#304: I Can Haz AI?
Everyone’s posting their AI projects online like proud pet owners sharing videos of their cat doing something marginally impressive — cute, occasionally clever, sometimes a little cringe. And the Analytics Power Hour is no different! In this co-hosts-only episode, Tim, Michael, and Julie skip the thought leadership hot takes and just… compare notes. What have they actually built? What broke? What surprised them? From a custom GPT podcast librarian to a full-blown show production app wired up to Neon, Vercel, Resend, and about five other things Michael is only sort of sure he set up correctly, to a Gemini Gem that simulates a client interaction so realistically it raises your blood pressure in a safe environment — there’s a lot of ground covered. Plus: why AI-generated communication has a Stevia aftertaste, why deploying AI context across a team is way harder than it looks, and why the LLM will absolutely tell you what you want to hear about your Meta spend if you give it half a chance. This episode is also brought to you by Stape , your all-in-one solution for server-side tagging. This episode is brought to you by Prism from Ask-Y —your agentic analytics platform for automating analytics, exploring data, creating repeatable workflows, and delivering accurate insights—all without the need for manual query writing. Links to Resources Mentioned in the Show APH Librarian Custom GPT * Hermes Agent Vercel Resend Neon * Surprisingly reliable, but it was something of an exploratory flight of fancy, so we provide no guarantees of uptime. Photo by Eugenia Pan’kiv on Unsplash Episode Transcript 00:00:00 | Announcer: Welcome to the Analytics Power Hour. 00:00:08 | Announcer: Analytics topics covered conversationally and sometimes with explicit language. 00:00:15 | Michael Helbling: Hi everybody, welcome. It’s the Analytics Power Hour, and this is episode 304. You know, there’s no shortage of high-level thought leaders that want to explain this or that about AI, and I mean, we’re not a post. But what about boots on the ground? What are people actually experiencing when they’re using AI in their day-to-day? I don’t know that any of us claim to be AI experts, at least I didn’t check anybody’s LinkedIn profiles, but I don’t think any of us have updated them to that extent. Updating that? No. All right. There goes that claim. But we have spent some pretty significant time in the tools, and we’ve got a pretty good background in this podcast of data and analytics, and I thought it’d be fun to talk amongst ourselves about some of the experiences actually using AI, the good, the bad, then the just okay. So that’s what we’re going to do. We’re going to get into it. And speaking of AI experts, let me introduce my co-host, Tim Wilson, AI Thought Leader. 00:01:15 | Tim Wilson: It’s actually Tim Wilson, M-A-I, like a master of AI. I’m going to start putting that at the end of my pool. I like that. Yeah. I’m like, hey. 00:01:25 | Michael Helbling: Well, good. And Julie Hoyer, are you doing? 00:01:29 | Julie Hoyer: Hello. I’m good. 00:01:31 | Michael Helbling: I’m excited to hear some of these war stories. Do you get introduced to clients as an expert in AI? 00:01:36 | Julie Hoyer: Oh, thank God I do not. 00:01:38 | Michael Helbling: Okay. Is that an AI strategist? 00:01:40 | Julie Hoyer: Nope, not yet. 00:01:42 | Michael Helbling: In consulting firms, a lot of times as an expert, you sort of just get slung in as an expert in whatever the topic is about. 00:01:48 | Tim Wilson: And it’s like, yeah, that’s actually inaccurate. 00:01:52 | Michael Helbling: Let’s kick off the whole conversation by just chopping that tree down. 00:01:56 | Tim Wilson: All right. NPR or the Daily Show and our AI correspondent. That’s right. 00:02:00 | Michael Helbling: And I’m Michael Helbling. All right. So let’s dive into it. I think maybe a good kickoff is just set the tables, so to speak, because maybe let’s just talk a little bit about the types of AI tools we’ve used, what LLMs, just very high level, very quickly, maybe just go around. We’ll just chat about that. And then I’ll dive into more specifics. Yeah. So why don’t you go first? Well, I’m glad you asked him. 00:02:31 | Julie Hoyer: Flip those tables. 