Your Product Isn’t Ready for AI Until You Fix These Software Mistakes

Garrett Fritz: It's a frustrating truth of the universe. Unfortunately, when it comes to a work product, you just can't assume that anyone's going to either read your manual, they're not going to read even the title of your application, they're not going to literally read anything unless they are a captive audience in some way, shape or form.

Aaron Marchbanks: Welcome to SaaS App Building B2B Web Applications, the podcast where we share real world stories, practical advice and tech insights for those building or thinking about starting a tech enabled business. I am your co host, Aaron Marchbanks

Justin Edwards: and I'm Justin Edwards. Each week we bring you the stories,

Garrett Fritz: strategies and insights you need to build

Justin Edwards: your SaaS or tech enabled business smarter, not harder. Let's dive right in.

Aaron Marchbanks: Hello and welcome back to SaaS that app. Hope everybody's been having a good time this spring and enjoying finally some decent weather with me. As always, my good friend, partner in crime, Justin Edwards. How's it going with Justin? Doing pretty well.

Justin Edwards: I got a new shirt for you.

Aaron Marchbanks: Reach for the stars. Uh, nice.

Justin Edwards: It's like a little nerdy and a little fatalistic because they've got the meteor here.

Aaron Marchbanks: Well, it's somewhat appropriate too, because in Garrett, we do have some aerospace background that we may chat a little bit about.

Justin Edwards: I, uh, totally planned that.

Aaron Marchbanks: Perfect.

Justin Edwards: Okay.

Aaron Marchbanks: I knew that you would. And speaking of Garrett, with us today is Garrett Fritz. He is partner and CTO at Meta cto, where he helps companies transform their digital products through through practical mobile and AI solutions. He has spent his career in a CTO role a lot, helping teams make smarter architecture decisions, deliver more with less, and build products that are reliable in the real world. He is also known for his nobody reads philosophy, which I have questions about. But it's a rather pragmatic approach to building software that explains itself, reduces friction, helps teams avoid the kinds of complexity that slow products down. In general, being a good steward with the client. So, Garrett, welcome to the show.

Garrett Fritz: Thanks, Aaron. Happy to be here.

Aaron Marchbanks: So, something that I did not get to ask you in our briefing call and I actually really wanted to because we were in the throes of watching Artemis heading to the moon. I know that you got background and some experience in that with aerospace, aeronautical engineering from mit. Were you very excited or were you just like, oh my gosh, I've had enough with the moon?

Garrett Fritz: Absolutely thrilled. You never grow out of that. And as you mentioned, I do have that degree. I don't use it a heck of a lot anymore, except to bring it up as early and often as I can. But uh, I did actually exercise that degree for five years or so after college. Built mostly in the aero sector. I was a prime contractor for the DoD building Helicopter simulators and stuff like that. It's a lot of fun. My wife who I met at uh, MIT is still very much in it. She does a lot of satellite work, GPS systems and so I get a little bit of a uh, contact high on a day to day basis. But in my actual working life now it's a lot more grounded tech focused. So it's always nice to see the big events.

Justin Edwards: I always have a lot of admiration for anybody who had to do more math than I did for their degree. Not a whole lot of degrees require more but uh, those intense engineering fields are among them.

Garrett Fritz: A lot of math. A lot of math.

Aaron Marchbanks: So speaking of math and now in software, how'd you make that jump? What was that progress like going from that into tech in general and software?

Garrett Fritz: Yeah, I mean the decision was very much career focused and came down to those practical things of my wife didn't want to live in Colorado so I left that job that I was at in Denver and it was just the first one that I found in Los Angeles which happened to be Android, uh, developer role for a uh, startup that was building social networks for big brands like Major League Baseball, Liverpool fc, stuff like that. So it was a pretty personal decision to just jump ship but like can't be too upset about that. I mean it ended up being one of the more marketable skills going into software dev work. At least it was 15, 20 years ago.

