24 | AI is the New ROI: Practical Ways AI Drives Profits and Growth for Your Company with David Hirschfeld, CEO of Tekyz

0:00Hello, and welcome to Leveraging AI. This is Isar Metis, your host. And in today's episode, we're talking to David Hirschfeld. He's the CEO of Techies. and he's going to share how his company is applying AI to more or less everything they do in the business, how he is creating an environment where his employees are encouraged to use AI and how they're creating a safe environment for them to test different AI's capabilities so they can then deploy them across everything that they do. It's a fascinating episode with a CEO that really gets it and that understands the potential benefits and that he's willing to invest the time in order to reap the benefits later on.

0:41And the later on is actually in the immediate future. If you are in any leadership position in thinking how to get started or how you can apply AI in your business, this is definitely gonna be an amazing episode for you. If you're anybody else in business, it's still gonna be really interesting. At the end of the episode, I'm gonna share some news as always on what happened this past week that is relevant in the AI world. And now let's dive into how you can apply AI to everything in your business. In the next few years, AI technology will change our world dramatically. Whether you are a business executive trying to catapult your business forward or just somebody who refuses to be left behind and want to advance your career, this is the show for you.

1:29I'm your host, Isar Maitis, a serial entrepreneur and an AI enthusiast. You'll hear invaluable practical tips from innovative business leaders, AI practitioners, and some of the brightest AI minds in our world today on how you can leverage AI in ethical ways to advance your career and grow your business.

1:53Hello and welcome to Leveraging AI, the podcast that shares practical ethical ways to leverage AI, improve efficiency, grow your business, and advance your career. This is Isar Maitis, your host. And as some of you may know, I have been in multiple tech startups in leadership positions, either as a CEO or as head of sales and marketing in different stages of my life. But developing new products in the tech world is one of my biggest passions, and I still find it fascinating and an amazing thing to do. I'm fascinated by the fact you can take an idea from somebody's head and make you do a product that actually helps people, provides value, and generates revenue to several different companies, including the company that developed it.

2:36AI makes a lot of steps in that process easier, but that's a very big statement. How do you take that statement and break it into actual building blocks that you can use when you're developing products? Our guest today, David Hirschfeld is doing exactly that. David is, first of all, a serial entrepreneur himself. He has founded and sold more than one business. He's currently running several different businesses as a CEO. And in the past several different years, he's running a company called Techies Corporation, spelled T-E-K-Y-Z, but sounds Techies, where he helps startups launch their products.

3:20So going from idea to revenue in the shortest amount of time. And in this past year, he's been focused a lot on adding and infusing AI into all of his processes in order to make that process for his clients even faster and more efficient. Now, because I think that every single company that develops anything, but especially that develops new products, will have to include AI in the processes or their competition will, and then they will not be competitive. I find this as a critical aspect that every single company has to learn how to do. And hence, I am really excited about the conversation that we're going to have with David.

4:00The conversation is going to be around what aspects, what steps in the journey from an idea to a working revenue generating product, you can have AI help you do faster and better and exactly how to do that. And since, like I said, I believe this is critical for the future of every company. And since I'm personally passionate about it, I'm really excited to have David as a guest of the show. David, welcome to Leveraging AI. Thank you for that great introduction, Isar. David, let's really start with, If you want high level, what are the steps, like bullet points that you help new startups go through?

4:45Before we talk about the AI, so people have the framework. I come to you as a potential client and say, David, I have this idea. I want to build this into a product. What are the five, six, 10 different steps that we have to go through in order to go from an idea to an actual running product? Okay, that's a great question. And there's a couple of different ways to go about that. with us, with techies. One's the classic way, which is where you have an idea. Sometimes it's five bullets on a napkin and it's going to be the next unicorn. And so - Always is. And it always is, right. And we spend a lot of time then basically distilling the requirements that the application will have that we need to build that will deliver these five bullets.

