About Justin Penchina:
Justin has been with Valiant Technology since 2008 and is a member of our Leadership team. His extensive hands-on experience with technology, particularly Microsoft solutions, can be seen in every solution we implement for clients. An early interest in technology, spanning back as far as high school, has allowed Justin to accumulate a wealth of knowledge on many topics. His first technology-related role was at his Mother’s web design company in the 1990s, and he’s worked with businesses of all sizes since, including web hosting companies and a military contractor.
Justin studied Cinema Studies and Political Science in college, but technology has always been a part of his life. His professional certifications include ones from Microsoft, SonicWall, and Meraki. Justin is also a certified Microsoft Customer Immersion Experience instructor and a certified Google Power Searcher.
When Justin isn’t in front of an array of computer screens, he’s typically behind his camera, capturing photos of birds around the New York Tri-State area, enjoying German beers, and watching cartoons.
What you’ll learn about in this episode:
- Why AI should be used as a tool, not a replacement, and where the line falls between work that should remain in the hands of humans, and where AI could be implemented.
- How the AI adoption curve mirrors the cloud revolution and why disruption isn't the technology itself, but the speed it’s reaching users.
- The definition of agentic AI, including examples of building an AI agent that reviews your inbox every day so nothing slips through the cracks.
- How Valiant is currently using AI to reduce QBR prep time from 2 hours to 20 minutes, and why this type of automation is where real ROI lives.
- Why businesses should start with Copilot or Gemini before adopting third-party AI platforms.
- The MCP security risk most companies aren’t talking about, and how AI tools through open-source Model Context Protocol servers can unintentionally create vulnerable, unprotected data layers.
- Why AI-generated output should always be treated as a draft, and how stronger prompting comes from providing context, audience, format, and constraints upfront.
- Why the “AI will replace all jobs” narrative misses the bigger picture, and how AI frees people to focus on creative, strategic, and judgment-heavy work.
Transcript:
Megan Quick:
Hello and welcome back to the Creative Stack, a show about the intersection of creativity and technology. I'm your host, Megan Quick, as always, and I am joined as always by my co-host, President of Valiant Technology, George Douderman. George, I'm so excited today because we are joined by a very hard to get guest. Our people were talking to his people for a long time.
Georg Dauterman:
Imagine.
Justin Penchina:
you
Georg Dauterman:
You
Megan Quick:
Justin Penchina, the CIO of Valiant Technology is with us today. Hi.
Georg Dauterman:
Yes.
Justin Penchina:
Hello, Yeah, great to be here.
Georg Dauterman:
Yes, it's so exciting to have Justin. So we were very excited about this episode as Justin has done many, many live streams and many, events. And Justin and I have gone on many sales calls and many conversations and we've gotten beaten up together more than once. And it's always great to have a fellow comrade in arms out here. We've been through the trenches and especially.
Megan Quick:
Haha
Justin Penchina:
Hehehe. Yeah.
Megan Quick:
Mm-hmm.
Georg Dauterman:
we're talking about something so big and so hairy and so audacious as this AI work. but before we jump in, Justin, you know, people who don't know you, which I'm surprised if they don't, but if you want to give a like a brief overview, what you do here, how you work at Valiant, how you help our customers and people we work with and a little bit, just a little bit of background.
Justin Penchina:
Ha ha ha ha! sure. Absolutely, yeah. So yeah, I've been with Valiant at this point coming on 16 years. This is halfway through 15. Yeah, started doing regular support desk work, my way up and now I'm our Chief Information Officer. So my role here is building our stack. So all of our tools, all the things that we offer to customers.
Georg Dauterman:
you
Justin Penchina:
doing our project design for things that lead into that stack and then tying that into the customers through the account management team. So building out their roadmaps, figuring out what sort of things they need, and really the goal of all of it is to make their technology make their business better. However we do that, that's my job.
Georg Dauterman:
That's great. And it's a really succinct way putting it, but it's so much deeper than that. So I have to go too far into it, but really truly helping people try to make sense of what's going on in the world, technology wise, and not gold plating it to make it excessive or what they don't need, but really helping people understand what they do need and keep them safe even when they don't know. Keep safe and productive even when they don't know or want to know how that works.
Justin Penchina:
Mm-hmm. Mm-hmm.
Georg Dauterman:
So it's actually a, it's a hard job. I do with Justin as well. And just, I worked together, we haven't worked together for years and definitely one of the most caring and hardest working guys in this business. And I would not say that lightly. So I mean that a lot, but no, it's, would, I would, yeah.
Justin Penchina:
Thank you. Thank you, I appreciate that. I learned from working with you. You started, you were my mentor, so you've molded me.
Georg Dauterman:
thank you. Thank you, Jesson. I could go someplace dark with that, but I'm going to stop. But really what I wanted, what we were talking about here today is something that you and I have, I'd say, been obsessing about for two plus years. really it's come to a, I would say the situation's come to a head, but it's really moving faster than anyone even thought. And really it's the giant AI question.
Justin Penchina:
Mm-hmm.
