282 - Best of 2025! How Fear and Regulation Impact AI in Finance with David Kaylor

0:00AI is helping me automate basic tasks in my work life. I'm on PTO. It's still going to run that whole thing. Anything possible with code. I'm not a programmer. I was just like typed a prompt into chat GPT, wrote it in Python, and I'd like manually ran it for probably a year. Every morning I'd click a button just to execute it. First try, no errors. It just worked. I was blown away.

0:27All right, David, super stoked to have you. I'm really excited to learn about some of the AI stuff that you're working on, where you are. Really quickly, will you give the people a brief background, who you are, where you work, what your role is? Yeah, for sure. So David Kaler, work at Alchemy Technology. We do banking services, digital banking services for credit unions, banks, across a variety of use cases from like the app you log into, to digital account opening, to loan origination, all sorts of stuff. My specific role is really around implementations of those platforms and services, essentially plugging our wires into banks' wires, making sure all of that works.

1:11Lots of fun. Yeah, you were saying that most, you know, credit unions or regional banks probably haven't built their own platform. They're using you or one of four of your competitors. Is that right? Yeah. I wonder, what's it been like over the last, let's say three years, two years? Two years ago is when OpenAI had their moment. It was like, oh, wow, chat GPT. I can ask it questions. Oh, my gosh. And like over that period of time, the companies have sort of shifted. All of a sudden, you're seeing roles like, oh, he leads our AI implementation team and et cetera. What's it been like working at a tech company that's servicing these fairly old legacy businesses?

1:51Is there a lot of talk around AI? Is it still nascent? How do you view it? I would say there's a lot of talk. But not a lot of doing. Yeah, yeah, yeah. I would say the doing is a little, not a fast follower. It's a slow follower. We're getting there. Honestly, I think we're doing okay internally. But I would say the industry as a whole, there's a lot of excitement. and we have a data product that does use some AI modeling that we'd send out to our clients and stuff. And that's really well done. But I would say as a whole in the industry, I would say there's more excitement than actual productivity.

2:29Do you think that's because of the regulatory environment within banking? Yeah. There's a lot of fear and apprehension. I would say in the banking sector, if you don't know if you're allowed to do something, people just assume you're not allowed to do it, which I would say is very different than most of the business world. If it's not in the business world, if it's not explicitly denied, then you assume you can go forward. And I would say in the banking regulatory environment, it's quite the opposite. Healthcare is very similar, right? It's like, hey, we don't move fast and break things in healthcare.

3:04We take our time. It was the same with banking. So tell me how you are currently using AI, both personally and within the organization. Sure. So I've tinkered with a bunch of different things. I like trying the different models for different things. I would say my best personal use case has been Replit, the AI engine agent model where it actually does all the hosting for you, which is brilliant. I'm somewhat technical in my background, but I'm by no means a developer or like a server engineer to spin all that stuff up. So it actually spins up the servers, the databases, it hosts it all, you know, serverless native in the background, whatever.

3:47Basically, you don't have to know any of that or think about it. And so I built a little mini game on Replit that is essentially like, it's a click based monster destroyer. I can bring that up if we want to look at that it's kind of fun it's a time waster you know it goes on for forever this is more of a like entertainment personal absolutely okay while you're pulling that up i like to ask people this question what models are you using and what are you using them for currently good question i really like gemini the google one which i know gets a lot of hate but it's good at one specific thing and that's knowing its own ecosystem so i write some google app script and who does google app script better than google and so gemini is pretty good at that i had a i had a crappy little python script that i was running manually every day in the morning on my laptop that would connect to some jira apis and send an email and i popped it into gemini and i said, hey, I want you to take this Python script and output me a Google Apps script that can then be scheduled and automated.

4:56And it did it. First try, no errors. It just worked. That's awesome. I was blown away. It's funny. I haven't heard people talk about Gemini, but that makes sense from that specific use case, right? It does its ecosystem really well. Right. Outside of that, I don't use Gemini. Okay. So what are the other ones you use and for what? I think Grok is really good at basic scripting, Python or Apps Script. Sometime I'll pop over there. I use ChatGPT a lot for like general stuff. I would say it's kind of slow. The output, it takes a long time to think and then like write a response. Especially after using Grok.

5:35Man, Grok is so fast. It's at least comparative to ChatGPT. Okay. Okay. So ChatTBT, Grok, Gemini, do you use anything else on a regular basis? No, I guess whatever's built into Replit, which I think is Claude 3.7 at the moment. Yeah, that's interesting because a lot of people use ChatTBT as like a thought partner. It's just general, hey, how does this work? And then they'll go to other things for specific use cases. Claude, the anthropic model, Sonnet 3.7, I've heard a lot of people using for coding, when you're using it for replets, so you're effectively using it for coding. Grok, I've heard, for real-time data and for some, not analysis, but the deep research function, supposedly, is really good.