5:35of wonder and to the world. And so we start there and we come up with an estimate for the project because we've got to have some idea of what it's going to take to deliver that. And then we try to scope it back to what has to be there in the MVP. And this is very difficult for a lot of founders to not only scope it, but then to maintain the discipline as we're building it, to live within that scope of the minimal viable product. So then we start to develop, then we do the design and we build a prototype, a high fidelity prototype that looks like the finished product with all the user experience so that we can see exactly whether this idea holds together and that the design is appealing, but more that the user experience all holds together.

6:25And then we build it and we try to build it as quickly as we possibly can and limit with some limitations in the first version to keep the cost down so that we can get something out quickly and see whether or not there's product market fit. And nine out of 10 times, there isn't. And now it's a matter of pivoting and hopefully finding that product market fit before you run out of money. That's one way. The other way is to test the market first. So we still do the estimation. we still do that high fidelity prototype, but we also go through a rigorous niche analysis, identifying that early adopter niche, their value proposition, and their top two or three issues that they need to solve.

7:12And we know they're the early adopter because those two or three issues that are their big problems have a higher perceived impact to them and a higher cost combined, those two independent numbers, than every other niche and all of the problems that you might solve for them. And so when you've nailed that down, you have the early adopter niche. And then after we build the prototype, we build a marketing stack and we go out and try to do pre-launch sales. And that way you find out very quickly, you're forced to face the truth. Do they really want my product or not? Because if you can't sell it in a pre-launch sale, you're probably not going to sell it once the product comes out.

7:51But now you haven't spent all that money, I haven't spent all that time, and you can pivot quickly and cheaply because you're not having to pivot a finished developed product. So those are the two paths that people can choose to go to. And surprisingly, a lot of people still pick the classic path. Even if they say they're going to do the other path first, and we go through the whole process, we get ready to market, they'd say, once it comes out, it would be so much easier to sell. I was like, no, don't do that. Go out and face the music now. I love that. And I think especially most people, when they develop a new product, they have a client in mind, meaning they have their previous employer, their previous client, an area of expertise that they have with relationships that they have, that they have the opportunity to do exactly what you said.

8:43It's somebody that will be willing to be their beta test for something quickly because they really have a real need and they really want a solution and they'll be willing to accept a not perfect thing. So I think in many cases, from a founder perspective, it's doable. And I'm actually surprised that most people don't take that quicker path. But let's go back now that we understand the framework and figure out where are you using AI right now and where are you planning in your process to use AI further in a way that can help other people who are in the same boat or planning to develop a product or developing a product to learn from that and implement them myself.

9:24So implement it themselves. So you started in the beginning with, you know, I'll let you follow your path of which steps in that process you're using AI and how. Okay. And, and it's probably best if we just, if we talk about it almost like our journey, our AI journey. Okay. I'll jump ahead. I won't go into all of the open AI stuff we did last year. But earlier this year, we started using it to write or to refactor queries just because we had some big applications in law enforcement in particular with one client. And we had huge stored procedures that were complex and some of them had some performance issues.

10:06So we started to literally take that code just for the stored procedure. Nothing is exposed to the world, right? Put it into AI and then say, we'd like this to run faster and more efficiently. And the first time we did that, it was 150 lines stored procedure and it's 40 lines about 10 seconds later. And it ran faster and it ran the first time. We were somewhat floored. This was in January, I think, with ChatGPT. So we had done other stuff with OpenAI prior to that, but that was our first experience with ChatGPT. And that was in, I think that was in January. So we jumped far into the process into actually writing code as far as the journey that you described.

10:49But I do have a follow-up question that I'm very fascinated by since we started talking about this. ChatGPT is limited with how long the prompt can be. So you can't drop, you cannot drop a whole chunk of code in there. You have to break this down into specific segments. Is this what you did? Like you took just shorter segments and run it through ChatGPT? No, we put the whole thing in. And it was that much code, 150 lines. But now you can put quite a bit into there now, like 10 pages of content. Actually more if it's code, but 10 pages of written content, you can put in a prompt. If you're using OpenAI, you have a bigger range that you can feed it.