00:02:32 | Michael Helbling: So yeah, I’ve been using AI since whenever ChatGPT3 came out. But yeah, I would say obviously using a ton of Chat AI LLMs. So that’s GPT, Gemini, Anthropics, Claude. So a fair amount of coding on Codex and Claude. Have experiment with some other stuff too. And then on the agentic AI side of things, I have a little Hermes doing little things for me using GLM 5.2 currently. So that’s sort of the latest and greatest. So yeah, I publish code to GitHub. I do things like that. And by me, AI does that, which is, it’s true. I think my first GitHub commit ever happened after I started using AI. So yeah, it’s definitely it opened the door. It opened the door to a set of possibilities I had hitherto lain dormant. All right. Okay, Julie, what about you? 00:03:41 | Tim Wilson: Me. 00:03:42 | Julie Hoyer: Okay. My list is not as impressive as yours, Michael. But similarly, I’ve used Chat GPT more for personal stuff since it came out. Like I was interested, of course, like anybody, but more so through work and more of the stuff that we’ll probably end up discussing today. I’ve done through Gemini and then Claude. So those are kind of my two main ones. Again, the easy foot in the door, I think the place everybody starts. That’s where I’m at. 00:04:12 | Tim Wilson: Thanks. No, that’s good. Tim. So, I mean, so I platform wise, I’m kind of across the, all through the big three, the Claude, Chat GPT and Gemini. I’ve gravitated much, I would say I’ve gone deeper in the Claude stack and is so many people out there saying people who are just using it for chat, you know, they’re, they’re missing the boat. And so I’ve kind of pushed myself, I feel like in the Claude world, I’ve actually used Claude code beyond just using chat to help me write code. I’ve gotten into Claude code a bit. I’ve done some stuff with co-work, which has been, which is the stuff I’ve done with co-work has been mostly podcast related stuff. So that’s a little, a little meta, but I feel like I’m, I’m hitting a point where I’m starting to sort of understand the distinctions between those and on the Gemini front, that’s kind of like, I’ve been underwhelmed by it, but outside of chat, I think I’ve played with Gems, which I remember when Julie, you, I think last called Gems multiple years ago, you’re like, Gems are cool. And I think it was another like year or two before I actually, I was like, dealt with one. I wrote down like, what are Gems? 00:05:31 | Julie Hoyer: Gems? 00:05:32 | Tim Wilson: But I mean, just a quick one, which isn’t even really in my list, but speaking of Gems, Stern at marketing analytics summit talked about, talked about AI, using AI for like training junior analysts and how to like engage with, with stakeholders. And I was, I, at the time I had a reaction that was like, but the humanity, how do you learn to do this? And since then I’ve had the opportunity to work with a couple of Gems that are purely there to simulate client interactions. And it’s been, even though I can read the instructions and I know what’s going on, like the realistic level, even to the point of punching record, like, so I’m like, Gems Stern, I know you regularly listen to this, I’ve now like done that. And I am converted that, wow, that’s a hell of a tiny little application. But it, it, it’s really impressed me. 00:06:32 | Julie Hoyer: Yeah, that situation made me sweat. The one that Tim, you like took the base one from, I can’t remember who made the original, was it Ryan Dupont? Ryan Dupont spun up the original one at further, and then you took it and tweaked it and made like that second iteration. But when you sent it to us to try out, I was like, whoa, I’m feeling like the pressure respond. 00:07:02 | Tim Wilson: It was good. That was like a wild little light bulb to say, oh, this is like really raising stress level, but in a safe environment. And the way I was like, ooh, I’m on board. So that was a very. Eye-opening one. But sorry, I went to a specific example that wasn’t even on my list. 00:07:19 | Michael Helbling: But yeah, Tim, save it for the next step of the process. 00:07:24 | Tim Wilson: It’s not in the outline. 00:07:26 | Michael Helbling: That’s right. No, well, that is what we’re getting into next is sort of like, OK, so let’s talk a little bit about projects we’ve done. And you kind of mentioned, Tim, the podcast, and that has been pretty fertile ground for a couple of us to try, you know, what I feel like is a low stakes way to try stuff out because it’s sort of like, OK, it’s not my business. It’s not a client. It’s a podcast and we’ve got real work that we’re doing. But like if it totally fails, like we can just throw it away and nobody has to know about it. But some stuff has actually stuck around. Like, and I think that’s kind of interesting and maybe informative in a way of like, yeah, if you’re not started yet on specific things, could you find something even sort of outside of your work that still would be a good use? 00:08:15 | Tim Wilson: