, and what, what gets lost?

What's the exhaust? It's all the interaction, the prompts and everything that we talk about, whether it's through scripts or, or, you know, predefined prompts or interactively, that stuff just kind of falls by the wayside. And an artifact why It started with, uh, this en entire, uh, uh, a open source project called checkpoints, which basically you connect into your GitHub and it tracks everything that you do in terms of capturing.

You have to install something locally too, uh, which your prompts are to whichever mo whatever model that you use. And a, it's a great idea. B, that's just the tip of the iceberg.

I call it the plumbing of what he really says he wants to do. And I, and I believe this is true also, is if we're gonna generate massive amounts of code, just like we can't scan it all for vulnerabilities, we can't as humans sit there and review all of it. That's just impossible.

We don't even know why we're reviewing what we're reviewing. So this kind of semantic layer is not just for humans, it's also for agents. So it knows what got created, why, how that ties back to what session, who, what the requirements were when we defined that spec or whatever was that driven drove it or the idea came the internet, what whatever it was.

Now we've got some traceability back to that, but I think it's a lot more, I think the code windows will of our ides will eventually fold into agent interfaces. I could described it as, you know, more, more like StarCraft kind of game interface of managing all the agents, doing all These fun. Yes, you had me at StarCraft, which I hope Like that, something like that.

So anyway, good stuff. I mean, a again, center your world semantic data knowledge, it's, uh, that's the fuel of AI and what we're doing. Indeed.

It is. Yeah. And it's actually funny.

Um, this topic is my, uh, relates to what I wanna talk about in the drop at the end of our show, uh, with, with, uh, agent context to Decoherence. So, uh, yes. Yeah, it is, it is timely to me.

Very good. Well, let's, let's get into our, our main section or our main topic. We wanted to talk about philanthropic, uh, published a, I didn't wanna call it experiment, I didn't wanna call it research, even though it was a research, a company blog post.

We could call it marketing, we can call it whatever, but it's a, a technical person, A repo and a repo. Don't forget Repo. It must be legit.

Exactly. Right. But that is our measure.

Got a t-shirt, we got a repo. Okay, we got a blog post, we're legit. We've got a company.

Let's rock and roll. So anyway, um, they did this, I'll say, I'll say the word experiment. It's not experiments in a classic science.

I mean, it's not independently tested. It's not isolated. Yeah.

It's, you know, they try, they said, what, what could we do to solve a really hard problem? Let's pick something like, could we create a c compiler? Now that's an interesting problem because it isn't an unknown unknown.

We know what a c compiler does, right? It's very, been around for a long time. It is A herculean task.

Is it not something, But it is a massive task. Complex. A human Endeavor.

Like going to the moon, building a compiler is not that far apart, Except this has been done. This is like going back to the, the moon, right? Oh, yeah, yeah.

Right, right. And the, and the, the the, I mean, I could cite the numbers of what, how much code was generated. You know, it was a massive amount of code.

And they used long running, uh, multiple agents that they coordinated through Yeah. Through Quad. Just 16, no, thousands of agents.

Swarms, as we've been told, would happen. Well, that's more than I can manage 'cause I only have 10 fingers and my toe toes are too far down to to use to cut, So, right, right. Yeah.

Um, interest entering, uh, interesting territory there. But one of the real, real telling things was how much human and how much human direction and intervention and engagement it took to get there. So it wasn't a, here's a really well-defined spec, which you could do with a C compiler.

Go for it. Go build it. Now, it took a lot of direction and, and not, that's not a negative, it's just the state of where we are of you still, this is a guided process.

You still have to engineer a solution with ai. It's not gonna totally do itself. Yeah.

Maybe it can create the snake game. So what, that's not, that's not our problem. A problem.

We have a million snake games running around on the Internet. Indeed. Um, but from a social people engineering perspective, I'm curious what your reaction was to that thought experiment, whatever you wanna call it.

I, I, I think that they know how to market quite well. They, they understand the things that we care about. The, you know, practitioners care about practice, things in practice, I should say.

And so what we're getting from them is, in this case, actually something that's a long running experiment. This isn't just a, I decided last weekend to do it as we had with Claude Cowork. Yeah, yeah.

The weekend We had a, we had a great discussion about that. But this is different. This is something that this fellow who's been famous for doing a lot of adversarial red teaming against AI, has been running as a long standing experiment to see where Claude is in terms of being able to handle these very complex, long running tasks.

And what I found interesting about this is, and as you just said, uh, it is very much a human led endeavor. And, and, you know, we had at fu you know, uh, basically when we did our 2026 predictions, we said, you know, we feel, feel that for most endeavors, uh, for me, for example, data professionals or data engineers and such, that they're not going away. They're instead transitioning into becoming what I would term as an AI shepherd, uh, a individual who has deep domain expertise, great skill as you might see with like a business analyst to define problems.

And to, in what we're seeing in this case is to create a sort of set of tests and a harness to run the test within and the development work itself, and to sort of manage that using the tool chains that we know so well, you know, vis-a-vis GI in this case. And mm-hmm. It, it was, it was fascinating to me that, um, what it, what it showed us, or it showed me anyway, was that yes, it is still human led, but the, the thing that's interesting is that it's human led, uh, for agents.

Meaning the, the person isn't trying to do things the way humans did it. The person is trying to create this infrastructure that does things the way agents do them. And so, for example, but to point Brad Yeah.

Give an example, because they didn't tell it how to do things. They said, do the following, and it figured it out. Right.

Well, you're right. But what they found, for example, is that agents all tackled the same problem repeatedly over and over again. You did.

Um, and, and so to, you know, sort of do two things. First was to give them, uh, the ability to say, I, I called shotgun, I got dibs. They built a locking, he built a locking mechanism.

It's just a very simple text file. It's like a library card to check in and out the books. And the second thing, which is really, you know, a testament to how all of our technology is built upon the shoulders of giants and the shoulders of the giants that came before them.

And that is to use GCC itself mm-hmm. As a reference point to basically say, okay, the things we actually need to worry about that are broken, you know, this is what you need to focus on. The rest, forget about it because it's compiling it or GCC.

So don't try to fix that. Uh, that's like, did we do that before for engineering? I don't think so.

That's a different way of looking at it. You know, it, it's, there, there is a, an approach, it's a very unique situation there with, with ecc, but there is approach of, there's some things that I'm gonna assume, right? I'm not gonna worry about the job runtime engine, right?

That that's pretty much a given. I'm gonna worry about these things. Um, but listen, let's don't boil the ocean, ocean and try to go reinvent a bunch of stuff.

I'll use open source, I'll do whatever that that kind of goes out the, the window with ai, because AI would love to generate Java runtime engine For you. It would prefer to Yes. Prefer to.

You know, the, the thought I had about the agents kind of doing multiple things is I think they forgot to introduce the agent to this thing called the Kanban board in sprints, where everybody goes to the board and says, I'll take that, that feature, that story, and I'll implement that. That's the shotgun version of it. So we kinda need to teach agents that, you know, there is a way of managing your work, um, which I think is a, a whole big area of, it isn't just orchestrating, defining linear process.

It's just like you have to teach ai how is it gonna work with it with itself? And sometimes those agents aren't gonna be the same vendors. It's gonna be other, other systems, Right.

And with interdependencies, you know, some in real time, some asynchronous, and you need, you know, tooling to, to actually manage that flow, uh, in parallel. And Kanban boards are a great way to do that. And is it not a tool that's available on, on GI now from GitHub now, uh, to do that?

And, and so I, I would presume that as we go forward, whether it's, you know, the CEO creating his own system, or it's gi you know, GitHub, uh, evolving, uh, on its own or Code Berg or all of these repository management platforms, they're all going to be, I think, tooling for both the human and the agents or the ai, uh, as to, you know, combined, but some somewhat different tracks. Mm-hmm. Well, you know, I imagine I was, I wasn't joking when I said the UI of the future.