11:32And then you can do things to chain it together, but that's all getting like way too technical, right? Okay. Interesting. So from your experience, the first thing you tried that was like, oh my God, was literally just letting it help you write a more effective code. It rewrote code that we had written. And I think about the evolution of that piece of code and the time to debug it and everything else that we had spent on it. And then just gave it to chat GPT. And 10 seconds later, we had 40 lines that worked perfectly the first time. And that's not always the case, but it did there. But it was so eye-opening, right?

12:15And now all of a sudden we're thinking, okay, maybe we need to go back and refactor a lot of our code, especially areas. And then we said, okay, we want our code document. So we'd paste the code in there and say, document this. And it would write beautiful documentation of the code. No change to the code. Things like that. So these are all like little baby step things that we did really early on. But we started realizing very quickly that this has to be the direction the world's going. Tech stacks keep changing. You're stuck in your tech stack based on your skill set typically, right? unless you're willing to just have lots of different tech stacks and lots of different skill sets on your team or get very senior people to have broad experience, which is great to have.

13:03And our people are experienced, but still your ability to, unless you're one of those brilliant developers that picks up a new language overnight, there's still going to be a limit to a fixed number of tech stacks. That sometimes is not the most efficient way to go. There might be other things that are better suited. And one thing we noticed about ChatGPT is we can start writing code in tech stacks that we're not so familiar with. And we learn that tech stack almost overnight because it's writing the code for us and we can see how it's structured an API call, how it's called a function, how it creates an object or references, whatever those things are, how it's building a class structure, how it's creating microservices in a particular tech stack.

13:47And then also, and we're starting to get competent, like very quickly, even though we don't really need to be competent. We just need to be good enough programmers so that if it runs into a problem, we can then have it debug itself. We can know what, have it. Anyway, the whole process is shifting and we have, and we keep learning. That's gets us to where we were a couple months ago when we started to say, we need to formalize this whole thing. In addition to that, so the process for developing anything is write requirements. Then you take those requirements and turn them into detailed requirements, a design, user stories, test plan, test cases, test script if you're automating your testing, deployment, and then all the various pieces in your deployment.

14:40and your CICD and your automation and orchestration and all that, right? And all of that requires different specialties and different skill sets and different experience levels. And it has to be really well planned and really well orchestrated. And in a competent team, this works really pretty efficiently as long as you don't throw any big changes into the mix. So just thinking about, okay, how does a competent team function? So where do we want to use AI to start to automate this process? And so we do a lot of point solutions like with writing user stories. So our QA group is using it to write all kinds of user stories and test plans and test cases, or our business analysts for the user stories.

15:32So I got to ask you a question before we go. So the user stories, a new product, are based on a deep understanding of the problem, meaning it's not just understanding the software side of it. Right. Understanding the specific niche for which you're developing the solution, right? It could be on one day you might be developing a travel app. The next day you might be developing a law enforcement app. The third day you might be developing an accounting app. how do you use AI to help you with writing those user stories? First of all, AI, the open AI, any of the large language models have an amazing deep level of understanding of all these different domains and of all these different users and their behaviors.

16:24And it's amazing, right? We still don't assume it's going to know exactly how we need to create this particular piece of software and test it. But we're not counting on AI, at least not yet, to do all the user experience design. So all of that is done first. And we do do deep analysis in that process with our clients and with that domain and with all of our past project experience and all these other domains, because it all comes in and applies. But when we ask it to write a user story, we're writing a prompt that gives it enough context about what we're trying to accomplish with this particular function or screen or whichever we're writing it for, or user story for an API, for example.

17:14And it writes that user story with a lot of competence. And some of the mistakes that we might make because we just miss things, it doesn't miss things. It may go in the wrong direction sometimes, but we're reading that and evaluating and saying, no, this isn't what I meant. And then we re-engineer the prompt and then we're getting the result we want. Question. Yeah. First of all, I love this. I think this is a really great way to use, just like you said, these large language models have read most of the stuff on the internet, at least the interesting stuff, and definitely software development stuff it has read.