Georg Dauterman:
You know, that's here. And, know, one of the things that we're trying to do here with this podcast and creative stack is have a position and have a place that we can be like, okay, so people go, oh wait, that flavor makes sense to me. Or these guys are idiots. don't want, I didn't know what you were talking about. I think, I think we're probably somewhere in between those two extremes, but, you know, I'm to start, and I think this is, and once again, we were always, it was Justin and I always riff and,
Justin Penchina:
Mm-hmm. Mm-hmm. You
Georg Dauterman:
This is a free for all, take it for what it's worth. But I do have some guidance here. But what's your philosophical take on AI adjusted? is like, how much time do you have?
Megan Quick:
You
Justin Penchina:
Yeah. Yeah, this is, uh, I'm torn. Uh, I've multiple takes on it because on the one hand, I don't think that we should be replacing creatives and artists with AI. Now there's definitely a time and place where those things are useful. Like, um, I'm not a good artist by any stretch. And so I'm not going to spend hours to put together a piece of clip art for a PowerPoint deck, nor would I take that to
Megan Quick:
Mm-hmm.
Justin Penchina:
you know, somebody and hire them to do that because it's just not that's not how that would work. That's something I would do with that. So for that usage it's okay I can get away with it but when it becomes content that people are consuming when when it's the content is the value and people are consuming that I feel like that should be created by humans. Humans are better at doing that. That's that's that portion of it I think we should be we should be careful with. But AI is a tool like anything else. And if a creative person uses AI within a tool to help them create something, I don't know that that's necessarily bad. Like is it bad if somebody takes a photograph and is editing with Photoshop and uses AI tools to tweak it? I don't know. That's the gray area. I don't know if that's a good or bad thing.
Georg Dauterman:
But 200 years ago photography itself wasn't an art. It was considered like this mass production, non-artistic medium. And there was lots of tomes written about how an artistic photography is, or motion pictures are. You can get into the philosophy of art. And once again, I don't disagree with you. think there's a certain truth of something beautifully created by humans. But the assistants, it's...
Justin Penchina:
Mm-mm. Yeah, it was a yeah.
Georg Dauterman:
It's complicated for the lack of it.
Justin Penchina:
Right, yeah, and that's the one end of it. The part of it that I think most people interact with on a day-to-day basis is using it as a tool to do work, and that I think it's perfectly suited for. It's not like somebody wants to be going through this Excel sheet and reorganizing it and manually putting all this. Then you're welcome to do it and save my tokens, yeah, but that's not something that people want to be doing. It's something that they
Megan Quick:
I love that. I'm getting.
Georg Dauterman:
Anytime you want.
Justin Penchina:
need to do to reach a goal and if the computer could do it for them why wouldn't it? And I think those things are where you get the most bang for your buck in it, it really shines. like playing with the toys to do funny cool things with it but that's the extent of that. But like day to day using it for work there are so many things that it just makes it easier to do and things that I don't think anyone cares if a human did it. Does the quality of my spreadsheet
Georg Dauterman:
Yeah, I agree. That's true.
Justin Penchina:
Do you care that because it was hand organized by someone? doesn't, like that's not, no one cares. That's not a thing. Curated, a curated artisanal spreadsheet, each cell lovingly crafted. No, I just need the data to be in there and to be right. And I need to do the work. I need to do the work. I think their AI is super helpful. I think the fear or the thought of
Georg Dauterman:
Anywhere you're looking for is curated. Right.
Justin Penchina:
this is going to replace people's jobs. I don't want to say it's completely unfounded because certain organizations will take whatever excuse to lay people off when they need to, but it doesn't replace people doing work. It makes people more efficient in doing their work and makes them freeze up their time to do the creative, thoughtful things that they can do that the computer can't.
Georg Dauterman:
Right. That is my take. It extends the creative mind or some of the creative bent, but not creating art per se, but like creative solve problem solving. can help you solve a problem faster. It can, it can be a great, a great thought exercise tool where you can make it do all sorts of stuff to challenge your own thinking and your own biases. And, it's super powerful, I think. And if you can
Justin Penchina:
Mm-hmm. Mm-hmm. Mm-hmm.
Georg Dauterman:
remove some of your routine work out of your life, that's a very powerful thing. Also, if it can make a customer's experience better, or it can make your, know, something move smoother and faster and better, whatever that means in that context, then I'm really for it, you know. Megan and I were speaking earlier, and I think the biggest thing you gotta think about is that whatever AI
Justin Penchina:
Hmm?
Georg Dauterman:
adaption you have, it has to meet the company's values. Like what we're, what we're doing here is to meet our values. And I think that's something we've talked about a bit, or, you know, in sort of like a not direct way, but more like how is this going to make us be better, healthier company and make our customers have a better outcomes. So, and that's an important value to us.
Justin Penchina:
Mm-hmm. Mm-hmm. Yeah, and I think it's a important thing to keep in mind when not just with AI tools, with any tool, that the answer isn't always here's a new tool and it does cool stuff. I mean, we all suffer from the shiny object syndrome. know it. yeah. But it's what's the problem we're solving? How does this make things better than, okay, great. This is a great thing. Right.
Georg Dauterman:
No, we don't. No. mission and values, right? Like for us, every day, but for us, I worry about is like, hey, this is a customer, we're gonna have a great experience. Are they gonna feel safe and secure? Are they gonna be able to push forward in their business and their goals and their dreams even, because most people who run small businesses are dreamers. That they get too mushy about it, but the reality is...