6:22I haven't used it for deep research. I use JGPT for deep research. So anyways, it's just interesting to hear the specific use cases that people use for each one of those. I'll bounce around too. Most of my deep AI work is scripting. And so when I run into an error, you tell it the error and it just repeats the same answer over and over. And it's like, okay, you're not getting the error. Then I'll switch models, paste the same code in, same error, and it'll approach it from a different angle and then usually get past the bug or the error or whatever. And it sounds like you're mainly using it for coding, like creating the automations, creating the workflows to help you.

7:00Correct. I don't actually have any automations using AI where I'm like pinging out to an AI to do something. it's the AI is helping me automate basic tasks in my like work life. Well, next month, it'll be a year since I started this podcast. I've had over a hundred interviews. It's been a ton of fun, but I promised my wife at some point I would decide whether or not this was going to be a business or we would just continue to pay for it out of our pocket. And I'm exploring monetization, to be honest with you. There's not a ton of costs, but I do want to figure out a way to cover the editing and the packaging and the promotion costs of putting this podcast together.

7:34So I am looking for sponsors. I don't know what that looks like. If you've enjoyed this podcast and you've ever found some value and you want to sponsor, reach out to me, Nick at cofounders.com and let's get back into it. Cool. All right. Let me see the game. I want to see the game you created. Yeah. So this was, I mean, it's not fancy, but it only took like 20 minutes. So I don't know if you ever played the paperclip game back in the day, but you could like build a paperclip factory and then do like took over the world. So you would essentially click. It's a clicker game. And I'll start iterating through this.

8:10And you can see as I click, the monster health goes down. And then it dies. And then we have a new monster. And you'll notice I have gold and a score. And it starts to go up. And so this is the same premise as the paperclip game that was popular. Oh, man, what was that? Like early 2010s probably. And built with a UI of like the 1990s. But man, I spent days. I had it running on a server in my last job so it could let the paperclip factory run overnight so I could like accumulate all these resources and take over the universe with with paperclips so anyways that was the inspiration here and I was like I wonder how hard it would be to recreate like a spinoff of that and so I have a I added a little storyline it's not great but I didn't write it AI wrote it I'm like hey give me like a thousand part storyline broken into a sentence is in this like JSON format and it did so we have a storyline that's awesome so I have gold I can upgrade my click so it like kills the monster faster now it's stronger I can do passive income so now if I don't do anything I just get two gold per second just by sitting here okay passive again so what did you create this on replet this is hosted on built and hosted by replet it took me like 30 minutes to put together can you like can you pull up the replet project that created this was it all natural language that you used to create this or were you actually coding no i i'm not a programmer by trade at all so you went to replet and you said hey replet i I want you to make this game that does XYZ things.

9:58And then you just iterate it essentially to this. Correct. Yeah. Let me see if I can get it in my repli account. I think that's the wrong Gmail. Yeah. No worries. No worries. If you can't, I was just curious if you had like the ability to show, Hey, this is what a first prompt and et cetera. Okay. It's on a different laptop. Sorry. No worries. No worries. Okay. So this is the game function, which is awesome. and I'm sure you use this just to for entertainment. What are some of the other functions that you're using AI for at the moment? So the actual useful use case is basic scripting. So there's a lot of things in life that we forget about, just tasks that need to be done, accountability items.

10:45One example I can talk you through is like time tracking for my team. It's always a big thing for various reasons, But, you know, everyone hates time tracking. No one's very good at remembering doing it real time. And so I had built a little Python script that would do it. I talked about that earlier. What it does now, it's this Google Apps Script that I had Gemini rewrite from the Python into Google Apps Script. and it essentially makes an API call out to Jira and it says, hey, what tickets did so-and-so comment on today or yesterday, actually? And then it gets that list and then it says, hey, what Jira tickets did they log time to?

11:26And then it does a diff against those two lists. So we commented on ABC, but we only logged time to A and B. And so the remainder is C, was a ticket that had work done, but no time logged to it. And so then it sends an email to that person and says, hey, you did work on this ticket, but you forgot to track your time to it. Do you have a flow I can look at for this or no? I don't have a flow. What I'm trying to understand is Jira is a project management software? Yeah, essentially. Essentially. And it sounds like that's where your team is logging their time just because they're associating their time with tickets within Jira?

12:05Yeah. Okay. So what you're saying is it will go in and take one time point, subtract it from another, and be able to match what they self-reported versus what they actually recorded in the system? Correct. Yeah. Okay. Or it'll look at where they logged a comment. So they responded to a client or just dropped an internal note. So they took some action on this particular project, but didn't associate any time to that project. And so it finds that discrepancy and then emails them about it. All right. And what is actually doing it? Is it a prompt? Good question. How is it initiated? That's not the AI.