Well, can we talk because this was over a year ago that that was when I headed down the path. I was using ChatGPT to help me write Python, Python’s not my strong suit, but I needed it because I was basically trying to save us some money and shorten some time frames on the transcription. So it started with like one Python script that ultimately got pushed into Colab, but it was purely like, help me write a script. I’m taking the code you’re giving me, putting it in VS code, going back and forth, kind of following along with what it’s doing, which is kind of a chat based version. So that was like, I don’t know, that was vibe coding because I was generating code that is still Python code that we run in Colab and we have three or four of those for different reasons, different purposes, but you did a different form of vibe coding for the podcast when it comes to actually developing an app. Like is it worth having you talk about that? 00:09:14 | Michael Helbling: Yeah. I mean, well, there’s a couple of things we could talk about, one of which our listeners can actually access, which is the very first, this is the very first coding, one of the very first coding things I ever did with AI, which was back at the end of 2025, which is crazy because that feels like such a long time ago, but it’s really not that long ago. It’s like so long ago, it was like a year ago, no, it was eight months ago. But no, we had all the transcripts from the podcast over the years, and I just started thinking about the fact that’s like, oh, well, AI is really good at reading through docs and stuff like that. 00:10:00 | Tim Wilson: I was like, wouldn’t it be cool to create a little AI library for the podcast using 00:10:07 | Michael Helbling: all the transcripts over the years? And so that got the idea started. And then I did a company hackathon with stacked analytics. We all went out and did AI hack work for a couple of days. And so I picked one of that as one of my projects and built a little custom GPT that basically the work or the underlying stuff behind it was sort of giving me a chance to learn about concepts like RAG and VectorDB and stuff like that, which are sort of organizing concepts that AI kind of access their chunks up data, probably not as relevant now the way that context windows have expanded and things like that. But for that time, it was kind of useful to figure out, OK, how do I get all of the transcripts in one place? So in this case, a Google Cloud storage bucket, and then how do I move that into a VectorDB? So I use Cloudflare to do that. And then how do I move that back across into a JSON schema that a custom GPT can talk to? And so building all that out and building out the flow of that was really the hard work. But it was kind of very useful because any time you’re dealing with permissions in Google Cloud, if you’re not super familiar with that platform and setting those up, that’s a good thing to learn about. You definitely want to know how that stuff works. And it was good. It was a really good experience because it took a long time to figure out, but it actually 00:11:40 | Tim Wilson: then produced something that’s like, hey, that’s actually pretty useful. 00:11:43 | Michael Helbling: And actually, if you go to ChatGPT and you go look at GPTs, you can find it as the Analytics Power Hour Librarian, and it’s available right now. So you can actually go find it and then ask it questions about the podcast, which is really circular because this right now will be recorded, transcribed, and then available at some point in that custom GPT. Weird. Tim, your company launches a new checkout and conversions suddenly drop. 00:12:15 | Tim Wilson: I mean, what’s your first move? Open 17 tabs, blame Safari, and schedule a meeting called Tracking Investigation, Urgent, or try 00:12:25 | Michael Helbling: this one on, connect the AI Assistant you already used to your Stape account through the Stape MCP server. 00:12:32 | Tim Wilson: That sounds less traditional. 00:12:34 | Michael Helbling: Yeah, but you could ask it to check the server side container, validate the tracking domain, compare usage before and after launch, break down activity by browser or client. 00:12:45 | Tim Wilson: So it can help answer, did customers stop buying or did our measurement just stop measuring? 00:12:51 | Michael Helbling: Exactly. And you can also review daily traffic, see which domains are generating requests and check for unexpected usage or possible surcharges. 