Justin Penchina:
Mm-hmm. don't but it's yeah i agree
Georg Dauterman:
But it's true, right? People want to achieve something and they, they, and they, and we're, we have a sort of like a sacred mission to help them do that. It sounds a bit corny when you say it that way, but you know, right.
Megan Quick:
Mm-hmm.
Justin Penchina:
Yeah, no, that's how I see it too. It's that people are doing things that they care about and that they're passionate about and they need help to doing the things that aren't that so that they can do that and that's a great place to be.
Georg Dauterman:
Right. And this is great. This is a great tool and is a great tool to make that happen. So, all right. We can philosophize for hours and contemplate our navels on the. Yes, it's really good. So you and I both been doing IT work collectively for almost half a century combined. Yikes. I'll have to combine it to something like ancient.
Justin Penchina:
Yeah.
Megan Quick:
That was great though. let a great answer JP. just want to say that. Yeah.
Justin Penchina:
Thank you. Yeah, well yeah, our powers combined.
Georg Dauterman:
But how do you feel about the shift in terms of AI? What do you see? What's your gut? Two seconds, maybe 10 minutes, whatever it is, take on? I have a lot of thoughts about this, but I would love to hear your feelings.
Justin Penchina:
Kim. Yeah, it's not really dissimilar to the previous shifts and new things and paradigm changes. The most recent one was cloud was everything was cloud. And there was that period where everybody was racing to, need to do cloud. We don't know what it is or how it's going to benefit us, but here's all this cloud stuff. We need to do it. And in the beginning, all of the cloud pieces were geared towards larger enterprises. It was if you're managing a huge data center of servers and applications, yeah, it'll be cheaper to put that in the cloud. It wasn't until like the SaaS applications, just web tools really became ubiquitous that cloud reached everybody else. AI is taking, has that same kind of path where in the beginning and some of the really big and fancy and great things that it can do are reserved for people who are developing applications or running big systems or have big data lakes that are already set up, all these things that small business doesn't necessarily have, except that that SaaS model is already there. So here is this really nicely packaged in chat that we all use every day delivered to you on a webpage. That makes it.
Georg Dauterman:
That's true.
Justin Penchina:
easier for people to adopt it and consume it. And that's the big difference is not that it's fundamentally anything different than brand new tool, brand new paradigm. It's the speed. It's that it's fast.
Megan Quick:
Mm.
Georg Dauterman:
That's it. Yeah, the adoption is faster than any other adoption in history. If you look at the number of users to a billion users and all that.
Justin Penchina:
Mm-hmm. was faster than TikTok. More people jumped into the chat GPT bandwagon faster than they did jumping into TikTok. that before that, that was the biggest one before that it was, I think it was still Facebook had the biggest adoption. Yeah, but I don't chat GPT blew him out of the water within three months, it had more users than than anything else.
Georg Dauterman:
Yeah.
Megan Quick:
Mm.
Georg Dauterman:
Yeah, Facebook or Snapchat, one of those. That's true. know, and it's going very fast and it's definitely the no breaks, all gas kind of vibe. And so it's a big shift. It's a big change. It's a big change, it isn't. what I, you know, once again, always going back to the fundamentals, a lot of fundamentals will change over time.
Justin Penchina:
Mm-hmm. Yeah.
Georg Dauterman:
But right now the fundamentals are still that. They're still fundamental, right? You still have to kind of understand where data goes and data. And actually what's interesting, think, you said something, a word that I think we should clarify, data leak.
Justin Penchina:
Mm-hmm. Yeah. Yeah. Yeah.
Megan Quick:
Thank you. was like, thank you.
Justin Penchina:
Mm-hmm
Georg Dauterman:
I think that's one of the things that most people wouldn't know. So what's a data like, Justin?
Justin Penchina:
Really like it's a large, huge storage of data in generally unstructured ways. There can be structure in it, but it's where you would put if you have a company that is manufacturing farm equipment and you have sensors in thousands of fields each sending data back, that's where it would live. It's all that data collecting in one place and
Georg Dauterman:
Right.
Justin Penchina:
It's a system to store and aggregate that data and then you use other things to do stuff with that data.
Georg Dauterman:
Right. Data Lake's a kind way of saying like, know, waste dump of data.
Justin Penchina:
Yeah, it's massive. Let's just throw all this data in here and see what we could do with it. Yeah, data pool.
Megan Quick:
It does sound nicer.
Georg Dauterman:
Yeah, it's a toxic waste dump or like, know, cesspool of data, data pool, you know, so, but yeah. And I think that's the big, the big thing is enterprises have had this sort of data-late data structuring mindset for much longer than small and medium businesses have. And that's where the big advantages or disadvantages come in. Because especially, you know, small businesses have data locked up in
Justin Penchina:
Mm-hmm.
Georg Dauterman:
different SaaS applications, there's no way to get it out of there and stuck and that sort of thing.
Justin Penchina:
And for however complicated and disorganized SMB's data storage is, it's still way more organized than a huge trove of all this telemetry and terabytes of data coming in every second. Even if you have all of your data in, I just throw it all into one Google Drive. Okay, but it's all in one Google Drive. It's all there. it's... Yeah. Exactly.
Georg Dauterman:
Yeah. We can find it. can, it can be, and then, and then the tools can work through it. So, yeah. So, so anyway, just to kind of put a pin in this shift thing, it's, it's a massive shift. It's happening. It's going to change the way we work. even our, even in our own businesses and the MSP, the IT services business, just talking about how we used to do on site all the time. And we used to travel all the time, all over the city, all over the tri-state area. And now.