17:52and has those expertise on how to do these documents in a very efficient way. Yet, like you said, you have to give it the right context. You have to give it the right formatting for it to work for your particular use case. Can you give me the structure of the prompt that you're using to do that? So obviously not the whole prompt because I assume it's a pretty long prompt and I assume some of it is proprietary to your client. But if you had to break it into components, what components do you put in a prompt like this to write a user story? And if you can read some of it of what you actually wrote in those components, that would be awesome.

18:29I think that will give people a better understanding of what you mean by prompting for writing user stories. Okay. I'm actually going to go a different direction than user stories only because I'm not the one, it's my people on my team doing the user stories. They're the ones developing the prompting. I will give you an interesting example, which I used to train my team how to think about this. And it's nothing to do with software. But it requires you to rethink your relationship with technology in terms of your relationship with AI, because it is a relationship. Because what happens is we've been doing things a certain way for so long, we don't realize that we have an expert available for everything at any moment.

19:16And what does that even mean? So we just moved recently to Vista, California from Scottsdale. I know we went the wrong direction. Everybody goes the other direction. Long story why that happened. But anyway, and my wife loves to garden. And here you can garden all year round. And it's a very different gardening here than it it was in Scottsdale. So she likes to do square foot gardening. So she wanted to come up with a square foot gardening plan in for our backyard. And so I said, okay, how about we ask ChatGPT? And she's rolling her eyes anyway. So while she was thinking through all this and she was on the phone with a friend talking about it, I started a conversation with ChatGPT.

20:01And I even put in, my wife wants to create a square foot garden in our backyard. There's two of us. How many square feet do we need? We're not vegetarians, but we like to have vegetables with most meals. And it told us, okay, for two people, and it told us how many square feet we would need and how we should probably arrange the beds. And I said, okay, great. Based on that, what are the different types of vegetables that we can grow in our garden and with along with their companion plants because when you're doing gardening you have companion plants certain plants like to grow with others and they don't like to grow with others so they'll hurt them versus help them and so then it built a companion plant plan together which beds this would all go into right and a nice list and then i said okay what about companion flowers because those are flowers you want to plant near certain vegetables because they draw the bugs that like those vegetables off.

20:57So then it gave us all the companion flowers. I said, okay, come up with a yearly plan that includes succession planting. So after one planting is done, something else will go in that same ground, but you want to put the right thing in there because the soil is going to like to grow something else. So then it did a whole plan for the whole year for three different seasons with the successions. I said, okay, consider a square foot gardening, number the beds and give me a table with all, and it built this all out. And then five minutes later, I said, is this what you're looking for? I just showed my wife and her mouth just fell open.

21:31There were a few more prompts along the way, but this is what I mean by a relationship with the AI. So while we're writing user stories or whether we're asking it to write code or requirements, there is this sort of evolutionary process that we go through in terms of making sure that our relationship and the way we're communicating with it is getting through in the right way, in the right context. And we're giving it the proper feedback to continue to improve on the result. So first of all, I really love your example. And I want to take it really to the direction that you're saying of generalizing it of how you want to use it.

22:09And if I generalize what you said, the first thing is you need to have some level of expertise in the thing you're asking about, because you need to know what questions to ask. The second thing, I know what you're going to say, but you need a good starting point. The second thing is you have to be curious, quote unquote, in order to continue asking questions to get solid information. And the cool thing is, like you said, in some of the stuff you're saying, I don't know what other plans I need, but you can ask it, okay, what else do I need to know in order to make this successful? And then it will tell you, like, I don't know about supporting plans.

22:49I didn't even know that's a thing. But if you'd say, okay, what else do I need in order to make this successful? It will tell you. And they're saying, okay, that thing that you told me, what do I need to know about it more and so on? So you have to, a lot of people that I work with, and I do this as a consultant and as a training in courses that I teach, they get the first answer like, oh, that's not really good. So I'm wasting my time. And the whole process here is a, you called it a relationship. I'm just saying it's a process. Like you got to go the back and forth and back and forth and be curious and committed to asking continuous questions because in question eight, you're going to hit gold.