00:12:59 | Tim Wilson: Fewer tabs, faster answers, and maybe that urgent meeting becomes an email. Well, let’s not get carried away. Fair. Click the link in the show description to learn more about the new Stape MCP server. Michael, what if your dashboard could do more than just sit there looking colorful and ask everyone to interpret it? Excuse me. That dashboard took six months, 14 meetings, and at least one resignation. Oh boy. Well, Prism’s new app builder turns an idea into a complete interactive application right inside your workspace. 00:13:38 | Michael Helbling: Okay, so I just describe what I need and Prism builds the app? 00:13:42 | Tim Wilson: Yep. You can review the spec, you can edit it, and keep refining the same app in plain English. 00:13:48 | Michael Helbling: Oh, so no rebuilding everything because someone asked for one tiny change is somehow requires a new data model? 00:13:56 | Tim Wilson: Exactly. You can also share the app with anyone through a link. 00:14:00 | Michael Helbling: Oh, useful, flexible, and no 12-week BI backlog. Frankly, it’s a little disrespectful to the traditional process. 00:14:10 | Tim Wilson: Well, Ask-Y is looking for people who want to replace a BI dashboard with an application without handling the migration themselves. 00:14:17 | Michael Helbling: Oh, so point the agent at your dashboard, Ask-Y reads the logical layer and builds it as a flexible, extensible application. 00:14:26 | Tim Wilson: That’s it. Sign up at ask-y.ai, that’s ask-dash-the-letter-y-dot-ai. Use code APH to jump to the top of that wait list. Boy, I like the sound of this. Give your dashboard a promotion. It’s been through enough. Which is, brings up, because like you built it, you kind of hacked it together, you’re like, hey, guys, I built this, and then like, and we were playing around with it, like, this is so cool. And then the immediate question was like, well, is that going to, is it going to stay updated? Because it’s like, do we want a snapshot? And that then sent us down a path, which I think is also representative because then the work I was doing on the transcription and needing to get, I guess I already had, I’d already done some stuff where it was going into markdown files with historical stuff. But then it wound up needing to move into a different process that was a combination of co-work and Python scripts to say, oh, we’ve got to coordinate between these two things. Like, I mean, literally just today, I just processed an episode and part of what it does is it, there’s a, there’s a handoff like the end of this one process is going to drop an additional markdown file in that folder. And then the GPT, that process you built is going to be looking and picking it up. And you were actually surprised a couple of months ago, you’re like, I don’t even think that’s working. Yeah. And then I was like, what’s the most recent episode? And then it get pulled up and like, it’s working. 00:15:56 | Michael Helbling: It’s going. Which is great. Because I didn’t build the next thing on top of that, which is like monitoring and data freshness checks in my GCP storage, which you would do if you were building something for a data purpose. And have built for other things. 00:16:11 | Tim Wilson: So yeah, that’s a nice thing about, yeah, podcast related things. You can just sort of work on it when you want to and then not update it later so that Tim 00:16:24 | Michael Helbling: gets really frustrated. 00:16:26 | Julie Hoyer: Yeah. 00:16:27 | Tim Wilson: There you go. 00:16:28 | Julie Hoyer: Speaking of the permissions part that you were running into, Michael. So one of the things I most recently tried to do, honestly, this was inspired from our conversation with Rob Colley, actually, recently, where he was saying where the data builders and where the people that are going to end up going in, again, using AI for more than just the chat and doing more than just analysis in it. And so I was trying to get into more of like, could I go use cloud code? Could I build an interface, an app, something that refreshes? And so from a lot of what you guys were just saying, I was inspired to go and try it for this learning community that I run at work every Friday. It’s called The Guild. And we have 183 recorded episodes or at least sessions that have happened that we have like a drive folder for Tim, which I was like, I hope Tim is proud of that because Tim was working with me when we were running that together. So anyways, I decided first I was talking to Claude just chatting and asking, you know, this is my situation. This is what I want to try to do. Can you help outline some of my options so that I could take all of these recorded videos? And there’s like content and decks in there. It all lives