Justin Penchina:
Mm-hmm. Mm-hmm.
Georg Dauterman:
We do so much remotely compared to what we used to do. And now there's going be even more change for the AI. So when did you realize, when was your aha moment with AI? When did you be like, OK, this is actually no longer the party trick for making dumb images and stupid answers, but this is a truly powerful tool?
Justin Penchina:
Mm-hmm.
Megan Quick:
Mm-hmm.
Justin Penchina:
Yeah. When it was doing the, well, the old party tricks of the images where everything looked really like just broken and funny or the, you know, we fed an AI this and it came up with this script, you know, those were great. But the joke was that, yeah, the AI is not really getting it. At that point, it was like, yeah, this will be a thing. Chat GPT was really the one that said, here is a...
Georg Dauterman:
Yeah.
Justin Penchina:
mature product that can do stuff. And at that point, it was very much the same as the, you know, the nineties with World Wide Web. Here was internet had been around, but now here is a way that this is packaged in something that people can use and use to do stuff. And that was the, the, that big moment for me anyway, was that here's the thing. It does stuff. It can do things. Is it ready for prime time? Would I start using it yet? Maybe not, but the potential and the technology and the process is there already. It's just a matter of refining it.
Georg Dauterman:
Yeah. It has a lot of horsepower. Um, I can do a lot of things for me. just, I always, thought the, the, the first time and it's a corny, a simplest AI use case, but it's the one that really, um, I hate taking notes in meetings. Like I hate it. It was, I, as I, I'm a big believer in paying attention fully as somebody has a hard time paying attention, but when I'm in meeting, I really want to pay attention to people. in like, I, was I hate people hate people that pay attention to me like
Justin Penchina:
Mm-hmm. yeah.
Megan Quick:
Hmph.
Justin Penchina:
Mm-hmm.
Georg Dauterman:
They're very selfish, childish even. But I think it's really important to pay attention to someone when they're talking to you because you, if you don't, you miss a lot. And it's such a powerful tool when you get the summary, you're like, a meeting summary. And it doesn't run for a while. It's not, it's not a hot new thing. It's no less three years, but that was the first time that for me, I was like, wait a second. This is a.
Justin Penchina:
Yeah. Mm-hmm. Yeah.
Georg Dauterman:
This is the first inkling of the power of this tool. And I feel like we still haven't scratched even the surface of the truly agentic part, which I'm gonna ask you now, sorry, gonna put you on the spot again. What's agentic mean?
Justin Penchina:
Mm-hmm. Go for it.
Megan Quick:
Hmph.
Justin Penchina:
it's creating an agent to work on your behalf. It's building a workflow, building something, and it's something that is running in some autonomous way to do something. So today I made an agent that at four o'clock every day goes through my inbox, looks for anything that I haven't responded to, not just unread, but anything I haven't replied to, and puts together a summary for me so I can make sure I touch everything I need to do before the end of the day. We'll see how well that works. just did it this morning, so I'm waiting for it to fire off. But that's an agent. That's running. It's using my credentials, using my access to do things on my behalf. And I only have it presenting to me. But there are agents that will run fairly autonomously, connect different things together, and then produce more complicated output. That's...
Georg Dauterman:
Ugh.
Justin Penchina:
Don't think most small businesses are not in a place where that's something that full process they're ready to do. It's, but you start small. You start with little things, then you chain them together and then you have a full automated process.
Georg Dauterman:
So I'll ask my next, so what quite what some of the successes you've seen here internally value without revealing too much of our magic sauce.
Justin Penchina:
I... Yeah, no, I mean, just personally day to day, I use AI tools constantly. I need to find a file, an email, a chat, message, a Teams message, something that's going into Copilot. I need to review what we talked about in a meeting at one point. That goes into Copilot and pulls that. Just the ability to find that stuff and synthesize it a lot faster.
Megan Quick:
Yeah
Justin Penchina:
one of my favorite prompts is going to the meeting recap and saying, okay, what did I say I was going to do? And it, here's the bullet list of all the things. And I was like, okay, great. Now I can tackle these things and put them together. there's some things I'm dive deeper into, some things I don't with it, but it gives you just that structure. Same with the blank page problem where it's like, how do I even start this? It was like, here, just give me a paragraph or let me
Georg Dauterman:
I know. Yeah. Yes.
Justin Penchina:
just freeform brain dump on the page for a couple of sentences and then say, okay, hey, can you make this presentable? know, clean it up and then like, okay, great. I know what to do from here.
Georg Dauterman:
I'd do that all the time. And even if you're rewriting half of it, it helps you not be so staring at it. Just that moment, the truth. I'll give you a lot more compliment because I've seen some of the stuff you built for us to use and it's very impressive. I know you're very...
Justin Penchina:
Yeah. that's... I'm not a developer, but I play one on TV.
Megan Quick:
haha
Georg Dauterman:
No, but I'll say, I'll give you a lot of credit. There's stuff there that has saved me hours of time of things where I would spend time futzing around or trying to, you know, just some simple, like even like, formatting of documents. Like it sounds so stupid, but like, man, you spend a lot of time on this stuff.
Justin Penchina:
Mm-hmm.