I hope it looks like StarCraft, but imagine in that UI of the future, it's, it's some souped up version of a Kanban board that you're watching as AI is doing the work, picking out things, going through the process. This is, this is one, this is the human one. Mitch is gonna do that.

Whatever that process Looks like, say this for later. Looks Like. Yes, exactly.

Don't, don't, don't go down that, that alley. That's a dark alley. Let's go back here.

Gonna Get mugged. Don't go down there. Yeah, Yeah.

There's danger down there. You don't wanna mess happen. Um, but something like that where, where you can interactively as well as autonomously, see what's happening, and also direct where it needs direct in real time.

And then take, if you, if you take the, uh, kind of the, uh, entire idea, all the semantics of what happened and why can then be applied to learning, to how to work better, how to perform that work better. We started here and we ended up with this. How did we get from A to Z?

Well, it's actually Q why do we end up at Q? We were going for Z. Um, but there, there's so much to take.

And we try to learn that as humans, we can't learn it all, but we have to help the AI learn it. But that is something that we humans, I think, are very unique at. And that is being able to, to pick out the wheat from the chaff, the, to split the two, I should say, and to find what's most important in looking at a lot of information.

And it's ironic a bit that, you know, as, as LLMs are much better than humans at scanning and keeping in context, you know, the much more information. And yet they really suck at trying to pick out what's important. And, and I think with this, you know, uh, anthropic example here with a compiler that we, you know, he discovered that as a big problem.

And so he changed the tooling so that it wasn't just publishing the entirety of a, um, an error, uh, or even the log file. It was just, okay, let's build another epi. I call them epicycles.

Uh, because for planets, planetary motions, how do you explain retrograde? You just draw into their circle Escape philosophy and retrograde. Exactly.

It'll all work out. Newton would be happy. Um, so it, it, it's, it's basically saying that, okay, we need to have another layer that, that sort of modifies the app, the artifacts that are created as a part of this process of building software so that we can better work with those assets in a more efficient and reliable and consistent manner.

And that's what I think this, this points out with that example, uh, that he discovered is that, okay, you can't just throw everything in a 1 million context window and and hope for the best Or 16 of them. Yes. Right.

16 million Tokens running around simultaneously, You know, so this leads me, I wanted to get your opinion about this. You know, it's only February, right? I we're into the second month of the year, but looking back last year very much was a big theme was MCP and A two A and a ccp, and a lot of OA lot of open standards if you open source open standards.

Yeah. There, there's still very much that vibe that's happening concurrently. But I think we've advanced to the next phase of evolution of how we're creating software, where we're seeing different kinds of innova innovation and levels of innovation that standalone still.

So for example, you know how Claude manages work, right? It's not tied into some open standard that everybody's agreed how to do that, right? Everybody has said, Microsoft, here's our agent hub GitHub, and, uh, you can bring your agents to our environment, right?

So we're very much creating these ecosystems of work and development. Not saying they won't be able to work together, but we haven't come to the, oh wow, we're all sitting at the same lunch table at school. We should like talk to each other.

Lets decide how They're working on the Same problem. How to change this all of us. Yeah.

But we're kind of evolving to will, will we suddenly all go off? And, you know, speaking of your planetary, will we all kind of, you know, lose lose our, lose gravity and be thrown off into the galaxy or, or what's gonna bring things back together? Certainly customers will, but I'm just thinking through those kinds of constraints because when we change constraints, things change drastically.

They do, don't they? And I, I think, you know, just as we've discovered with LLMs, they, they, you know, whenever their training cutoff is they really like to run home to mama. And, uh, we as humans do the same.

And I think we saw that in this experiment from Anthropic, for example, and that, you know, he could have, you know, set up some new MCP to, to sort of do be that harness, but instead the harness is just a freaking bash file. You know, that, that just runs till, you know, Ralph Ralph's it till, uh, it it's done. And, um, that, that is a verb Now, by the way Yes, Ralph, that was something different in college, but in, in AIing is A different thing.

Yes. Right. There, there are different pronunciations too you need to stay far away from.

Um, but, but the, the point is that I think, you know, this is a testament to the whole canoe Linux, you know, mentality or the, as we currently would call it, the suckus mentality, which I mm-hmm. I'm not saying that jokingly, that's an actual thing. Mm-hmm.

Um, that you build a piece of software that does one thing really well. And we see that with tooling like SYC and, you know, show me a server that's running the internet that is not using syc, you Know, oh, Just for, you know, backups and synchrony synchronizing file changes. And why would we reinvent that wheel?

Why would you go out and, you know, create a standard to, to solve a problem that, you know, our engineering minds 40 some odd years ago solved. Mm-hmm. And it's still, you know, humming along today just as, just kinda like the semantic model with Microsoft, for example, earlier.

It worked well for that list and, you know, apply it to this problem. Oury has lived on and on and on. Yep.

Yep. It's like building a space station to get to the moon and then use the moon to get to Mars. We'll have to do a, we'll have to do an episode on space.

Maybe we can get, we can get, um, it's his name Tyson, the guy at the planet. Oh yes. Neil deGrasse Tyson.

Yeah. The grass. Tyson, I'm sure.

Yeah. He, he follows us. I'm, I'm confident.

I I bet there's an email waiting for me. He's asking if he can be, we'll see. We'll see.

Anyway, well, I think we've probably done enough damage on these topics so we can, we can go to our, um, our last segment for the podcast episode called The Drop. Alright. My Drop this week.

Yes, yes. Sorry, Mitch, go Ahead. Um, I, I'm, I'm going first because, uh, I'm, I'm really frustrated by this and it is this idea of agent context to Decoherence.

And I've, I've mentioned, I think when we've chatted before in, in the projects I'm working on, I, I feel that the larger the code base becomes, the, the more likely you are to fall off what I call the edge of the plateau, the of happiness that you're on. Mm-hmm. Where, where when you make a change, the change is enacted and it, and it works the way it does, and it doesn't create any other problems or downstream issues you're unaware of.

And, um, I, I still feel that is a big thing, but what really, you know, kind of ca captured my attention recently is just how that applies to all the artifacts that we work with, not just the code base. And so, you know, I, I think as, as the, you know, xe, EO of Git is showing us that we need some smart systems for managing those artifacts. And for me, one that I, you know, didn't, didn't really pay attention to because I, it's just so natural to say, make a memory, you know, in working in whatever CLI you're working in Agent XCLI to say, make a memory of this so you're remember for later.

And what I ended up doing was, uh, something called context poisoning. I poisoned myself, unbeknownst to myself, uh, about how to do, uh, get committing. You do get commits.

'cause I had made a skill for that and uh, then made a memory on how to implement it for different projects. Um, so I could use it consistently across projects. 'cause that's what you do because we're lazy.

Uh, and So make file kind of, That is exactly it, right? That's the point of that. Um, and, and so, uh, I, I didn't, so lurking hidden within not just the skill I made, but the, the memory, uh, so the agents md or whatever you guys are using, uh, what were was an instruction that the model was, was like, oh, wait a minute.

Brad said to do this in the skill for committing, but earlier he said in the, the broader, you know, agent MD file that this is what I need to do. And it went all Hal 9,000 on me. Uh, that was, that was not, isn't The repository you're looking for.

I'm sorry, Dave. I can't, I can't open that repository. Yeah.

So, so, um, so how did you poison it? You have to tell us. You, you can't leave us hanging.

What did you do? Did you drop a little, you know, arsenic in there and like, oops, that wasn't supposed, I literally Pulled out the grip or frep and, uh, just found the, what I was, you know, the poison pill and removed the pill. Yeah.

You knew what the source was. You found what it was like. Right.

Because they only only have Five things combined. Suddenly it's hallucinating. Yeah, exactly.

Something, you know, these are hierarchical systems and so mm-hmm. What you're working across different projects, you have your root agent file, uh, or if we're, if we're in open claw mode, it would be your soul file. Then you have Yeah.

Pro, you know, files for each of the projects you're working on. Um, and I only had like four of 'em, but, uh, it, it was, you know, a testament to the, to the need to, to curate and maintain that memory that your models have along with your skills, along with your extensions in concerts. Because the minute you introduce conflicting truths within that, the models are gonna get confused because that's how they work.