Justin Edwards: Yeah, sometimes having those really clear bright line requirements make things easy. Like well we're not staying in Colorado.

Garrett Fritz: Yeah.

Justin Edwards: We've eliminate one option outside of it. Yeah.

Garrett Fritz: And so I mean it's great. I'm an engineer, I'm a problem solver so it doesn't really matter if we're building simulators or if we're building software solutions or if we're building a model rocket. It's all the same thing. We're trying to come up with a solution to a problem, trying to engineer something, be creative to whatever extent possible. And so anything that uh, allows me to do that and in my current space allows me to do that on dialed up to 11. I'm pretty happy it exists that way.

Aaron Marchbanks: Gotcha. Very cool. So I'm imagining that life in aerospace and to Justin's point, lots of math but certainly working for the DoD there's a pretty high bar for quality and precision and sometimes, well often that over speed and it's kind of been my experience in software. It's almost the reverse of that where speed often wins over the degree of quality or perfection that you're looking for. Have you found that to be true or was this like just barely a step over in this direction and you just carried everything with you?

Garrett Fritz: Yeah, it's wild. It's definitely the traditional prototypical triangle of things. You have the uh, quality, speed and cost. And you can pick two right, for things that you can actually optimize. Definitely as a government contractor you have a lot more of that regulation red tape. Certain processes you gotta follow but that also comes with the cost associated with that. And when we're dealing with fast tech, you usually, especially in the startups don't have so much of that pricing flexibility. And so it's really quality and speed that it gets enhanced. And so really it just kind of depends on what that, that appetite is. And two different companies, even if they're both tech startups, will value those differently. You have one providing financial solutions, there is no question you still need that quality and the assurance and capabilities whereas having, if you're just building a dating Apple, who the hell cares?

Justin Edwards: Just whip the thing out as fast

Garrett Fritz: as possible and then get it through. And so uh, for us it's really a matter of like, okay, well how do we get a team in place that could support either one and just set your dials and kind of operate in that space. But you know, there's a little bit of whiplash sometimes. Like this worked totally fine for these other people. Why are you upset now that we don't necessarily have documentation when you told us you don't want to pay for documentation, stuff like that. Whereas that sort of thing is all pre negotiated and upfront. When you're dealing with the government it's a little more loose what the private contracts.

Aaron Marchbanks: Gotcha. So you have been CTO or you are acting as CTO still and across many years now in many different organizations. I'm just kind of curious, what is kind of your approach when you're either stepping into the CTO role or you are helping people who need fractional CTO or who need your services. First time dealing with the product, dealing with the team. What is your approach there?

Garrett Fritz: Yeah, ah, you mentioned my first principle of software development already, which is nobody reads my second principle of software development is the previous guy wasn't an idiot. So which is tends to be a guiding factor when any developer or any expert comes into the reins. It's just assuming that everything before was wrong. And you got to start over. We try not to do that to the greatest extent possible. Like no matter what, if you come to us with nothing, or we're taking over from another org, or like there was a previous person in place that was the cto, or if this is just a brand new idea. There's a lot of thinking that went into whatever, however you got to, wherever you are at. And my goal at the very beginning, when stepping in, is to understand as much of that as humanly possible. Take the time to really appreciate the think work that went into that. Where the bodies are buried and how they got there in the first place and what decisions put them in the ground. And then after understanding where you've been, I can help you get to where you want to go. And so it's really easy to just say, hey, forget all that and just let me build you something brand new. But the harder job, but the more probability to succeed is to not stumble over those same rocks that got you to where you're at. So that's my answer for how I get kind of started. There's a lot of questions and not so much suggestions on the beginning is how I try to, um, keep that focus.

Justin Edwards: I absolutely love that. I think that understanding is the basis of all good recommendations. You have to actually understand the situation, what's going on. Do you have an example or a fun story for something that was very peculiar that you later found out was important for a reason, probably.