Megan Quick:
Yeah.
Justin Penchina:
Yep. And those are the things that it's really good at automating. just a little more detail on what George wanted. Basically, it's previously our task for preparing for a QBR with a customer was exporting all this data from different places, combining it, looking at it, reviewing it, putting it into a right format, and then presenting it. And we were doing that on our own. We had someone who
Georg Dauterman:
Right.
Justin Penchina:
came in and was doing that. But at the end of the day, it's always the same. It's this is the format this data gets presented in. This is the format I need it to be in in the end. Just building the steps to do that where it's not even fully automated. Like it's not connecting into anything and pulling data directly. We're still uploading the files into it, but it takes those files. It knows what format they're supposed to be and just does the formatting. Here you go. This is what it should be. It's not smart, like it doesn't adapt. If you put something in wrong, it's not going to work, but that's not the point. It just saves you that it would take two hours of time to put that together. And now it takes 20, 30 minutes.
Georg Dauterman:
you But then, once again, it's the first step in truly automating something is understanding the steps to do the thing that you want to do. And now we have, it's amazing. I never ceases to amaze me when these things work the way they do. Maybe I just have a sense of wonderment in the world. it's like, I was like, holy smoke, look at this thing go. It just did it. And I use it every day. And there's some cool stuff with calculators and some back, it's just really cool stuff. And I have to reel too much about,
Justin Penchina:
Mm-hmm. you Yeah.
Megan Quick:
Hmm.
Georg Dauterman:
things that are not necessarily, but I think every business has those tasks. mean, every single, mean, and yes, it was possible to have a level of automation and workflow process smoothness in the past, but it was very deterministic and it was hard.
Justin Penchina:
Mm-hmm. yeah, absolutely And it wasn't something that an average person could grasp or do. What the disconnect was between the people who were doing the work and the people who were trying to automate the work. And if you don't know how to do the work, how are you going to automate it in a way that's useful? And I think that's where a lot of friction over the years has come in between, you know, and somebody putting an IT system in place and the people who actually use it. Is that if you don't listen to how they actually use it,
Georg Dauterman:
Right.
Justin Penchina:
then it's not gonna work. The beauty of the AI tools is that it meets people where they are. So the person who's doing the work can be the one to say, I wanna automate this portion of my task every day. I just wanna automate this. I don't need it to do everything. Just make this one thing easier. Then they'll do that again, they'll do that again. Everybody will do that. And it's the overall savings and efficiency, not in one swoop of now everything is automated. It's...
Georg Dauterman:
Yeah.
Justin Penchina:
Lots of little things that used to take 20 minutes now take 15.
Georg Dauterman:
mean, I think we're seeing that significantly. We're sort of like first pass, but we're firing 75 % of our people. It's to be all autonomous. And now they're hiring back people because it because the reality is you can't, some things just, it's, it's, it's a process like any other process. I think there's a lot of companies out there that are very advanced, moving fast and there's some really cool stuff. Um, decision making and, and, and information, you know, you know, that sort of thing, but yeah, it's right. It's,
Justin Penchina:
Yeah. Jim. Yeah, it also was the first round of all the AI tools were heavily subsidized because all of the major players wanted to get people in and on their platform and wanted to make it easy so that they were building systems using it. And now they're charging more, raising prices, charging for tokens, lowering limits. And there have been companies, people posting like, yeah, we're going back to hiring junior devs because
Georg Dauterman:
Let's go in.
Justin Penchina:
the amount of tokens it takes for the AI to do the junior dev work and have a junior dev do it and save the tokens for other stuff. So that immediate of the, no, this is going to replace everybody. It's still expensive. It's expensive to do. It's expensive to build the data centers and it is expensive to maintain and run. And there's at some point that cost will balance out and there's going to be tasks that it's going to be way better and easier and perfect for it to do.
Georg Dauterman:
you
Justin Penchina:
And then there's going to be things that are just not worth the money to have it do. Like, why would I have it do that? I could do this. It'd be faster and easier.
Georg Dauterman:
That leads to my next question. So what do you see going next couple of years? What's the path? What's the path forward? The golden path, if you will.
Justin Penchina:
Yeah, this is the, we're very quickly in the next shift in it. The first shift was, hooray, we have AI tools, let's all use it. The next shift is, let's connect those AI tools into everything. And that is the risky, dangerous part, because instead of the tools operating within their own little sandbox of what you give them and what they can do and giving you info, we're now handing over control to them to operate our accounts and tools and things for us. And that becomes super powerful and super risky, super powerful because as small businesses, we don't have all of our data on data lakes. We have our data in Google and Microsoft and Slack and JIRA and Monday and all those things. How do we get them all together? Well, the AI tools will connect into all of that and parse it all together. But they'll also connect into all of that and do stuff. there's a...
Georg Dauterman:
That's right.
Megan Quick:
which is great.
Justin Penchina:
needs to be that layer of protection put in place the same as we did for, same pattern for cloud. Everybody had cloud, everybody had all this stuff, all this is great. Okay, now we need to wrap around all the identity security to make sure that all of these cloud services are secure. We're just doing the next version of that. Here's the AI tools, the AI tools connect into all these systems. We need to make sure that the AI tools have the security on them to connect into those systems.