23:34Right. That's right. And what I, so let's say you wanted to do gardening, right? But you didn't know about it. You've heard about square foot gardening and you could start with the prompt. I'd like to do gardening. And by the way, when I gave this prompt, I said in Vista, California. So it was giving me all of the planting guidelines for the weather in our area. But I could have asked it, which questions should I ask if I want to start doing square foot gardening in my backyard? And then it will list all the questions. And if those are questions that I can answer, then I'll say, okay, given these answers for these questions, and then I could just continue to go through this process, So I don't even have to know anything about square foot gardening.

24:18I could have easily ended up in the same place maybe a minute or two later. I love that. I think that's a great addition. So let's go back to your company and let's go back to the process you described in the beginning. You said that the first thing you do with clients is to try to estimate the scope of the project. Do you use AI for that as well? Yes, we are now doing that. Right now, I know this isn't the most important thing, but we are building an estimation tools using AI to estimate our project, which is a many-step process in AI to do that. But in the process of doing that, there's a lot of other benefits that come out of this.

25:05Number one is it expands the requirements to fill in gaps. and one of the steps are questions that come out of this that we might need to clarify with our clients because there's not enough known in a certain area to have enough detailed requirement to build something. And we ask AI to identify those patterns so that we know that the estimate we're giving is more complete and richer and comprehensive. it. So then we want those requirements to then be expanded out in a full functional spec, and AI is doing that for us. Then we want to build a traceability matrix out of that, a requirements traceability matrix, and it does that.

25:54And then it asks us what tech stack we plan on using, right? And it pulls all these tech stacks out. And we don't have to use those. we could actually write in one that's not in the list. And then based on that tech stack, it starts to come up with risk factors and effectiveness scores and all of these other things. So we can see whether this is really a direction we want to go or not before we commit to that. Anyway, so it's this multi-long step process. And at the end, we end up with a full breakdown of the project and effort estimate for each of the modules in that, or each of the elements in that project.

26:34You're not even on the module level, but at a functional level. And then we have two different AIs that we use so that we can cross-check the results. It also tells us team size. It'll also tell us, and if we give it the duration of the project, it'll tell us team size. Or we say, here's the team we have on this, and that'll tell us what the duration of the project will be. Incredible. Okay, let's break this down because what you said is mind-blowing by itself. even if you don't touch any of the other points. First of all, how does it know? So I'll start with the first thing you said. The first thing you said is understanding where there are gaps in the requirements.

27:17How do you do that? Do you feed it the existing requirements and ask it, do you see any gaps? If this needs to do one, two, three, and these are the requirements, where do you see gaps? That's the idea. We, yeah, so we feed it a set of requirements. And if it can, it will just fill in those gaps and build out the rest of the requirements. If let's say we're starting with five bullets, then what it will, then we can just ask it to what questions do we need to ask to get a full set of requirements from our client. but if we feed in let's say five bullets we say just go ahead and create a functional spec out of for this for a full plug and it'll do a decent job but then we'll take the result of that and put it back in and say okay now take this and continue and it'll build it out at even a more detailed level until we feel like two times through is pretty much it you don't really need to go beyond that until we know we've got everything that needs to be there.

28:22Now, the estimation part of it, it has to know how to estimate the development, which means it needs to know either your team or an average team. I don't know if that's worth anything because you need it for your team because you're trying to make money in the process. So did you take previous project information and fed it somehow into ChatGPT or whatever tool you're developing as a training process so you can get better results? There was actually an easier way to do it. We just look at our team in terms of what we've done and how long it took us to do it. And then we asked ChatGPT to say, how long should it take to build this thing?

29:05And where our numbers were coming out, some percentage of that number, right? Some projects a little bit more, some projects a little bit less based on what ChatGPT knew. So that gave us a good understanding of how it was estimating projects based on what it understands. Oh, I understand what you're saying. So it's a lot easier. You now have a benchmark knowing that for the projects you have done and you know exactly how long it took, you asked ChatGPT how long it thinks it would take. And now you know percentage-wise up or down, depending on the type of project, this is where your team will score compared to what Chachupiti thinks it will take?