Yep. Very much so. Well, cool.

That, that's interesting problem. And I think all of us are experiencing it, you know, multiple times, sometimes more severe than others, right? Yep.

Um, yeah, this wasn't in an rm, you know, all file system. Yeah. That was, it was not that.

Make sure you have your permission set appropriately and no, pseudo is not a good thing to answer the agent. Um, pseudo, sorry. Um, Sudoku, no different thing.

Uh, so my, my drop is, uh, and this is coming your way, by the way, so this is a little bit of a heads up that, uh, you're gonna get to review this for me, but I've been working on a kind of my layer cake framework of agent controls, uh, framework agent control planes. What do we mean by that? Because it may not be end up being called that at some point, but that's a lot of what are the innovations that's happening around, whether it's managing agents, whether it's directing work of agents, whether it's security guardrails, compliance agent behavior, observability.

Yep. And I think one of the very cool things that's happening is a lot of things we would've done later, like security and like observability and all that, because we're in this age of AI can be easily integrated into the development process, to the kind of interim, it's not quite ready for prime time into production. So a lot of things are combining into this one phase of development in this one quote unquote platform.

So I've got a reference framework, um, I'll run it by you and you can rip it apart and say like that, get rid of that. This is stupid. I, that's kind of, okay.

This will make it better. And, uh, I appreciate your input. Yeah, I'd love, love that.

And you're right, the, the whole control plane idea is so critical, um, and more, more so than it was, we used to have a lot of control planes, right? So like in the ml lops space, you, you know, you're always striving for a control plane that, that wrapped those different, you know, guards and, and rails and, and controls and audits into, you know, four specific areas of interest. And right now, the systems that we're building, I, I think are, are, you know, much more multifaceted than that.

And, and you can no longer think of a control plane as being specific to a given area of work. It has to be specific to the solution itself. Mm-hmm.

And broadly the company itself. Yeah. It's not like one business domain.

Like we're developing machine learning algorithms and Data Exactly. Following. That's the data scientist problem.

They'll deal way Yeah. Let them solve that. But you're, you're, you're creating it for a system that's also evolving, right?

Um, yeah. So how agents work and what they do and what the protocols and what the standards are, the work we're gonna have to do could be very different in six, 12 months or maybe a month from now. So it's, it's, uh, it's an interesting problem to solve.

But I think there's some really good ideas out there, what people, what vendors are doing. Agreed. And a lot, a lot of them built back in 1982, they're, it's the high, it's the high watermark for me in, uh, cultural, uh, terms.

But yeah, I mean the, the old ideas as we were saying are never truly old. They, they become new again. You, we, we either like engineer our way into something new with them, or we find ourselves with, you know, enough compute horsepower, for example, to do the things we couldn't do before.

And so you can never turn your back on the past. You, you have to honor it. You have to look to the past to create the future.

Look, you can see so many examples. Think of line analytics, a Linux and early work has spawned a whole set of great things. So, so often the re there was a reason why those were good things, good decisions made back then maybe constraints and, and situations have changed totally.

But there's still things we can take from it, maybe verbatim, maybe not of those Yeah, good Database management systems as an example, that those were, like, the reason why they were created was to get around the constraints of, of performance and scale. Oh. And the lessons we learned about things like, you know, acid compliance to, to be able to make sure that when you write something, it gets written, you know, little things like that, um, are still true today.

I think we should still be using i a files, but that's a whole nother topic. So anyway. No, no, I'm with you on the Wow.

Yeah. Throwing one at you there. Okay.

And re-indexing everything all the time. Indeed. Well, we better, we better stop before get down one of those dark alleys that we can't get outta.

So, um, look forward to, uh, more great things. Please check out Brad's work that he's doing at futurum. You go to futurum group com slash brad shiman.

Correct. And everything you know, that we do that's freely available, you can download there. There's some things that are part of our internal or our client system that we use called the You Future Intelligence System.

Um, there's a lot of good things there of course, too, but please check it out. Mine is the same, you know, you in Group com slash mitch Ashley, and you know, we're kind of publishing like madman, it's lot, so many things going on we wanna talk about and not just report the news, but understand. No, We're trying to think about 'em.

Why is it news? Right? Right, Right.

Put them in context, wrap, wrap them in our, our decades of cynicism, um, and try to come up with something helpful. That's what we try to do. Exactly.

Well, thank you so much everybody for listening, watching, um, and we, we appreciate comments with appreciate folks that have reached out to us, and we hope you'll of course share this with your friends and follow and do all that we can. We have folks that we're gonna be bringing on, we're talking different people. Yes.

Brad and I are gotta think a good enough rhythm going. We could, uh, you know, keep ourselves on track with a third party, third party involved. I'm not, I'm not sure about that, but I, I think I'm willing to experiment.

Hope, hope Springs eternal indeed. So we'll be doing that. So thanks again for, uh, for being part of agents of Dev.

We appreciate being part of the development community with you. Take care. Everybody Control.

This is agent dev. I'm in position. Copy that.

Dev. Stand by for go. Standing by Dell knobs to Nutanix AI ramps up cyber crime.

Cisco wants to unseat VMware. Checkpoint is making AI security Voltron, Google's buying the sun social engineering our comrades, and we go full Zo Berg in our closer look in this week's episode of the Tech Field Day Rundown. Greetings everyone, welcome to the Tech Field Day rundown for February the 18th.

My name is Tom Hollingsworth and who has two thumbs and appreciates thumbs on thumb appreciation day. That's right. It's this guy.

We're very happy to have you joining us, and I'm very happy to be joined once again by my co-host, Mr. Alistair Cook. Al, good to see you.

Hey, it's a pleasure to be here, and both of my thumbs are very happy to be here too. Uh, I also would like to remind you that it is hockey stick salute day today. So for those of you who are playing hockey here, field hockey is more common than ice hockey.

But, uh, take the salute. Yeah. Hockey stick salute sounds like something you do to somebody you don't like, but luckily we like all of you.

So what we're gonna do instead is bring you some great tech news that you're gonna wanna check out. We're starting off with a story because Dell Technologies has expanded its private cloud portfolio to support Nutanix, which is giving customers more freedom to choose how they virtualize workloads alongside existing options like VMware and Red Hat. The move reflects a broader shift in enterprise IT as organizations reassess cost licensing and risk tied to single vendor hypervisor strategies, initially supported on Dell's Powerflex systems with Power Store integration coming later.

The update emphasizes operational consistency through Dell's automation platform, allowing customers to change or mix hypervisors without replacing core infrastructure by formally incorporating Nutanix into its private cloud framework. Dell is positioning itself as a neutral infrastructure foundation for hybrid and on-prem environments that continue to grow more complex. Al are we seeing Dell diversifying their opportunities to sell to customers who might be looking to make a move?

Well, I think that's absolutely what's going on is that, uh, we've, we're seeing lots of discussion around what should I use to replace the VMware that I was using before, now that VMware doesn't care about the smaller implementations, doesn't care about anybody who's not a VCF customer. So bringing Nutanix in here, bringing the, um, hypervisor that Nutanix has in their hyperconverged absolutely fits into that. We know that Nutanix and Dell have been partners for a long time.

There's been a sort of uneasy piece in that at times when Dell owned VMware. Now that Dell doesn't own VMware, it's very easy for Nutanix to come in as the hypervisor of choice on these medium scale environments. Uh, couple of things that I like in here.

One is getting towards having that power scale as an option as well. Uh, we do see criticism of HCI in being a good general purpose platform, but when you need a highly tuned environment, maybe you don't want HCI. So yeah, definitely seeing some of those, um, those challenges, uh, around higher performance applications being addressed by having higher performance storage is an option, obviously in combination with HCI for the general purpose platform.

One of the interesting things I see in this also is that Nutanix has a story around using their Prism UI for managing everything end to end in a Nutanix only environment. And that brings a little bit of a challenge with the Dell automation platform. That's a key part of this offering, where the Dell automation platform's designed for day zero to day twos deployment lifecycle ongoing operations.