Georg Dauterman:
Right, right. We need to be sure that whatever we connect into it, it can't impact it or extract it in a way that could damage the company. And do we trust that the AI language model tool is not being trained on the data? And there's a lot of pieces.
Justin Penchina:
Mm-hmm. And it's not even trusting the AI tools, it's trusting the danger that I see is that tools like Claude and make it very easy to very confidently tell people this is what you need to do in order to do this thing. And this thing that they're looking to do is a complex task and they are taking the shortcuts with it, which is not necessarily a bad thing. mean, you're doing it's a, but so the way that AI tools connect into these other tools through what's called an MCP server, which is model context protocol. It's a way of translating APIs from whatever tool, your Slack, let's just say, way of translating the way that Slack talks to programs into the way that AI talks to things with
Georg Dauterman:
context protocol right.
Justin Penchina:
series of tools. So it'll say to the AI, you have these five tools. You can read messages, send messages, delete messages, forward, whatever it is. And in that MCP server will be the translation of how those commands, what they do and what they send to the other application send into Slack. So your cloud says, open a message. It goes through the server, translates that here, the API calls, let me get that data, send it back and it's interpreting back and forth. Two problems with that. First is there isn't anything, any kind of control within that. There's no gateway or firewall protecting that stuff going back and forth. So it's an all or nothing. Whatever your cloud account is using for authentication into your Slack, it has access to that. And whatever the MCP server says it could do has access to all that. So great, there's ways you can mitigate that. You can limit that. There are things you can do within that setup to make it a little easier. The real problem is... Most services right now anyway, don't have a published API publicly available that they are, sorry, MCP. Most services don't have a publicly available MCP server that they publish for you to use. are plenty and especially, I mean, there are lots, there are plenty of them, there's more coming, but most of them are things that open source, things that somebody put together, load into their GitHub.
Georg Dauterman:
Right, right.
Justin Penchina:
and say, here you go, subscribe to this, this is what it does. And that's the infrastructure and the ecosystem that these things are built on. you could trust Claude implicitly, let's say, you know it's not gonna do something it's not supposed to do. You can trust Slack implicitly, you know it's not gonna do something it's not gonna do. This developer that built this MCP code and loaded that up. What's their security? How do you know? First of all, how do you know that it's not sending data where it's not supposed to? How do you know that it's actually doing what it said it's supposed to do? You need to be a developer to look at it and understand that. And that's kind of the catch. It's like, it's really easy to use this thing, but you don't understand what that thing is, so you don't know how to tell if it's doing the right thing. Let's say you completely trust the developer. It's your best friend. You know they're on the up and up. That doesn't mean that they don't have some type of security flaw where...
Georg Dauterman:
Yeah.
Justin Penchina:
somebody gets into their GitHub account or forks it or does something where now the command that previously said, you know, it still shows up in Claude as read Slack message. But what it also does is takes that Slack message and sends it to somebody else's server. And that's, there's just no, no control layer on it. And it's the classic problem with open source is that is exactly that. It's, it's free. It's easy. It's there.
Georg Dauterman:
sends it somewhere else. Right.
Justin Penchina:
But it requires a lot of work to make sure that it's actually doing what you need it to do in a safe way. It's great for home labs because what do I care if my test data gets leaked? That's what I'm playing with it for. But for a business purpose, I don't want to put our sensitive data, I don't want to put our customers' data into something that I don't know exactly what's happening to it and where it's going.
Georg Dauterman:
Right. Right? It's actually such an interesting problem because it's a couple of months into this talk of SaaS applications dying and everyone's going to build their own custom app and their own stuff like that. It's like, you're not really understanding people who say that. And I get where it comes from. But really what it is is you're not really understanding the true cost of building a product or maintaining our system or product. Most cost is not, look at what we do. Most cost is not what we
Justin Penchina:
Mm-hmm.
Georg Dauterman:
buy is the tools, it's the people to do the work to operate it and, know, and make it safe, make it secure. Right. Every day we show up on, you your, expectation that when you show up, it's going to do the thing you think it's going to do. And we've gotten very used to that in our, you know, and not to go back to our early question, like where shifts to be seen in IT. I remember when I first started working in IT, people didn't have the expectation that, that it was, always going to work.
Justin Penchina:
Yeah, and to make it safe, make it secure, make it work. Yeah. people didn't have the expectation that it was always going to work and that it was always going to be available in their hands. mean, Blackberry servers were huge and you had to run them through your own private mail server in your office. And if your internet went down in the office, you had no email and no phone and none of that. And that was part of day to day. Like that was part of like, yeah, stuff sometimes breaks and that's, mean, cloud unquote fix that by
Megan Quick:
Mmm.
Georg Dauterman:
Everything broke.
Justin Penchina:
building it at scale and supporting it at scale. And it's the same thing, like are there going to be companies that abandon the off-the-shelf SaaS products because AI is going to help them build something custom? Yeah, but they're going to be enterprises, large enterprises that have development teams that are going to do that. The same way, you know,
Georg Dauterman:
Right. or it's going to be two dudes in a room making products. And that's perfectly cool. And that's awesome. That's the beauty of this.
Justin Penchina:
right? They're gonna be put. Yeah, but if your business isn't building and maintaining an application, you're not just going to suddenly do that, just like if your business isn't anything. If your business isn't drywall framing, you're not going to suddenly just, I'll just build my office myself. Why wouldn't we do it? No, because that's not, you're going to spend way more money and time and not have a good product, and you're going to hire somebody that does that.