Garrett Fritz: But, uh, a lot of those are more sad stories. Before I had that principle in place, I didn't get this philosophy of the previous person. Wasn't a knucklehead just because I had this great idea. It's because I, uh, occasionally was a, uh, more abrupt and super confident young whippersnapper. I want to say I can do this absolutely better in a day. And it's not until you start turning systems off that you realize how the criticality of those. And so there may have been one or two scenarios in my early 20s where I may have deployed a brand new greenfield version of a platform and heard from some angry customers who said, hey, what about my ability to upload the CSV? And I said, well, why are you uploading CSVs? The interface is so smooth. Just use that instead of. It doesn't work with 10,000 different records that are all malformatted and have their own fuzzy mapping. So, yeah, not so much anymore, unfortunately, because of the new philosophy. But that was a lesson learned.

Justin Edwards: I like approaching that with the last guy. Wasn't an idiot. And even if what's right for the business now isn't what was right for the business then, that guy was working with different constraints or that guy was working with different requirements and different stakeholders. And so coming in and just assuming that they were incompetent is not probably the best place to start. And especially when you rob yourself of the ability to take all their hard won learnings and all their hard won lessons forward, even if you are going to rewrite, it's like to understand first. So that's absolutely fantastic. Real quick, this episode is brought to you by Delta Systems, which is what Aaron and I do when we're not talking through microphones to you, the people of the Internet. We've got a really, really great software team here and we love to work with cool people on cool projects. So if that sounds like you and you've got a problem or you're in some kind of a jam, go to deltasystems.com grab a time with us, we can beat up on your problem together and if there's a fit there, amazing, we'll help you out. So Deltasystems.com, grab an appointment and hey, maybe we can work together.

Aaron Marchbanks: So that was kind of the birth of your second principle. Talk to me a little bit about the first one, the nobody reads thing. That's very interesting to me.

Garrett Fritz: It's a frustrating truth of the universe unfortunately. And from a broad sense, what I mean is, of course people read books and novels and sci fi and all that kind of stuff, but when it comes to a work product, you just can't assume that anyone's going to either read your manual, they're not going to read even the title of your application, they're not going to literally read anything unless they are a captive audience in some way, shape or form. And even beyond just your users, the people who are working with you and developing your product also, you can't assume that they're going to read the specifications or the requirements or the use cases that you want to handle. And so how do you get over that? You can't just stop writing. You still need to have the kind of, that cover of the base of what it is we want to build, where we want to go. But what's core is getting uh, understanding in some other way? Is it by intuitive interfaces, is it by a, uh, conversation with whoever your audience is? Is it some sort of viral video that you create just to spit out and say this is how you use my product. There's just needs to be some other transparent way to get across. And I would say that's never been more true than in this now age of AI slop, where text is just very cheap to create these days. And I'm sure you're no stranger to this. It is easy to spot when something is generated by AI. And not only do you not read that, but you don't even focus in on the text. It's just fuzzy and blurred and through and you just kind of scroll all the way to the bottom and okay, well, there's no information here. And I feel like just even outside of our space of software development, I think the layman regular person is also becoming numb to that sort of block of text experience. Right? And so we gotta find better and more interesting ways to, you know, solve these problems visually and creatively that don't make us blend into the AI's CR now a little bit even worse, because not only are people not reading, but it's starting to become very easy to detect, like AI generated design. And so now every little bit of software that you're seeing out there is starting to become this amorphous blob of AI where you've got these rounded corners, you got the colored left bar edges, and all the menus kind of look the same. And so standing out of that world is now difficult because, well, what do you want to do? Do you really want to pay cost design to really be aesthetically pleasing and jump out, or are we going to find some other clever way with maybe UX to get things across? So it's just easy to get blended in when you're relying on what traditionally you should be able to rely on, which is being able to communicate through written specs. It's not the world we live in.

Aaron Marchbanks: Does that extend into documentation on code or repos or things like that for developer sake, or is it mainly a customer focused kind of philosophy?