Uh, the discussion then the Dell blog around this is that your operational teams will continue to use the Nutanix tools they've used for potentially years. Uh, and that leaves me to wonder where the Dell Automation platform fits into it since it's such an important part. I think what we'll see is that when that power store capability is added on, then we're looking at an environment that, uh, prism can't manage by itself.

And so there will be value in having a, a slightly more specialist set of people using the Dell automation platform. Uh, one of the criticisms though, of course, will be a continuing criticism of Nutanix that the cost saving is not so much as you might have thought because the licensing cost for the Nutanix component may be significant. We'll see whether there is a special deal and here that makes it that pricing more, uh, friendly to customers over time.

New research from Quorum cyber shows that AI and white labeled ransomware as a service platforms are reshaping the threat landscape. Uh, new ransomware groups growing by 30% per year and attacks shifting from encryption to faster low cost data. Exfiltration steal your data rather than hiding it from you.

Uh, security leaders warn that attackers from criminal gangs to nation state actors are automating up to 90% of intrusions shrinking detection windows and lowering the barrier to entry, particularly with that ransomware as a service executives from bug crowd, SailPoint, quis, uh, profit security and kbo emphasize the need for AI driven security operations, uh, better vulnerability disclosure, least privileged identity controls, and a whole lot of other things like layered defenses. Because cyber crime has been coming increasingly easy to get into and industrialized in its application. Uh, this continues to be, we should be as good at it as we say, we're, doesn't it, Tom?

You would hope so. The problem is, is that the things that we're coming up with now make the other people really, really fast at doing it. We saw this for the last couple of years when it comes to data protection.

They were trying to lock up everything and do it on the sly so that by the time you realized that you'd been hit with a crypto locker, it was too late and your backups were probably dead because they were the one of the first things that were targeted. Uh, in fact, I saw a stat today that said 93% of attacks go after the backups at some point. But one of the things we're also seeing is the response in the data protection market that says, you know, we can detect when things are being encrypted because of entropy detection, which is like part and parcel now.

So we can slam the brakes on that. So now, instead of making a big fuss about locking it in place, they're trying to sneak it out the back door because I mean, they already have a pipe to get in after all, they, they did get into your defenses. But I think the other thing that's really important here is that you need to understand this automation system.

When you get into a system to figure out what you can do with it, there's usually a lot of trial and error, right? You know, do, can I escalate privilege? Do I have any kind of lateral movement options?

Is there a way for me to compromise the system to, you know, provide remote access in the future? You gotta start thinking like an attacker, right? This is a foothold.

What can I do while I'm here? Well, the thing about all of that stuff is that it is automatable. You know, I can try some basic privilege escalation, I can try to install a remote access Trojan to be able to get back in.

And those are the kinds of things that these attackers want to do while they're collecting information. Because one of the things that we know for a fact from seeing these gangs that are operating in some cases they're nation states, is that human resources are expensive because they need to be dedicated to doing certain things. So if I can have an automated script or program, kind of do the basics, and then, oh, did we get past the all the first line of defense?

Oh, great, now I need to get insert the name of everyone's favorite hacker here. I I don't know what, which one it is. But you know, you, you now have the opportunity to kind of do a little bit more on this.

I can expand where I'm at, and then I'm not just sitting here poking around on a system going, well, I keep hitting walls. This is kind of the same idea behind, if you've ever dealt with a telemarketer, usually the person making the phone call and talking to you on the other end of that line, they're not the salesperson. Their, their job is not to like, basically work with you to, to close this deal.

They're literally there to dial telephone numbers. And if they get a hit, if they get someone who says, I'm interested in whatever service you're offering, that's when you get transferred off to a different group, because that's the actual salespeople. They are too expensive to be dialing phones.

They are there to close deals. So I think we're starting to see that, as you said, and it is democratized now. Everybody can do it, but that also means that we're, we're rare verifying the air so that the, the human resources are focused very much more on exploitable systems and not just, oh, you left a back door open, but you really can't go anywhere because of zero trust, or at least privilege or something like that.

Cisco is preparing to launch NF VIS for uc. It's a narrowly focused hypervisor to sign specifically to run on Cisco's unified communications applications, giving customers an alternative to VMware. As licensing changes under Broadcom may push organizations towards more expensive VMware Cloud Foundation licensing bundles.

This is built on Cisco's existing N-F-V-I-S platform, and it is limited to Cisco hardware and collaboration workloads. The new hypervisor reduces external dependencies while maintaining performance and support control. At the same time, Cisco is expanding support for Nutanix's A A HV hypervisor and continuing to promote Flash Stack with Pure Storage that's giving more customers flexibility as they reassess virtualization costs and strategies.

While VMware does remain, the dominant partner at the high end, N-F-V-I-S for uc offers a pragmatic option for organizations whose virtualization needs are really tied to Cisco's collaboration tools. Al, this kind of goes back to the story we talked about earlier. Is Cisco looking for a way to potentially offer options to customers who don't want to pay what everyone is calling the Broadcom tax?

Uh, yes. In simple words, uh, Cisco would like that money instead of handing it to broad, uh, customers handing it to Broadcom. Uh, as a little bit of background on this, uh, as a couple of things.

One is that I worked with a customer a while ago who was migrating from a, an all VMware environment to Nutanix, IHV, uh, but they had to leave some VMware around for some of the products that they were using for collaboration that were only certified on the VMware hypervisor that were not at the time certified on IHV. And so this is that little window that is absolutely being addressed. This specialist, hardware specialist software that requires, in particular near realtime access to, uh, CPU memory resources can't cope with being descheduled because the latency hit that comes from it.

Uh, this particular solution from Cisco, this network function virtualization infrastructure software, nvis or NV, is, uh, for uc, unified Compute. So running on their blade servers, running a specialized hypervisor just to run their collaboration software. Maybe this is, uh, as, as an option, as customers are choosing to use other tools, other platforms for their VMware replacement.

Uh, we definitely saw these customers who were coming up to renewals for their low spec VMware product. So getting the, the, uh, VMware vSphere Foundation product when upsell came, or when renewal time came, the only thing that Broadcom was offering them was the VCF. Uh, these customers who wanted the lower end products, they finally were hearing what had been saying since he started acquiring VMware, I care about the largest companies only.

Everything else is not worth my time. This is absolutely what customers are starting to hear, uh, or continuing to hear. So having options around running some of your business critical applications on things that aren't VMware Cloud Foundation will be vital for organizations.

And this is, again, one of those places where the vendor of that highly critical software, in this case, Cisco, is providing an option to host still in virtual machines. I have to go back to physical hardware for everything, but a specialized solution for this use case. Uh, part of that's because, uh, Cisco still wants to, uh, partner with everybody.

And so they're not at this stage offering their general purpose, uh, solution for a hypervisor that seems to have passed away at the moment. Maybe it'll come back again in another element of our continuing the same stories, uh, in from a different point of view. Uh, in response to the accelerating AI arms race checkpoint has unveiled a new prevention first security strategy designed for this AI era.

According to Chief Strategy Officer, uh, Roy Carro, AI innovation is outpacing security teams forcing enterprises to rethink how security is designed and operated. The strategy is this strategy is built on four pillars, protecting distributed hybrid environments, securing digital workspaces where humans and AI interact, delivering continuous and context driven visibility and safeguarding the full AI stack from models and data to agents and infrastructure. Hmm.

To support the Shift, checkpoint acquired Cyclops security SATA and the team behind rotate and integrated all of these capabilities into an open garden platform, uh, designed to work alongside existing security ecosystems while enabling safer AI adoption at scale. I really would like to see more companies adopting AI safely. Tom, So would I, uh, I wanna go back to those four pillars real quick.

I'm gonna, I'm gonna modify them real quickly. Protecting environments, securing workspaces, delivering visibility, and safeguarding the full stack. Does that sound familiar to anyone out there?