in a spreadsheet that we maintain and keep up to date, but also then it links out to all the drive folders where all this content is. And it gave me a couple options and I wanted it to be for searchability and like discoverability. So not only somebody to say like, oh, are there sessions on this topic or these key words, but could it also surface some things? Maybe they didn’t think about that could be relevant. So I was really excited about that. But as soon as I went to try to start like building it, it kept like throwing this error. Suddenly I was like giving it a link to the spreadsheet. And it’s like, oh, I can’t build off of this because I can’t publish because the account owner is different than the spreadsheet owner. And like this weird stuff, I’m like, what are you talking about? So then I am asking it, where do I find this out? Who owns the account? It’s my account. I don’t understand. So anyways, I find out that it’s the leader of IT from our parent company is the name on the true account. And for some reason wanting to publish this up to date like app is different than all the other ways I’ve been using Claude and giving it drive links. So that’s been like my big stop gap. And then I was even asking it like, is it appropriate to ask your IT leader to give ownership of the account to me so that I can publish this? And I was like spiraling. 00:19:06 | Tim Wilson: Help me draft an email. Help me. I’m at that point. I’ll draft an email for you. 00:19:11 | Julie Hoyer: Yeah, but I’m really hoping I can do it because to your point with the librarian, I’m like, we have so much content here that people never go and look at, but there’s so much applicable stuff in there that I’m pointing people to it constantly. Like, oh, there was this one time we talked about it. Here you go. 00:19:27 | Tim Wilson: But you said early on, like, which I think I’ve now learned, which is a little bit of an adjustment for me. I’m used to like, oh, I know, I know how whatever this thing is works. So I sort of have the architecture and here’s how I’m going to plug in the pieces. And because I’m still learning, I’m finding myself often saying, this is the end result and pushing myself to like articulate the outcome that I want. And then I will sit there and say, now, I think I could approach it this way or that way. And, you know, you tell me, like, actually, I had a case where I think it’s pretty known out there, like fax and feelings is kind of drifting away. And so Squarespace was going to charge us a big old fee to re-up in a bit. And I was like, oh, well, we’ve already stripped the site down to just be like a splash page. So couldn’t I just it’s like, it’s just a simple page. Can I just have, you know, Claude helped me recreate it as a static HTML. I’ll put it on Netlify. It won’t cost me a thing, whatever. And so I said, hey, can I do this? Want to walk through it? What are the cases? And then it brought up a whole bunch of really good points about managing the timing on the DNS cutover that I hadn’t thought about, which then sort of gave me like a much better plan before I kicked off to do it. But then I also learned like having it vibe code, trying to look at screen captures and inspect an existing page built on Squarespace, which has all sorts of added bloat on it. It was I still want to be doing a lot of inspection and CSS tweaking. Although even in the middle of that, I kicked Claude, I kicked co-work off to go do some like dom scraping to to suss out some some settings for me. So that whole like, don’t just assume what you’d know, like the chat ones are really nice to say, what might be the options? And can I use, should I use, you know, Claude code for this? Or, you know, should I use co-work? And at times it’s come back and said, well, yes or no. And given the pros and cons. But when we talked to Rob Colley, a big thing he was pushing was like you, you should go and build apps. And it’s kind of funny being on the podcast with Michael, because I think we all would acknowledge that Michael, you’re kind of tend to be a couple of steps ahead of us, generally, when it comes to what’s the new things that are doing. So you when Rob was talking about that on that episode, I was like, oh, yeah, that’s what that’s what Michael did with, like, literally building an app for us to do our back end workflow for the show. Yeah. 00:22:16 | Michael Helbling: So what was interesting about that was obviously, like running a podcast, there’s all this stuff in the background that we do, which we don’t need to get into. It’s not that fascinating. 