Garrett Fritz: Mostly customer focused. However, it's interesting when you're talking about like a developer kind of space and specs in the sense that nobody reads, yeah, that's true for humans, but AI sure as hell does. And so now it's actually okay that your specs and your comments and your documentation, when it comes to technical documentation, is an ocean of text. In fact, it's almost preferred as long as it's relatively accurate. And so it's almost leaning the other way when it comes to code documentation and product specification. If you're leveraging these AI tools to get across them, sure, you're no stranger to that and neither is your audience, but that is effective now. It's never going to move the needle with your end users unless you're starting to also be on the very cutting edge of your product, which is starting to market and provide your products and services to other agents. So that is not, uh, totally unheard of. I mean AI LLM, um, search optimization is a thing, right? We use it on our site, we generate content specifically to get scraped by ChatGPT searches and things like that. In that case, yeah, oceans of text are very much read, but not the humans that you eventually want to capture with it. So we kind of do a little bit of both. We have sort of the hidden mechanisms of text to get caught as the experts in the field and get that generative optimization, but then be flashy and stand out for the eyeballs that invariably then follow that lead.

Aaron Marchbanks: Yeah, we just got back from microconf and that was a topic that came up. Obviously AI in general was just ever present, but specifically where it's going, directionally speaking and probably in the near term is that whole agent to agent thing because everybody in the room is worried that they're about to get replaced by a very, very cheap version that somebody throws together in a vibe code over a weekend versus just extending what they actually already have, which may have decent market share and provide a good service and treating another AI or another agent as another customer for their services to be able to continue to leverage it in interesting ways.

Garrett Fritz: So yeah, if ah, you're worried that the Internet wasn't dead already, it's now going to be completely agents Internet facing with agents.

Aaron Marchbanks: It's interesting, I mean the statistics on how quickly the corner is turning even just as a search tool. How many people are moving away from the traditional search stuff and just exclusively using whatever flavor of AI and LLM that they're tinkering with as the search tool.

Garrett Fritz: Absolutely. And that was something that I was reminiscing with my business partner. We were in Vegas last week for the Google Next conference. Same thing. It was just agents up and down the board. But we were discussing like from the big coming to market moment of ChatGPT3, four years ago, right. It was, oh my God, Google's dead. They're never going to survive this. Right. No one's ever going to search. But then at the same time, those AI search results absolutely sabotaged us during that trip when we tried to find a vegan restaurant for us to make a reservation for. It sure, go to this place. They even helped us get the number to make the reservation. We, we called it up. We made it. And then the next day they called to confirm the reservation and we say, oh by the way, we're vegan. I'm not, my business partner is. And they said, oh well, we don't have any vegan food and we can't serve you. And oh by the way, there's a cancellation fee of $200 each. Oops. Oh my God, are you serious? And unfortunately the excuse that we gave them of, well, this AI bot said you guys do serve vegan, didn't actually fly.

Aaron Marchbanks: I'll bet not.

Garrett Fritz: It's a valid excuse. So in that sense, I think traditional search on Google and the uh, Internet still has a place, especially when it comes to confidence. I mean how many times have you searched on Google and you've just had to scroll past the Gemini results at the top because you just either knew it was already wrong from the get go, there's no way it would be correct or when you looked at it, yes, this is actually something I know about and this is also incorrect. So all these hacks of how do I remove the AI from the search minus AI or whatever. So I'm sure there is a practical future in the very near future where that becomes more reliable, but still they ain't figured that bit out yet. It's a fun pain point that we recently experienced. So thanks Google. You know me for.

Justin Edwards: Yeah, haven't had AI Libre quite so badly. I had Claude dump one of my development databases recently. But 200 bucks a pop for a place that doesn't have vegan fruit is that sounds awful.

Aaron Marchbanks: That's true.