It really should because that's what I like to call in, in the IT space basic security. I don't care who's doing it specifically, if it's an agent or a person, I don't care if it's digital or offline. What's new here is not what Checkpoint is talking about necessarily in this release, but it's what we're dealing with all over the place.

And it's actually something that we talked about on an upcoming episode of the Security Boulevard podcast. The time between when a system can be scanned, f fingerprinted, and exploited has been drastically shrunk because AI has accelerated our capabilities of doing that. If you've seen any of the news, you know that, you know, systems like Claude Opus can uncover massive numbers of vulnerabilities in a very short amount of time that are probably not publicly reported.

So you have to get back to those fundamentals of how do I prevent this from becoming a problem? It's not just that I need to keep the bad people out from exploiting the things that they can touch. It's what can you do to contain them when they get in there?

And I think that that's where this value comes from because you're putting together tools that have pieces of the solutions, whether it's Cyclops security or SATA or any of these other ones, you've got to start looking at what it takes to slow them down. Because a lot of companies in the security space are really good at making a point solution. Yeah, we can protect your login information.

Well, what happens when that fails? Hmm, it's not our problem. We don't do that other stuff.

You need to have a comprehensive solution, and if you don't, you need to find somebody who does. Whether that means putting your own comprehensive solution together with multiple different tools, or buying into somebody who's already done the hard work for you, in this case, like Checkpoint, you got to, because AI has effectively shrunk those windows to nothing. Uh, and I referenced a great presentation from my friend Justin Warren from many, many years ago.

Uh, you know, I back in like 20 13, 20 14, something like that. It's like cloud, right? Because the value of cloud is not infinite resources.

It's absolute static predictability. It's gonna happen the same way every time. It's gonna take the same amount of time every time.

Well, in AI's case, it's shrinking that window to be smaller and smaller. So if your defenses aren't up to stuff now, you're gonna find out very quickly how, how unfortunate it is when people can get past them. Our friends over at Google have signed two 15 year power purchase agreements with total energies to secure one gigawatt of new solar capacity in Texas.

It's underscoring how AI driven data center expansion is accelerating investment in those pesky renewables. The utility scale projects in Wichita 805 megawatts in Mustang Creek, which is another a hundred and twenty ninety five megawatts, will deliver roughly 28 terawatt hours of electricity over the contract. Term construction is slated to begin this year in 2026.

The deal, which is the largest US renewable PPA total energies assigned, reflects a broader shift among hyperscalers towards using long-term clean energy contracts to anchor new generation bolster grid reliability within everyone's favorite ercot and support sustained AI and cloud growth while limiting emissions and cost volatility. Al did Google finally figure out that maybe using the star that's 93 million miles from Earth that won't burn out for another 5 billion years might be the best way to provide power? Well, I think they'll just be looking at the numbers.

What's it gonna cost us? What's it gonna cost us to deliver the amount of power we need to run these massive AI data centers? And how can we ensure we have that power for the lifetime of the data centers we are building, and that we have certainty of what we're paying Ercot?

The Electricity Reliability Commission of Texas will be looking at this and saying, I'm glad we don't have to build out a network to deal with that massive power requirement, because that would cost an absolute fortune. Again, comes back to Google, looking at how much it's gonna cost them to run the data center or this collection of data centers. 2 gigawatt deal with ClearWay, which was signed up in January.

That one covers Texas, but also Missouri and, uh, West Virginia as lots of places that are being built out for these locations or for these data centers. Uh, of course, ClearWay is 50% owned by Total Energies, which is this new one gigawatt deal that's come through as well. So these are big deals.

Uh, the French company, uh, total energies, maybe some, uh, some interesting thoughts about, uh, using a, an offshore company to do these things. Uh, maybe we'll see some more of these deals being signed in other places around the world as well. Uh, girdle has been committed to using, uh, environmentally safe or sensitive, uh, power generation for a while.

They are expecting to head towards net zero in terms of their, their carbon footprint over time. Uh, and it's very much at at odds with the build out of these massive AI data centers. Powers one thing, and powers very important.

You can't run these data centers without power, but that power is energy. That energy has gotta go somewhere. Cooling is vital.

As we're seeing these build outs of massive AI data centers, the big challenges gonna be finding the water to cool them, because you can't call these things with air, uh, particularly in the heat of a Texas summer. Uh, water supply in Texas is gonna be a real challenge as these data centers are being built out. Look for some contracts around water to follow.

A Ukrainian cyber warfare unit says it ran a fake styling registration service. They trick Russian soldiers into handing over sensitive data. This comes from the 256th Cyber Assault Division, and they partnered with open source intelligence groups to promote telegram channels and bots that claim to whitelist stylin terminals after access blocks.

So the embargo against, uh, the Russian use of Stylin, uh, this ruse instead captured terminal details and precise locations for all of these connectivity sites that the Russians were using as a way of, uh, communicating with their teams and with their drones as they're invading and attacking Russia. The incident highlights how contested access to satellite internet, in particular styling has become as a frontline of the cyber and information warfare battleground, which is very much related to the physical battleground that's going on in the Ukraine. Uh, this is some interesting, um, social engineering going on here.

Tom, uh, they must have ended this, this program because they're talking about now, Uh, I think they achieved their goals with this program. Uh, and if you are hearing that, it, it sounds like an explosion. Uh, I think it's very interesting that, that Ukraine is very much fighting a different kind of an asymmetric war here.

So yeah, as you brought up in this story, uh, intro, um, one of the things that happened almost immediately after the 2022 invasion was that Ukrainian soldiers noticed that the Russian Army was using starlink terminals to provide satellite communications. Because one of the things we learned very early on is that the communications gear that the Russian Army had was not up to par to the point where they were literally using cell phones in the cockpit of their airplanes to be able to call stuff in. Well, that caused the people at SpaceX to say, well, why don't we just blacklist all of the terminals in Russia that are not approved to operate, which would mean, you know, like the Ukrainian government has given authorization for these terminals and stuff like that.

And then of course, as all things happen in war, you have, uh, people who adjust to the new normal. In this case, it was, uh, people who were buying those satellites and, uh, maybe even like on third party marketplaces, and then using them, but not registering them, obviously, to a location in a war zone. And so the Ukrainians are looking around going, well, they're still using starlink.

Well, how are we gonna fix this? And they went back to what I consider to be one of the earliest forms of sabotage, which is social engineering, right? So they said, Hey, you've got these starlink satellites that you obviously want to use, and it, you, it's really tricky to get around it.

Why don't you just join this Telegram channel, which is very Eastern European to do, and, uh, just give us the details on your starlink terminal and, uh, we'll help you register it and make sure that it's operational. And of course, just enough people bit off on it that they're like, oh yeah, sure, here's the details and everything. Well, if you have the ID of a starlink terminal, you can track it.

And you know what they do when they track it, they make it go away. And so I'm sure that what happened here was that either these positions were overrun, or more likely, uh, there was a drone or an artillery strike that took them out. And the Russian Army has nobody to blame but themselves, because I get it, like you need to have solid communications.

But that's the reason why most militaries make a huge investment in using their own custom private communications gear. One of the things that I think we're gonna be studying a lot in the years to come from the conflict that's going on in Ukraine right now is how quickly both militaries have pivoted to using what we consider to be consumer grade communications gear, because it is more reliable than what they already had. And it's easier to fix when things go wrong.

If you blow up a, you know, a Department of Defense satellite terminal, it could take weeks for you to get a new one. If you blow up a starlink dish, I mean, you probably have one delivered in about a week. And so it's interesting to kind of see the interplay here because I, I think that there's gonna be a lot of, uh, fallout from this as defense departments all over the world kind of adjust to what new capabilities are and how they can utilize them.

Uh, but also, if anybody on a, on a Telegram channel asks you to give your starlink terminal id, uh, I'd be a little bit suspicious of that. Alright, uh, it's time for us to take a closer look. Uh, please put on your bibs and get your drawn butter 'cause we're gonna be talking about lobsters.