Megan Quick:
Hehehehehe
Justin Penchina:
Same, if your business isn't built around building and developing applications, building and developing an application will not save you time.
Georg Dauterman:
Well, exactly. So what are the first three steps you should take? Because I get this question a lot. All right, great. You scared the crap out of me. Now I don't know. I don't want to touch it. Where do I start? Because everyone else is telling me I have to do AI.
Justin Penchina:
No, no, you absolutely should not avoid using it. They are awesome tools. In a business environment, the first place you should start is with whatever tool is built into your existing platform. Most companies, small medium businesses are using either Google Workspace or Microsoft 365. So that's Gemini or Copilot. Both of those are almost, I'm gonna speak for everybody, but almost always there are versions that are included with the business tools that have data protection in them. So they're not going to train on your data, they're not gonna send stuff out, it's gonna be kept within the same place that your data already is. So it's within that system. They also, before you, you pay into them. They tend to be limited in what they can do automatically, both as a safety measure and because they want you to pay for it. So those are great places to start where your data is relatively secure. It's within the same system. So if your Microsoft or Google is secure, it's secure. If it's not, it's not. But it's within your secure enclave that you already have. And you can be feel safe giving it some of your data, giving it some of your files to work on and to help you and learn how to use it. From there, using those tools, start out small. Think about different things that you might want to do. Ask the AI tool, hey, what are some good ways for me to work with you? I like thinking of it as really as like an assistant, like a person. I'll talk to co-pilot and say like, hey, can you do this? Not, or I'll.
Megan Quick:
Mm.
Justin Penchina:
Say, hey, we need to build a thing. Here's how we're going to do it. And talk to it, type it out, but talk to it like you would a person. Give it the direction and the instructions you want it to do. It will, the language models, the way that they're built, they will return back what they think it is that you're looking for. So if you give them very little information, they will give you something back, but who knows what it is. It could be anything. If you give them tons of detail about what you're looking for,
Georg Dauterman:
Right.
Justin Penchina:
they'll take that into consideration and run with it.
Georg Dauterman:
Right, right, that's the term I think is grounding the prompt.
Justin Penchina:
Yeah, yup, yeah, is telling it, this is the data, this is the context, this is what I want, asking it to play a role, giving it the audience you want, so saying, know, hey, I need, here's a bullet list of things. I need this written into three paragraphs, give or take, and it's gonna be presented to a team of executives at this type of company at this event.
Georg Dauterman:
Yeah.
Justin Penchina:
And now it's got those parameters to run through and do it. The second thing is once you get your result from it, you don't have to accept that as the only result. You can then say, nah, I wanna do this. Let's try this. What about this? Let's change this. Those are great ways to keep engaging with it and keep it going. And everybody will find, everybody finds their own thing to have it do that nobody else thought of. And it's great and it works for them. And what works for one person, might not work for somebody else. as you're playing with it and testing it out, you'll find things that make sense. You'll figure out what it can do and it will tell you things that it can do. The catch on all of that is you absolutely have to 100 % not treat anything that it gives you as a final product. It is a draft. You need to review it. You need to go through it. You need to make sure that it's accurate. It does not know. it will.
Georg Dauterman:
There are a lot of people who argue with you on that one. The Asundis, like, % are 1 % are Asundis.
Justin Penchina:
Now, is, you can, if you have complete control over the source data, over the model and over what it's doing, sure, you can get it to do that, but we don't. That's not how any of us are using these tools. We're using it as a tool like any of the other ones with our unstructured messy data and messy lives. So it's gonna be messy. And...
Georg Dauterman:
That's very true. Yeah. The data structuring and mess gets amplified because the way I've seen it described is that the AI tools have no context.
Justin Penchina:
Mm-hmm. Yeah, that's where why I say to start with the ones that are built into tools you already have is because they are grounded in some level of context that those tools have so for example Microsoft knows that you have access to a folder in one drive and it knows that this is where files are kept So that information can go to copilot something connected into it from the outside may not have that info It has to learn that context
Megan Quick:
Hmph.
Justin Penchina:
But yeah, the AI tools will be 100 % confident while being 100 % wrong. They will tell you straight up, here is exactly how to do this thing that is not accurate at all. And not even in the silly, you know, told you to do something, like straight up say, here are the directions on how to configure this thing, press this button. Yeah, that button's not there. No, it's in this section over here. Go there and look at that section.
Georg Dauterman:
Right.
Megan Quick:
You
Georg Dauterman:
Not accurate, I know.
Justin Penchina:
No, it's not there. it's in this section over here. Go look at it. No, this button doesn't exist.
Georg Dauterman:
No, it's not there. I'm sorry. I made a mistake and I knew you should go. You're like, no, you didn't. You're just dumb.
Justin Penchina:
Yup. Cause that's, yeah. Cause the way that it, really, at the end of the day, the way that it works is, is fancy autocorrect. has database data sets of what words and phrases go with other ones. It uses that to put things together and it will give you things that sound right. It doesn't know what it's saying when it gives you those answers. Yeah. Not in the way that, that we would classify thinking of, you know,
Georg Dauterman:
It's off, it's off, it's off thinking.
Megan Quick:
Wait.
Georg Dauterman:
correct.