Garrett Fritz: Yeah, I mean I prefer enough my dev database. Although I saw this morning there's another article of it dumping a production database for some poor lowly developers trying to use it. You see that all the time.

Aaron Marchbanks: That's not going to stop anytime soon.

Garrett Fritz: It's not going to stop anytime soon and it's not even going to slow people down. Honestly, as much as we get this all the time because half of our business is the startup Y type people. The people who've got some idea, they've played around with Replit or something and the other half is more mid market enterprise context engineering type stuff. But for the startups, eight out of ten of them have tried something on Replit, they've gone on there, they tried to build a proof of concept, they just are unfazed with those worries of hey, if you connect this to your stripe account, you may end up getting a lot of billing or do you know what the cost of hitting the Google places API is. And if you just try to scrape for all of Southern California, you're going to get a $10,000 bill. It doesn't bother, right, because it's like it's hard for them to kind of internalize what the risks are about API, key leakage and utilization. And no one really has a concept of software costs. So I think the warnings aren't slowing people down so much. But when it comes to like trying out these tools and like pretending to be moonlighting, um, as a software developer, what I think will slow people down is the cost of these tokens. You're seeing more and more of these platforms dropping these kind of flat rate $100 for all you can eat tokens. And like just yesterday, Copilot is switching to usage based tokens because they're all just losing money hand over fist.

Aaron Marchbanks: Right.

Garrett Fritz: And so the gravy train of just like token up everything and every model that becomes even more token greedy to get things through is just going to be now a bigger barrier. So over the last couple of years it's been the promised land for all these folks of just, yeah, just whip it out and get your proof of concept out and it's going to become pretty close to now those proof of concepts might, $10,000, might cost you $50,000. Maybe you want to talk to somebody who can use those tokens a little more efficiently and actually get you a product that isn't a big throwaway at the beginning. So it'll be very interesting to see the cost models as they progress because you can definitely spend those tokens fast and that's not Even misutilizing the APIs that those underlying uh, go across. Right. So dollars will make a big impact in this market I think over the next six months even.

Aaron Marchbanks: Yeah. And I think that they saw that coming quick enough. There's a notion now, of course, with token maxing and for every hire that you make, you also need to hire an equivalent amount of tokens for that person. And yeah, they, I think they saw that early enough that they're able to make that adjustment. And I mean good on them for doing it, but holy smokes, I'm just reading some of those numbers. I don't even know how you can use that many. I mean if I was trying, I don't know, I could do that.

Garrett Fritz: Yeah. And like, just to give you a sense of like our environment, we do have a lot of developers and they're all leveraging these tools and I'm not the implementation engineer on any of these projects. But I've got two computers here, four monitors and everyone has two Claude sessions on it, right? And I'm still comfortably in the max plan of cloud. I mean I get some over utilization but I'm not powering a bit. But what ends up happening, and it's very dangerous, is it's very easy to connect your API key which is tied to your token utilization, which most people have uncapped to other third party integrations. And so just like you just hit the big uh, approve key, right? Yes, approve. It's like agreeing to terms of service and privacy policies, right? Once you get that first hit at getting some benefit, then you're just out there and all of a sudden you're connected your cloud key to your GitHub repo and now it's checking and doing pull request reviews on all of your third parties that you've pulled in. Oh shit, now you've just spent $10,000 for completely useless stuff, right? That's not product development. That's just like completely tangential to the work you're trying to do. It's just technically you asked them to do it. So fun stuff and you can really run up a bill and I don't think they're super cooperative with refunds on those. I have not yet heard anyone who has been able to call that back.

Justin Edwards: There's an actionable takeaway for anybody listening. Put a cap on your token utilization. They will make it easy for you to raise it if you need to. But just put one in place because you might get surprised. You might get surprised that it might be ugly. So with your customers and the people you're advising and working with in terms of bringing AI into product or having product features that are AI, uh, not AI, uh, as a coding tool or as a marketing tool, what do you think that's working or what are people doing that's kind of interesting.