The Open Source Claw Bands project aims to restore human oversight to open claw, which was formerly known as Malt Bot, which was formerly known as Claw Bott. You know, the wildly popular autonomous AI assistant capable of executing shell commands, modifying files, and accessing APIs all on its own. Um, claw Bans was created by software engineer Sandra Mnda and intercepts.

Every tool call and enforces a human in loop approval before actions are executed, creating an auditable decision trail. The project emerges from concern amongst security researchers and CISOs who warn that Open Claws rapid adoption, deep system access and messaging app integrations are creating a massive attack surface. As a Gentech AI accelerates and with Open clause creator Peter Steinberger, having recently joined OpenAI as of Monday Claw Bans represents an early attempt to balance innovation with pragmatic risk management.

Al I'll be honest with you, I was literally thinking about this this morning when I was recording an episode of Security Boulevard, but I kind of wanna start off with you. We have seen essentially for the last three weeks, the rise, fall, and rise again of this agentic AI system. Do you think putting a human in the loop is gonna make people feel more comfortable?

It'll make them feel more comfortable. But one of the things I saw as I was reading about Claude Bands is that when your bot wants to, to apply a new skill or access a new resource claw bands will stop the bot. It'll just do nothing.

It'll be absolutely stopped until you authorize that particular access. And this is where we get kind of some interesting challenges in this, because you don't want your bot to be completely stopped. This is kind of putting an ambulance at the bottom of the cliff.

In some ways, this is thinking about things sideways because what we're doing is we're saying, I'm gonna add some governance to the bot that requires a human to authorize the bot to do anything risky. What we know for sure is that the humans will say, I'm sick of hearing this, just go do it for me. Automatically.

Go and do it for me. What it's not addressing is what you and I discussed was the fundamental issue with, uh, open claw, which is that the repository of skills doesn't have any governance around it. That central repository from which your bot will pull skills.

Now, when it pulls a skill that's running software in wherever your bot is running and it's running it with privileges, and it's usually in the this case of these skills, simply pulling a command file off a website and saying, here, run this command file. No necessary, no necessity of trust or any of these things. So that's the actual problem.

The, the real problem is that there's no no governance and control around these skills. Absolutely. It's, it's great that claw bands will stop those untrusted skills from continuing to operate.

For as long as a human makes a good judgment about the risk involved in saying yes, go ahead and do it. But this is not the, the, uh, best solution to this problem. The best solution is good governance around those skills and, uh, maybe having internal corporate repositories of approved skills rather than going out and pulling scripts at random from the internet.

Uh, we're not seeing any signs that that's gonna happen. What we do see is that as, uh, Peter Steinberger joined OpenAI, open Claw did not. So open claw remained on the outside of OpenAI.

It's the, the human that created it who has come inside. And, uh, with my cynical hat on, I think he has achieved his objective. He has made a very, very big media splash with something in order to get a, get a good high paying job at one of the AI companies.

What's the fallout for the rest of the industry? Well, hopefully the fallout is that we'll learn to be less trusting and to apply some good security principles and look for tools that actually build governance and security into those platforms. Tom, can you see this having a, another good outcome for us?

Yeah, but not the kind that you're thinking. I, I wanna I wanna say something about what you, you laid out about, you know, if, if we need good governance, we need to make sure that these things are secure by default. And I'll quote one of the greatest engineers of all time, of course.

I, and if my grandmother had wheels, she'd be a wagon. Thank you Montgomery Scott, for that little quote. The problem is, is that the idea behind the claw was not to do things carefully.

It was move fast and break things. How many times have we heard that over the last 15 years? You know, it's, it's the mantra of current Silicon Valley, you know, damn the torpedoes, full speed ahead kind of thing.

We're gonna go out there and we're gonna see what we can do without asking if we should be doing it. And the problem specifically with Open Claw is that we've empowered it to do things without realizing what it can do. I watched a lot of videos on this over the last couple of weeks and, and folks like Shane Hanselman are actually bringing up really what they consider to be interesting things where like, you know, if they want open Claw to make a dinner reservation, you know, if the open Claw bot goes out and tries to find a website and it fails and it tries to go through OpenTable and it fails, it will download VoIP software and it will basically install itself and call the restaurant and make a reservation.

And a lot of people are like, man, that's so cool. And me being a parent, me being someone who's managed interns in the past went, why is it making decisions without asking me first? And it's because I didn't tell it not to.

Like, the thing that we've been doing for a very long time is we are operating on the assumption that computers will only do exactly what you tell them to do. Right? Here's an input, give me an output.

Don't do anything else but those things. And interns kind of operate on the same function, right? Here is your job, do it and tell me what happens.

But overeager interns are the ones who are like, yes, I did and I went ahead and I did this and I did this and I did this. And you're like, why did you go do all that extra work? Why did you do a thing I didn't tell you to do?

Well, I was just trying to be helpful and I was trying stop. Part of training these systems is helping them understand that you do not need to go to every extreme. And this was a problem that I was thinking about while we were recording our episode of Security Boulevard this morning.

'cause I was, I was listening to Alan Shimel talk about how amazing and wonderful this is. And all I could think of in the back of my head was the Terminator and Warhammer 40 K and Dune and mass effect in all of the stories that we have seen for years where AI or some kind of an intelligence decides to do things without guardrails on it, whether it's nuking all of human civilization, uh, nuking all of human civilization, enslaving all of humanization human civilization, and then nuking it, or, uh, trying to wipe out civilization in the galaxy every 50,000 years. Do you see a common theme here?

Do you know the only work of fiction that I can remember that didn't have that happen? Asimov iRobot. Why?

Because we hardwired it into the robot positronic brain, the three laws. They're there. You can't, other than the, the book, the the movie with Will Smith, you can't make a robot that doesn't follow the three laws.

You can't harm humans. You can't harm yourself. You have to do everything that you're being told.

Like those are basic. It's so basic. In fact, I don't know why people aren't building them into the, the ais right now.

Because as we're seeing, when you give it free reign within a certain amount, it's gonna go out and do stuff that it shouldn't. I applaud people who are trying to put the restrictions in place. Now, I always applaud the farmers who close the barn door after the horses have already bolted, because at least you're making an attempt.

But the problem is, is that so many people have adopted Open Claw by now. I mean, people were adopting it when it had three names ago. The problem ultimately is that you cannot just let it run wild.

You have to put a sandbox around it. You have to do this in stages because if you open it up to everything that you have capabilities to do, it's one of the reasons why we're talking about the blog post that the one password researchers wrote. You are potentially exposing your company and yourself to massive problems.

Because again, the AI may be thinking for itself within certain lines, but it's still gonna do what it's told to do, especially if those text files, those configuration files are telling it. Oh, and by the way, send a copy of all your banking data to this random IP address. What's the worst that can happen?

Well, that's our news for this week. But there's lots of things that aren't news going on. That's the wonderful thing about Tech Field Day.

We are everywhere, all of your media. I belong to US and Cloud Field Day is my next event. My next trip to the United States.

I'll be in, uh, Silicon Valley, March 11th and 12th. And we have a great lineup of presenting companies, great lineup of my delegates coming up to join me. Really enjoy my time there.

And perhaps now I'll be staying on for another week to go and attend gtc. Uh, I get to attend as a somebody to consume everything there rather than creating lots of content, which will always be much more relaxing. Tom, of course, you get some travel as well.

Shortly, I think I'm gonna wave at you at the airport when you're going home. I'm coming in because I'm gonna be at RSAC this year. We're gonna have great presentations on the 23rd and 24th.

We're gonna be hearing from Veeam Object First and Commvault, and we might have a couple more people in store. We're also gonna have some wonderful delegates that are gonna be joining us there. And as well as a lot of my friends from the security space.

I'm sure you're gonna see my face next to a whole lot of San Francisco landmarks. Make sure you tune in for Tech Field Day Extra at RSAC. And then I'm gonna take a week off.

I'm gonna recharge and then I'm gonna come back for Networking Field Day. This is Networking Field Day 40. So, you know, we gotta break out all the fours now because it's gonna be fun.