12:47The AI helped me write the Google Apps Script that actually does the work. So it's a Google Apps Script that runs on a scheduler every morning. and so it's just executing like appscript code calls the api gets the list gets the other list does the logic sends the email and so that's the function i wanted to work i knew it was possible like if you can logically think through it anything is essentially possible with code i didn't know how to write it i'm not really a programmer and so i was just like typed a prompt into originally chat GPT wrote it in Python and I'd like manually ran it for probably a year every morning I'd click a button just to execute it and I got tired of clicking the button and so I popped it into Gemini and I said hey rewrite this Python script in Google Apps Script because Google Apps Script has a scheduler and that was the key that I no longer have to click a button runs on like some server that Google has.

13:49I'm on PTO. It's still going to run that whole thing. When you're saying Google app script, I don't know what that is. So is there a platform that you can write these, I guess, function? I don't know what the right word is on Google. Yeah. So they're attached to four folks that have the Google cloud, like work suite. It's built in. And I think it's built into the personal Google too, probably. If you've ever used like Google Docs, Google Docs or Google Sheets. I usually attach mine to Google Sheet. Okay. So I would just open a new Google Sheet and I could do this? Yeah, absolutely. Shut up. It's like totally free.

14:26Don't lie to me. Okay. Any Google Sheet. And then you go to extensions. Okay. Google Sheet, extensions. And then App Script. Oh, so this is how I would get it automatically populating into the sheet so that I can see the data. Yeah, I mean, sometimes like this one, I don't even put through a sheet. It just does all the logic in memory. But there are use cases. So I have a different use case where I just want like a daily dump of all the Jira tickets that match a certain filter. And so every day it increments to this list that's in a sitting in a Google sheet. and it just adds whatever the diff was from the last time it ran to the end of the list.

15:09All right, now's the part of the show where I feel the most uncomfortable, but my therapist says I need to face my fears. So here we are. I've started a newsletter and I want you to subscribe. And what you're going to get every single week are the aggregated conversations from that week that I have on this podcast with an overview of what their business actually looks like. I'm also going to throw in a review of one or two businesses that are listed for sale. I'll give you my opinion on whether or not the EBITDA multiple is good or there's customer concentration whether it's red flags or green flags.

15:34And then lastly, I'm going to give you one piece of actionable advice every single week on how to buy your first business. So click the link below, subscribe to my newsletter, and let's get back into the show. You can do all sorts of stuff. You can do logic. You can manipulate the sheet. You can build a, if you want to like take the data in a Google sheet and like build a Google slide deck out of it, you can do that. Really the sky's the limit as far as it goes between like Google integrations. But it's also not limited to just the Google ecosystem. Like I'm making an API call out to Jira from Apps Script.

16:09So Apps Script isn't necessarily populating in your Google Sheet. That's just where you go to get into the Apps Script. Yeah, it just like is hosting it. I have like a dummy shell of a Google Sheet that's just like it exists. And then I have like all these scripts that have schedulers attached to them. Oh, so it's almost like your database of app scripts. Yep. And I'm sure there's a better way to do it, but this is simple and easy. Oh my gosh. There's no, I guess, AI. There's no context happening when you're running these, right? It's not like, hey, look at this and make a determination whether or not XYZ happened.

16:48It's literally if then statements being executed. Yeah. Yeah, basically. Yeah, it's pretty simple stuff. I wouldn't be surprised if Apps Script has like a Gemini hook where you could like ping out to Gemini and say, hey, do some logic or something on this. I'm not doing any of that, but I wouldn't be surprised if it's possible. Theoretically, I could use this to say, okay, fine. Tell me how many abandoned carts. I guess I could probably have an automated Shopify message to me. I'm trying to think of what use case I might use this for. Could I use this for my sales team? Could I use this for...

17:26You could use it for like, let's say you have a report, let's say in Salesforce, that you want to run on a daily basis that calls out certain accounts that are maybe in a red status or trending red based on certain factors. And you want to email your sales reps that are attached to those accounts, but you don't want to email everyone. You only want to email the ones that are trending red based on like, you know, four factors that may or may not show up in the report. Well, the script can like loop through that logic and then send the email or not send the email to like, Hey, go do your job. Dude, this was, this was awesome.

18:02I had no idea about, about app scripts. I'm not like, I can't wait to start playing, playing around with this stuff. Anything else you want to show? Anything else that you're like working on right now? That's the main, it's like, I do a lot of the same thing over and over again. So it's like that kind of pattern just across a variety of different use cases, but that that's the bulk of it. And that might be a good framework for thinking of how to bring AI in is like, if there's something that you're doing repetitively over and over and over, it's probably a way for you to automate it. Oh, definitely.

18:32And it's not that you're automating it with AI making decisions. It's that you're figuring out how to do the coding that automates it. It's like empowering you to be better at basic task automation. It's fantastic. This is fantastic. Thank you. Yeah, absolutely.