Garrett Fritz: The big thing that we're driving it to and uh, we're finding quite a bit of appetite for is the uh, kind of governed utilization of agents. So this is more for like the mid market enterprise type customers. The folks that are already pretty cautious about their data, the exposure of that, they have compliances like ISO, SoC2, HIPAA. In some scenarios it's non negotiable. They can't just hook up a robot to all their data and just say go to town, right? But the whole utilization of and concept of having these agents and LLMs accessing is to enable sort of loosey goosey development and product implementations, right? So the real question is, well, how do we enable our especially non technical staff to be able to leverage this agentic AI across uh, our systems without like sacrificing our compliance and our customer faith in our data security layers. And because they do have those enterprise layers of security on. So that was the topic of conversation at Google Next conference last week was they've, as you would expect, rebranded their Vertex AI system into Gemini Enterprise. And it was all about how you can leverage those enterprise level securities, role, uh, based access, IAM type policies and your agentic workflows. And that's been what we've been hearing from the market for a while is how do we turn that corner and not be just running off of our dev team's laptops and getting those kind of ad hoc reports, but how do we enable our entire organization and still have that confidence that we're not overspending, we're not developing stuff that just sits in POC limbo, but we're actually developing production ready agents that can be used across our internal workforce to enable them to do their jobs but then also potentially monetize and expose their agents to their public as well. So yeah, governed agent access and almost making it somewhat deterministic in its deployment and utilization, which is sort of counterintuitive, but it is putting guardrails around agents which is I think going to be the bread and butter of companies like us, moves us away from having to say oh yeah, we're not just replit or chatgpt, right, which the price point is just a fight to the bottom. This is. Now how do you deliver enterprise value to those customers that can afford to pay for it, can value it and really benefit from that sort of lover of security when it comes to a

Aaron Marchbanks: gentech workflow, would you say that you're still in the space of someone comes to you, a founder, CTO startup or whatever, and they have grand notions and they AI AI, I want to put this in. And you actually then have the conversation with them and discover AI is not the solution that they need or it's not the output that they need versus well yeah, you're absolutely right, it needs to do all of these things. Are you still having those types of conversations?

Garrett Fritz: Yeah. First of all, I'm not sure there's been any startup in the last six months that doesn't claim to be an AI company. Every single one. Some of them are fancy if statements and some of them should still say if statements for years and years and years. But yes, there's no one who comes through the door right now that doesn't ask for AI? Uh, and it's up to us to kind of determine where that fit is. And I'd say, like, half the time it's just wildly inappropriate. Or like the AI component that they're asking for is what we would call like a phase seven or phase eight, something like that. Because ultimately it's just not everybody wants a chatbot. And your users probably don't want a chatbot. A lot of times if they end up developing that sort of AI component, it's like, why don't your users just use ChatGPT to get what they're trying to accomplish?

Aaron Marchbanks: Right?

Garrett Fritz: And so what we're seeing and what we're trying to drive people towards is more of the niche implementation of what they're actually trying to accomplish instead. So get to the root of that problem and solve. So the answer is, yes, we still do quite a bit of that. Our market positioning is obviously evolving and moving around, but I'd say about 50% of our revenue is still like app products in some way, shape or form. Is it mobile app, SaaS products, stuff like that, where we're solving a specific problem for a specific user base. And half of those are very AI enabled. Like, they have their own agents, they're all deployed. They are not like chatbot necessarily, but like, you're getting recommendations and streamlined workflows and using LLMs and those function calls and tools and MCPs and things. But the other half, you know, still traditional apps, they're just like, very specific to the problem they're trying to solve. They have a customer base already ready to utilize those, and they're just happy that the company has AI in the domain.

Aaron Marchbanks: Have you run across any that are, you know, you get the requests, you kind of validate the request. Okay, yes, I think this probably would be a good opportunity to inject some form of AI into your services or systems or whatever. But it's been more challenging or more engineering heavy maybe than the founder was thinking. Like, they came in the door with, oh, no, I just want to do this. And you have the conversation. It turns out to be, like you said, seven phases deep worth of effort.