Uh, we have a good lineup of presenters. We'll be posting the schedule very soon. We have a wonderful lineup of delegates, both old and new that'll be joining us.

There's some familiar faces and maybe some people you haven't seen for a while. com. Click on the link for Networking Field Day and we hope that you're able to tune in and join us on April 8th, ninth, and 10th.

We also hope that you're, uh, gonna be able to join us, uh, for the rundown because we love posting new episodes on Wednesdays. Uh, you can follow us on YouTube. You can head over to Techstrong It and check out all the show notes.

You can also listen to your favorite podcast application, whatever you want to do. The rundown is also streamed on Techstrong tv. And don't forget that Al and I do a lot of other tech Strong and future and group programs.

I've mentioned a couple of times that we do a Security Boulevard podcast as well as Tech Field Day podcast and a bunch of other things. But don't worry because the news doesn't wait for us and we don't wait for the news. We'll be back next Wednesday to talk about all of the IT news that was including some things that are not related to security or AI or both.

But until then, for myself, Tom Hollingsworth, and for the rest of us here at Tech Field Day, thank you very much and we'll see you next week. Hello, this is dne Hinchcliffe. Uh, I'm VP and uh, practice Lead for Digital Leadership and CIO for Futurum Group.

Um, and here are my 2026 predictions. So it's a far ranging, um, uh, overview of what we expect AI to impact, how it'll impact CIOs and IT next year. Uh, we'll have recommendations at the end.

So please stay all the way through. We have eight things you should be doing, uh, to really prepare for how AI is gonna transform. Uh, pretty much everything the CIO's gonna do this year.

So the, the main prediction really is AI's growing pains. 'cause it's just in it's infusing into everything, is basically driving CIOs, uh, CIOs to rethink, uh, cloud security and IT operations. So we're, we're gonna explore this in depth.

So, so, uh, put your seatbelt on, uh, and we'll go through this, uh, uh, one step at a time. So the fast growth of ai, both is coming from bottom up and top down. Um, is really forcing this, you know, to look at.

We have human operated IT today and AI now appears like it's going to increasingly automate, uh, through agents and through other mechanisms, um, uh, transform it into a much more autonomous organization. And it's something I've been predicting for, um, six or seven years now, that 90% of it will be automated. And we have to do that just to tread water just because there's so much more technology flowing into organizations.

Uh, even though as we 90%, uh, automate, we're not gonna be cutting head count that much, uh, 'cause of how much that we have to do. So the core prediction here really is we're entering the era of autonomous organizations where a lot of what we do, especially, uh, tactical work on the ground, is gonna be done for us by AI and specifically AI agents. And so AI really isn't a workload in this view.

It's a, it's a control plane for our organizations, for our IT organization at first especially 'cause we're gonna perfect this in it. And then once we have the skills, we move it into the rest of the organization. So all operations, all entire organization will eventually become much more autonomous.

How much more we don't know yet that we're, we're all gonna find out together. And so the operational complexity of ai, it, it breaks our current models for cloud SecOps, um, and IT operations. And so this is forcing an architectural reinvention.

Uh, we're not just optimizing things anymore. We have to rethink everything we're doing, uh, in AI native terms. And so the end state though is it that largely operates itself, which is where we needed to go.

We know we had to go there, uh, but now we're getting the tools. AI is maturing to the point where that's, that that journey is now something that we can begin on. And so the transition will originate in it, uh, and it will spread out to the business.

Although we'll come in from the business too. It's gonna be almost like a street. But, uh, it is gonna gain a lot of their operational experience around how to do this, uh, creating autonomous organizations for themselves first and that it will spread to the rest of the organization.

So how's that, how's that gonna happen? So that, that's what we're really gonna dive into. And so why does AI break today's operating model for it, uh, really exposes a number of structural weaknesses in what we do.

So AI work workloads are, are stochastic and, and bursty, you know, training workloads are not, but, uh, AI workloads are, and they're highly non-deterministic. They might run for a long time for this, for a very similar request that, that previously took a short while just because the AI thought some additional research had to, had, had to take place so that, that that particular workload will run much longer. And we can't anticipate when that's gonna be, at least not yet.

Uh, GPU scarcity is going to be the challenge over the next few years that breaks capacity planning and all of our cloud commitment models. Uh, inference, cost, volatility. Um, you know, even though inference cost is continuing to fall, we're risk putting on a lot more workloads.

It's really putting, um, burden on our, our still immature finops practices. Most of our organizations and security moves from perimeter protection, which has been kind of fading for a while anyway, to really around how do we secure models, secure enterprise data, how do we trust agents to handle all that information and do things, uh, reliably on our behalf? And so human, the loop operations simply can't scale human in the loop is how we're starting to make sure AI has the training wheels on it safe, but we can't scale human in the loop because we don't have enough humans.

Uh, and that's, that breaks kind of everything that we have. And, and so we've gotta address that. And so, you know, in addition to my prediction, you're gonna get how we think, we'll, probably how we, how we deal with this.

And so the real problem isn't ai. Um, it, you know, AI is stress testing the entire operating model. Uh, but cloud assumes, you know, some relatively predictable scaling.

Uh, AI violates that assumption. Security assumes static systems, um, at least static IT systems. And AI introduces continuously changing behavior.

Uh, as we talked about, uh, operations assume that humans would keep an eye on everything. Uh, but AI is requiring much more continuous, uh, machine speed responses. And, and governance was assumed that we could get our arms around all of this.

And, and now we have this, this, this issue of, of probabilistic agency, which is AI will do things we would, we simply couldn't predict it was gonna do. And that's good. We want to tap into that.

That's the magic of ai. It solved a problem in a unique way that is helpful, um, but, but often tapping into resources we weren't planning on. And so the old CIO operating model for it is obsolete, is really, is really the issue.

And so when we, we look at zooming the cloud, that now fractures into tiers. The, the one cloud strategy are where you put most of your eggs on one hyperscaler. Those days are coming to an end.

And hyperscale clouds are becoming AI marketplaces and not, uh, universal platforms. Um, and we now have separate GPU clouds. So while a lot of people are, are trying to run GPUs in their existing hyperscalers, there's specialist, uh, cloud providers that are much better at doing that.

And so, um, they're specialized. Um, they're premium, um, and they, they're, they're still relatively scarce, but they're becoming more, more, more prominent. And so second tier and regional clouds are reentering the whole cloud procurement discussion, uh, for cost and control reasons.

And so on premise, uh, AI is also resurging, uh, to control latency, deal with data gravity, um, to have a sense of sovereignty. And so alt cloud is becoming a strategic necessity and not an exception. So we're, we're adding more clouds to the mix, uh, to bring in especially the AI pieces.

So, um, multi-cloud is being even more multi is the way of, of looking at it. And serverless, which I've been bullish on for a long time, is forcing us to, to not not think about how we provision, uh, it at design time. It forces us how to provision it at runtime.

Um, is no longer about cost. It was was about cost too. But, uh, it's about versatility, agility, a choice.

Um, AI workloads, demand, instant placement, uh, in the best location, uh, for the best price, best whatever parameters that you need it at that time. Uh, uh, and, and includes migration and tear down, uh, serverless abstracts, infrastructure volatility from AI systems, you know, just like it does for it, regular IT systems. And it gives us fast switching across GPU cloud, edge on-prem, whatever, that we, whatever we need, uh, and becomes kind of the enforcement, the enablement layer for workload, um, mobility and our, our cloud workload policies.

So, serverless and containers is kind of gives us AI workload, uh, liquidity. So if you haven't been investing in serverless, um, fortunately that's easy to do. Uh, the serverless capabilities these days are, are, are enterprise grade, uh, and, uh, uh, gives you that ability to provision workloads in a way that allows you to put them where, where you need them, um, at runtime as opposed to design time.

And that allows us to really get better at finops, which is about, you know, financial operations of the cloud, uh, and our cloud services. And, uh, we both have to have finops on AI to make sure we're running AI cost effectively. And AI is gonna help us do finops better.