00:22:25 | Tim Wilson: No, let’s talk about it. 00:22:25 | Michael Helbling: I don’t want to document this as a very detailed process document. 00:22:31 | Tim Wilson: I wasn’t going to go here all night. It’s right. It’s all been transitioned into runbooks and co-work. Yeah. 00:22:37 | Michael Helbling: But but the thing was, is like the idea was, how do we both expand the capability for everybody to be part of that process, as well as take some of the stuff off of Tim’s plate? I don’t know that we’ve been effective at accomplishing either of those two goals. But the idea was to start a start a process and also like experiment with AI as a result of that. But a couple of really interesting learnings from that. One was I spent a lot of time up front writing documentation of like what I thought this app should be able to do. And actually was pretty unhappy with my design, if you will, or the architecture the first go around. So I was using chat GPT to kind of like kind of help me write those and like think it through. And it was just AI tends to do this thing where it kind of latches on to ideas and it’s real hard to get it to let go of them. And then it sort of gets all in the rest of the stuff and it gets, you know, it’s sort of like a what’s that thing where like when you’re cooking something, it changes state. So like an egg is cooked, you can’t uncook the egg, right? It’s so it’s sort of like that feels like that with AI sometimes. Like something gets in there and you can’t uncook the egg. It’s now like a different form and it’s just really in there with your ideas. And you just have to like tear it up and start over. So I was at that point and so then I switched over to Claude because I was starting to use that a little more and I was like, all right, this is a good first step with Claude. And I used us a tool to design it and using the design of it to do that. But the other thing I thought was the kind of interesting was it actually took me back into a learning process about specific things because one of the pieces of the app is an integration with our Slack our Slack channel in the podcast. And I realized like as I was thinking through like how this would work, I really didn’t know what was possible with Slack integrations really. Like I’ve seen some but not a lot and I’ve not done any development behind it. So like it was sort of a blind spot about what here’s what I want. But like I’m not sure how how much this is or is not should interact. And so working with the AI to kind of like even educate myself a little more about that was actually part of the process. So it’s sort of like you hand stuff off to AI to go do. But actually you’re getting educated a little bit along the way on this as well because while you kind of know like what you’re trying to get to, it’s sort of like, well, do I even know what’s possible? And that’s actually pretty critical learning because then that shows you like, oh, if I don’t know what’s possible, like I’m not really the AI is not really extending my capabilities as far as they might be able to go if there was a more experienced hand. And this comes out, I think, when you do other kinds of work with AI. I was chatting with a CEO of a company a week or two ago and they’re like, yeah, I really love using AI and it’s great. But I feel like it’s not really ever helping me to like innovate that next step. Like it’s not helping me kind of like go that one step further. And I was like, oh, yeah, I think I get what you mean. So AI is it’s really it was a great experience. The app works really well. It actually looks pretty good too. 00:26:04 | Tim Wilson: Can I ask about because it’s it’s a it’s a little bit of a mystery to me. Like I get the the coding part, but like ultimately, this is an app that is hosted on Vercel. Like what is the actual and you’ve done a few iterations because we have like sort of a feature request and bug tracker. And so I guess two questions. One, the actual I get when you have code, I’ve got it when I’m in our studio or I’m in VS code, or I’m just asking for it to generate and that’s code. But when you have what is it, what are you actually doing to get the code run and deployed somewhere? That’s question one. Question two is when you’ve made updates, is it discipline that it’s touching specific sections of the code or is it kind of going we in like regenerating from scratch? And this is like reading Ben Stansel a year or two ago and the way it was being described. So that’s a two-parter question. 