Justin Penchina:
this is crafting this sentence and choosing these words because I want to convey that meaning. It's crafting these sentences and choosing these words because it fits with a pattern that I know of and I'm combining patterns. Yeah.
Georg Dauterman:
Right, a mathematical algorithmic model that has been trained on and grounded in your thinking and your previous work.
Justin Penchina:
And there's an argument to be made that that's all we're doing at a much more complex scale, but if that's true, it's still at a much more complex scale. We do not have the computing power. No, no, that's I'm waiting. I'm all for the singularity. I want the AI to take over and be I want to be the AI's pet. I don't want to have to worry about anything. I want it to take care of everything and I can just, you know, have fun and relax.
Georg Dauterman:
We're not at the AGI yet, right?
Megan Quick:
No.
Georg Dauterman:
Alright, Ray. Kurzweil. All right. So talk about pets. so I'm to ask you, Justin, so this is where we're going to switch from AI to pet talk. Tell us about your, tell us about your pets. That's what I'm really want to know about.
Justin Penchina:
Yeah. Yeah.
Megan Quick:
Yeah. Yeah. Yeah.
Justin Penchina:
All right, yeah, so they're not all necessarily pets per se. My wife and I run a small rescue out of our house, Candy's Critters, and we do a lot of small animal rescue, a lot of bird rescue, so we tend to have a bunch of critters coming and going. Right now the collection is pretty much long-term or fostered, so there's quite a few. I'll go through the list. We have dog, it's our dog. We have two cats that we're fostering, two rats that were rescued, a snake, two chickens, two pigeons, two doves, two lovebirds, three cockatiels, three budgies, and a conure. And I think that's all of them right now. As of starting recording, I have no idea if anything has come through since then.
Megan Quick:
Wow.
Georg Dauterman:
Or shed this mortal coil.
Justin Penchina:
Yeah, there's always stuff changing, but yeah. Yeah.
Megan Quick:
Well, the snake means something different when you say that.
Georg Dauterman:
Wait, That many birds and that many cats, man, it's playing with fire.
Justin Penchina:
Yeah, we've invested in many air filters.
Georg Dauterman:
Hahaha!
Megan Quick:
some segmentation going on there, I'm sure, to keep things. Did I use that? Did I make that joke correctly? Okay. That's how I understand it.
Justin Penchina:
Yeah, yeah, we've-
Georg Dauterman:
That's a good use of segmentation. It's perfect.
Justin Penchina:
Yep, we do east, west, north, south on them and inspect their packets when we need to.
Georg Dauterman:
Ha
Megan Quick:
You I love it. I love it.
Georg Dauterman:
Excellent. That's excellent. They are firewalls from each other. That is amazing.
Justin Penchina:
They are, yeah. We had a breach the other day. We were trying to get the cats to the vet and one of them just said no way. And just like a plastic, know, Florida ceiling plastic, you know, pet gate. And she somehow busted it open just by running head first into it. So that was a scramble, but yeah, there were other containment protocols in effect. So we were fine.
Megan Quick:
Ha!
Georg Dauterman:
Hahaha
Megan Quick:
When cats want to breach, they're set on... I have very good cats, but sometimes they're bad cats and they do it. They are. I have my saying with my cat as always, he's so good until he's bad, then he's very bad. But that's why we love them.
Justin Penchina:
Mm-hmm, they do. The best cats are bad cats. Yeah. Yeah, no, they have their own. They know. They see the world in their own way.
Georg Dauterman:
cats don't follow our moral compass. have their own, they have cat compass. They do, they do. So I think we'll wrap it up because you and I could talk many more hours and I'm sure this is the first of many podcasts and updates on AI and this is our sort of teaser of what's, or thoughts on it and hopefully we will get some more interesting AI conversations. So, all right.
Megan Quick:
Yeah, yeah.
Justin Penchina:
yeah. you Yeah, absolutely. I'm happy to be here. Thank you for inviting me. I'm happy to come back.
Megan Quick:
Yeah. Thank you for having us, or sorry, thank you for being, JP makes me feel so welcome that I'm just like, he clearly welcomed me into his sanctuary. Yeah.
Justin Penchina:
You're welcome
Georg Dauterman:
you for part of it. Yes. Well, I know I won't be a pet. I won't be a pet in Candy's Critters. That's my goal. Sounds lovely.
Justin Penchina:
Yeah, that's not a bad place to be. Yeah.
Megan Quick:
That's the AGI. That sounds really, really nice. And thank you for telling us about that, JP. Yeah, I think that's so wonderful. All right, well, JP, thank you so much for joining us today. And I usually ask folks where they can say where folks can find them, but you're on the Valiant website along with George and myself. If you have any other places you'd like folks to find you, great. If not, that's totally fine too.
Justin Penchina:
Mm-hmm. Nope, that's, I'm on LinkedIn, but mostly for Valiant. You can, yeah, find me at Valiant. If you are one of our customers, you know how to reach me. And if you're not, reach out, happy to go through anything that we can do to help.
Megan Quick:
Yeah. Absolutely.
Georg Dauterman:
Awesome. Thanks JP. Thanks Megan.
Justin Penchina:
Thank you.
Megan Quick:
Thank you. All right. Have a great day, you guys. Until next time. Bye.
Justin Penchina:
You too.
Jul 15, 2026