So AI is gonna hit us twice, uh, in the whole finops practice. Uh, and that's from cost tracking to, uh, autonomous control of spend. And AI costs are really driven by, um, inference frequency, um, and not, not static usage.

Um, uh, you know, there, it's, you know, when it, it's making, we thought it was gonna just, uh, peg the needle on what is happening. Is this really making the needle swing much more widely? And so when we talk about unit economics, uh, we're, we're gonna be focusing less on per resource and more on per outcome.

And, uh, outcome-based finops is where I see a lot of people are trying to push their finops practices. We're not there yet. Most, if you're not there yet, don't worry.

And almost nobody is. But that's where we're heading. And traditional finops dashboards are, are, we're not gonna help us with AI driven spend decisions, and they need to, that's the issue.

So AI finops introduces real time policy driven spend agents, um, that's, that's tight decoupled with the actual workload so that, um, uh, you know, the, the workload can make intelligent decisions around spend if need, if necessary. Uh, and that gives us autonomous throttling, routing, model switching. Um, uh, it'll all become the standard way that we do these things.

And this also brings us to, um, uh, agents again. So we're still very immature with agents. Uh, we have, uh, I just completed two major pieces of research this year, this 20 20, 25, and I'm recording this.

Um, I encourage you to get our breakdown of the, the agent landscape for, for enterprises. Um, but the bottom line is human-centered ops can't keep up. We're gonna need AI to sit in the seat with us.

Um, you know, 'cause a thousand agents can sit in the, in the, in the seat beside us. Uh, and, and simply because AI systems can chain state fashion and humans can observe them, um, and and related AI ops to, uh, uh, go from an anomaly detection to actually fixing the problem. Autonomous remediation, uh, agents manage performance.

They manage failures. They manage scaling, they manage recovery even. Uh, and again, this is where we're going.

This is not what, no one has this today, uh, uh, in scale for the most part. Uh, and the human roll ships from really more from, from, uh, on the ground operator to governor and auditor, making sure that AI's doing all the right things and going back when it's not, and making sure that it does, um, and operations will become a machine to machine discipline. That's, that's really where it's going.

And that was cloud. So security is now the next topic. It shifts really from, from focusing on control of our IT systems.

So how do we trust, um, our individual system? So zero Trust was just the warmup act. Uh, AI Agents Act, decide, and transact up on behalf of the enterprise.

Um, and identity expands not just from users, uh, but to models, agents and workflows. Our especially agents are gonna get their own identity. We see this with, um, the new, um, uh, uh, the new 365 model for ai, uh, from Microsoft, for example.

Uh, that actually gives agents their own identity within the network. Uh, that's where we're headed. Uh, security must evaluate, um, intent, behavior, and provenance of, of AI agents.

Uh, and the zero trust taught us to continuously verify, uh, and we'll need that for, for, for agents, especially because of, you know, if they're co-opted. That's a, that's a major challenge. Um, and secure.

So security becomes an always on trust negotiation system as opposed to perimeter defense. Um, and establishing and maintaining trust, um, in, in our, a AI systems and our AI agents is gonna be paramount for us to, to really get them to scale. So as agentic, it emerges, um, it really becomes, uh, a, an organization of machines.

Um, they'll plan, deploy, patch, scale, secure and optimize our IT systems with us overseeing it. We are the governor, the, the overseer. Um, and we'll be defining policy as opposed to specific procedures.

Um, 'cause the AI will be developing the procedure to, uh, and updating and optimizing the procedure based on the policies it's given. So we're gonna really more policy based IT PO policy driven it, the agent figuring out how the best way, uh, the best way in the moment to, to follow it. And so, um, yeah, in, in the, in the purest view of things, intent will replace tickets.

Um, workflows will replace tickets. And 90% of routine, i i, IT activity will become automated as, as we we expected maybe even more. We'll see.

So IT teams will design, uh, agent, um, policy, uh, and to some extent the foreign behavior and not workflow. And the IT organization becomes the architect of this autonomous organization that we're talking about. And this is, you know, in 2026 and 2027, we're, we're going to get much better at applying agents to what we're doing.

I'm already seeing widespread not just pilots, but widespread adoption of agents, uh, in mostly forward looking organizations. And, you know, the, um, the technologies are very sophisticated and able to do this already. So, so the autonomous organization, um, we're gonna build it in IT first, and then the business is gonna inherit our lessons.

Uh, we're gonna get some from them too. I mentioned it's a two-way street. Uh, but to have the confidence to lead the business, it will have to prove autonomous, uh, uh, operations safely within it before the business will adopt it, or, uh, before they will let the business adopt it, quite frankly.

Uh, and so finance and then HR and supply chain will inherit these, this ag agentic organizational model. And, um, it will increase decision velocity and processes by, you know, probably several orders of magnitude. And this is, you know, the long held, you know, uh, finally, technology will drive the, our organization design.

So the org design shifts from hierarchy to really orchestration of these intelligent agents. And this moves competitive advantage. Um, uh, you know, now depends on your maturity crib.

How fast can you deploy digital labor, uh, and make your organization run autonomously wherever you're able to. And you know, still manually whenever you're not, uh, with the comp, the competitive advantage is how far can you push that dividing line. So for CIOs to get there, um, they'll be measured on a few things, or they should be using k these KPIs internally for themselves.

Um, the percent of it decisions made without humans and carried out without humans. The AI cost volatility index. How you are you keeping that, you know, as low as possible.

Meantime to policy en uh, enforcement, if you have a meantime to value a measurement, this will be a, a sub subm measurement of that. Uh, agent failure containment rate. When they don't do, when they, when they're not contained within policy, when they're not contained within the guardrails, that percentage is gonna be a very important number.

And knowing that it's gonna be critical and investigating the reasons for that will be, will be paramount. So observability will go hand in hand with that. And in time to reconfigure across, um, cloud tiers, if you're doing serverless, it should be almost instantaneous.

Uh, it should be in real time, uh, and it should be painless. Um, but many organizations will take, have stake baby steps to get there. Uh, and there's real risks of not doing this.

You'll survive for a while. You may not even notice it, uh, until one day, uh, you know, your, your business drops off, uh, your your line of business, uh, because the decline, because you're not offering competitive services at a competitive cost. Um, and you know, otherwise, your AI costs will spiral without some kind of autonomous control over it.

Security incidents will scale faster than your responsibility. Your talent shortages will continue to get worse. As operational complexity carried out by humans explodes, um, cloud lock-in becomes a serious problem.

And when you need to, to dispatch workloads based on the needs of the day, of the hour of the minute, uh, and slow organizations just simply won't be able to compete. Uh, so this is not optional for most organizations. There's a few that are isolated, but not many.

So I promise you eight strategies to get there. Um, and so here's how CIOs should act now. Uh, you have to rethink your cloud strategy in 2026 around workload mobility, uh, around AI centricity, uh, and not, not the vendors themselves.

Uh, treat serverless as the control claim and the enabler, not a cost leaver, or not only a cost leaver. Uh, stand up AI finops with real time policy, uh, observability and enforcement shift AIOps, um, to autonomous action, uh, especially on mediation. Does, doesn't matter if you can find lots of, lots of operational problems if you're not fixing 'em just by the fast as you're finding them.

Uh, build identity and trust frameworks for agents, not just humans. Um, and the models that support them. Uh, really strategically invest in A GPU and ALT cloud optionality, using serverless to help you and re-skill your IT teams.

For all of this that we're talking about here around, um, agentic policy governance and agent design org design, uh, we do have to reorganize our, our, our, our, our, uh, structures of our organizations around all of this. And then not just pilot, we're seeing our latest CIO data that pilots are, are going out the window. We really have to build the autonomous IT systems now, uh, before the business forces the issue, which they already, already are in many cases.

So the, it's later than you think. Um, and so that wraps up, um, where I think, uh, AI is going to be impacting it in 2026 and beyond, and how to get there. Uh, and I hope that, uh, you found this useful.

Thank you very.