00:27:09 | Michael Helbling: So I don’t understand the whole thing I learned at the hackathon. Well, and before that too, but the hackathon we did, which had better developers than me at it, was I learned more about how to use GitHub. And so I use GitHub pretty heavily. So basically when I do a coding project, basically my process is, and hey, listen, I’m not claiming anything here. So like this is totally whack and broken. Well, you let us know in the comments, but this is what I do. I start a local folder. This is where the code is going to live on my computer. And then as it starts to develop into something, I then go create a GitHub repository. And then that code starts to live on that repository. And so basically for small projects like this, it starts locally and then moves to GitHub. And then GitHub becomes the versioning and the control. 00:27:58 | Tim Wilson: And then from there, OK, I’m just going to put it, put it, put it a little, let’s just call out that Michael did not explain how Git works well on that. 00:28:06 | Michael Helbling: I don’t know how Git works very well. I just want to put in the qualifier. Yeah, no, I already qualified it. I was like, I don’t really know. But he didn’t say it might be whack. Yeah, exactly. He said it might be whack. 00:28:20 | Tim Wilson: And what I mean by that is it’s probably whack. But that’s the code. I mean, I published from GitHub. Yeah, two GitHub pages for like web. 00:28:29 | Michael Helbling: But how do you then when you when you start up? So then basically, there’s sort of the DevOps piece of it, which is how do you take code that lives in GitHub and is maintained there in a repository and make it an app out on the web or wherever. So that’s the DevOps piece, which is where you use something like we in this case, use Vercel and Vercel basically can go connect to your GitHub, get the code, deploy it. And basically any time you change GitHub, it automatically deploys to Vercel. And so you have an automatic deployment process. And so the Vercel interface stores all of your environment variables and all the stuff that makes all this work and all the API calls that go in and out. And so this app basically connects to resend for email. 00:29:15 | Tim Wilson: It is the database living and where’s the actual database 00:29:21 | Michael Helbling: where the database is neon, which is the got bought by Databricks. So yeah, that lives out there. And so basically all this stuff is sort of interconnected already. You basically when you write all this code, you’re basically connecting all that stuff via API keys and stuff like that. And then the app knows when to call. 00:29:40 | Tim Wilson: So Claude told you, OK, go make a Vercel account, enter these. Here’s how you can go make a neon account. 00:29:46 | Michael Helbling: And yeah, basically, I mean, you can say to it, like, I want to use this versus that, you know, so it might come to you and be like, well, this is an architectural decision. Do you want to use SuperBase? Do you want to use Neon? Do you want to use Convex? Like you can kind of say which one you want to use or you have one you need to use to say that. So it’s not like you can’t. You can just use whichever one. So like there’s either other versions like Vercel is one example. But I think there’s many others that do the same thing like rep lead. I think probably does the same things. So yeah, since I know absolutely nothing, I just went with whatever Claude said. So that’s where it lives. 00:30:22 | Tim Wilson: Well, so I mean, so that that’s helpful because I yeah, I’ve got things that are published to GitHub pages and to notify. But I’m generally it’s just like static stuff. Same thing if I if I do a new push to GitHub, GitHub is going to it’s going to detect that and it’ll redeploy and kind of refresh it. But I was like, but that’s always been like, that’s just rendering Korto or our markdown or Hugo. But so it’s but that’s really helpful because I was like, OK, but you’re extending it saying, well, there might be two or three different systems that all have to kind of play together. 