Garrett Fritz: Yeah. When it comes to, like, generating and benefiting from a, uh, lot of these concepts with the AI, I mean, the first thing that anyone's telling you is, oh, we'll just train it and it'll get better over time. It's like, okay, great, but you're not overtime yet, you're on day one. So you have no user data yet. And so yeah, we can leverage AI to consume your data tables and your user patterns and all that, but honestly, just look at it yourself like there's not that many rows and you get a lot more specific inferences just by talking to your users and getting that information. And so where we're seeing that it's worth the squeeze on day one to really implement these agents and access into your data systems are for larger companies with lots of users and lots of data, right, where you can't just look at it and pull it via CSV into the spreadsheet and just create a chart right where you've got hundreds of millions, billions of records across different geospatial areas and things that you're not just specifically asking a question that you could just do with a SQL import. It's more of are there anomaly detections that we should be aware of or fraud happening on our system where it makes a lot of sense, Whereas a lot of these starts are saying, oh yeah, we'll just use AI to figure out where our most valuable users are. What's the lifetime value of ones that click this button? Well, that's a SQL statement, that's not AI. And so do you really want to pay us two weeks of effort to put in this really like flexible piece, uh, of software in place for you to query your data, or do you just want to ask GPT to write one SQL statement for you? And uh, a lot of product teams, small product teams are kind of falling over that crutch of is this really AI first that I should be doing this? Because that's the pressure, downward pressure they're getting from the CEO, from the board, how are we being faster, all that kind of stuff and they're just blind guesses. Well, maybe I did this part of my job last week myself, maybe this I should spend the time doing it in AI and sometimes that's okay. But really there's better usages of your time to get things moved across that way. So it's certainly contextually relevant for where the request is coming from. But more often than not it's. They're probably barking up the wrong tree on some of those scenarios of this must be an AI job.

Aaron Marchbanks: So on the internal side, I'm curious how it is that you handle your team, your team specifically, but also I guess in general development teams that are building these things out. Not necessarily related to AI, but sure, everybody's using the tooling to varying degrees and I haven't talked to anybody in a long time. That doesn't do something with AI, but just in a general sense, how is it that you govern your development staff or your development team when building out these features or these products as it relates to AI?

Garrett Fritz: Yeah, that's where some opinionation really comes into play and is required. Let's step back. Ten years ago, before AI, right, you asked 10 different developers to do the exact same task. They're going to do it 10 different ways, completely wildly different. Now you ask 10 different developers to build something, the exact same thing, you're going to get 20 different ways because they're all going to do a couple different variants with different AI tools and they're always going to try to like jump across and do something that's kind of fancy. And now without having some sort of governance or mechanism in place to sort of enforce, encourage and guide your own team, you're going to be skinning this cat every way possible. And none of it are kind of following the letter of really what me as or us as the leadership was really expecting. So how uh, we're doing it is we've got text files, right? All this AI world is just text files and skills and things like that. But they're opinionated in terms of like, if we're going to deploy a cloud based API, this is the infrastructure we're using. First we need to decide if the customer is opinionated on aws, Azure, Google Cloud, anything like that. If aws, then we will deploy using these services that AWS has. Oh, by the way, here are the script files that deploy all those. Oh, by the way, this is how we do environment variables across that system. And oh, by the way, this is how you do start up a local developer, right? So like if a new developer comes to the system, they just run the one line and they're up. And that also is powered by Claude as a tool. We also are prescriptive about our developers, particularly going to use Claude. These models, we track their utilization and prompt calls that they all have. And as a backup we have codecs because just like today, Claude was down for two hours and your whole business can't be out of commission if you don't have a backup there. So we spent a lot of time kind of generating that infrastructure. Now does that mean that our infrastructure of skills and opinions for how to develop these systems with AI is the right way to go? It's a way to go. It's not the best. We're not advocating advocating that, but it's consistent. And that means that when our developers and our team and even our product managers, our QA people are working in this space. They can not only bounce from like project to project and have consistency in knowing that if there's a problem, I can look at this system. If I want to look at console log, I can go do that. If I need to ask for access to a system, this is the way that I go ask for it. So it's more important that we're just internally consistent versus being ensuring we're not just at a local maxima in terms of productivity and importance. We don't need to have 100% optimized everything we just need to be consistent is way more important. There's always going to be a flasher tool and we'll evaluate that when it comes through. But until then, we're relying on the decades of software experience to know that these systems work well and we know that the AI, uh, agent can work well with those. And we're kind of sticking there. And that's where we continue to push our expertise and where we're bringing our value. Add is not so much on the writing of the software, that's not so much the mechanism, but our ability to continuously evaluate new providers as they um, come out, cost efficiencies on their platforms and then providing like product and business insights on those deployed systems once they're out there.

Aaron Marchbanks: Gotcha. Awesome. Well, kind of coming up on time here, but Justin, I'm going to give you your last question, man.

Justin Edwards: I'm going to tweak it slightly, but it'll be the same that it usually is. Which is what is one thing we didn't ask you about that would be valuable to our audience or valuable to people who are working in this crazy AI development software SaaS space.

Garrett Fritz: We covered a lot of ground though, didn't we?

Aaron Marchbanks: Yeah, we did.

Garrett Fritz: So anyone who's jumping to this, I'd say target the folks who are maybe not a developer, not a CTO, but maybe your CFOs, your COOs, or someone who's just a non technical founder who's like, hey, you know what? AI's uh, got to be my solution here. My recommendation is try it out, dip your toe a little bit in, but then don't be afraid to ask for help. If you're really coming in with no background experience on the whole world, you will not catch the rest of the package. The learning curve is too steep, the tools change too often and it's okay to get a sense of where things are at. So you kind of feel the lay of the land. But don't be afraid to reach out for some expert guidance when it comes to how to leverage these most effectively. You'll find that consultations are pretty cheap and will get you kind of set down the right path and even help you think about new things. But AI uh is going to enable you, but it's enabling your competitors and your audience as well. So don't just think that you can turn the keys that everyone else is turning and get that Ferrari to purr. So we're here to help.

Justin Edwards: Love that.

Aaron Marchbanks: Take the test drive, but maybe don't back all the way out of the driveway.

Justin Edwards: At least put your seatbelt on.

Garrett Fritz: If nothing else, put your seatbelt on.

Aaron Marchbanks: That's right.

Garrett Fritz: That's great.

Aaron Marchbanks: Awesome. Garrett, again, thank you so much. We have been chatting with Garrett Fritz, partner and CTO at Meta cto. What is the best way for people to get in touch with you, Garrett?

Garrett Fritz: Uh, our website, metacto.com, metacto.com has got a lot of resources for you up there, A lot of descriptions of how we're doing, the things we're doing, some of the case studies that we're doing on those agentic workflows and those kinds of assessments, a lot of resources. So, yeah, metacto.com is the place to be and it was a lot of fun. Thanks for having me, Aaron and Justin.

Aaron Marchbanks: Absolutely. Justin, a pleasure as always. And we'd like to say as always, too, thank you so much for our listeners and subscribers out there. If you like what you're hearing, do hit that subscribe button. Let us know if you like what you're hearing. You want to see something different? You got some ideas, Just throw them our way and we'll catch everybody next time on Sas, that app. Thanks for cruising along with us on sas. That applies. We hope you grabbed some insights that were inspiring, actionable, or at least entertaining.

Justin Edwards: If you enjoyed the show, don't forget to subscribe and leave a review until next time.

Aaron Marchbanks: Keep building, keep growing, and keep those apps sassy.