00:30:55 | Michael Helbling: Yeah, in the case of that app on the back end, there’s about five or six systems or API calls that go out. So like we have API calls out to different LLMs to process things within the app. It reads, you know, we have an incoming email and I look through those to evaluate, like, is this a good pitch or is this not a good pitch, you know, those kinds of things for us. So those are all things that are happening in the app. And then the email processing itself goes through a company called Resend and, you know, so on and so forth. So there’s a bunch of those little things. But basically, all of those decisions get made back at the beginning when you’re defining what is this thing is going to do and how does that work? It’s like, well, here’s what happens. We get topics from the website. We get topics from emails. We get topics from our Slack channel. Now we need to bring those all together in one place so we can evaluate those topics and and make decisions about them. 00:31:49 | Julie Hoyer: So what you’re saying is my earlier app library for this guild is laughable because I have a lot of steps I’m going to have to still go figure out. Well, I wasn’t even thinking like, oh, I’m going to have to probably connect it to GitHub and do this. But like, I don’t know, the example you’re saying is giving me thought of like, I have quite a few pieces to put together in this puzzle. 00:32:12 | Tim Wilson: There’s a nice little nice little fable anecdote that you have when like you asked didn’t rather than it was a fable. What is it? Yeah, fable that you asked to say, hey, could you evaluate this? 00:32:22 | Michael Helbling: Is this kind of pretty low level authentication path? Yeah, when fable five came first came out, I had to look at the code base for that app. Not with anything. I was just like, I’m thinking of making some changes. Could you evaluate this code base and give me your thoughts? And so fable is pretty well known as being like a pretty high end AI or like pretty state of the art right now. And it immediately identified three security flaws in that app, which we then patched immediately. So it was like, oh, yeah, well, before we do anything, here’s three things that are broken about your app. I was like, well, your little brother is the one that coded it. 00:33:01 | Tim Wilson: So we escalated that because this is a podcast. You can bring the podcast production process down. 00:33:10 | Michael Helbling: Bring it to a screeching halt. I don’t know that was actually what was the problem. But there were like a couple of things that definitely should have been fixed and we did fix them. But but you know, that’s that’s sort of an emerging thing. And also one of the things like, you know, I’m over here doing experimentation with things like Hermes, which is an agentic AI that kind of is like doing ongoing AI tasks for me on different things. And like this is actually an area of big concern, which is the security piece of that. So like setting that up is I’m super cautious about that because like that doesn’t live on my local network. It lives far away on an external box that is not attached to me at all. It cannot access certain things. It cannot write things in certain places. 00:33:56 | Tim Wilson: So basically I keep it pretty pinned down for now 00:34:02 | Michael Helbling: because I just don’t as I learn, I don’t want to be exposing like my work and all these things to an agent that may then just go dump all that information on the first, you know, random hacker that gets the opportunity to kind of prompt inject it with something that it thinks is cool. So like there’s some real risks out there with AI, too, which you know, I’m trying to be cautious of, but still also experiment around and have a lot of fun. 00:34:30 | Tim Wilson: Yeah, I mean, it’s what it’s the luxury like it is kind of nice sandbox. We’re not having a crunch a bunch of data, but there is a lot because what I tackled was I said, oh, we have this lengthy 40 page document of our process. That should be something I have a background as a technical writer. I’ve kept it maintained, throw it in and let co-work run with it to do everything because we record a little bit of stuff happens. Things go off to Tony to do the editing. He sends back like a final file. And that’s like the starting point for a whole bunch of stuff that has to happen. And that’s documented in the steps. And we had several of us who were like, we can do this process. I do it most of the time, but we had a couple of other people with fallback. It’s a Google doc knew how to do it, threw it into co-work. And a couple of things have emerged like one, how many steps there are that the human in the loop, partly because, oh, you got to go click somewhere that it’s just not going to go click on its own more. So I need to check in and have you, you know, pick which tags we’re going to put on the show. I’ll recommend these or these the right ones. Or these are three options for the show bar because frankly, I don’t think the show we’ve written enough by hand. We have enough training. We can now use it to write that copy. And it is now a the two things that have kind of emerged. One, what used to be an hour, an hour and a half of focused time that was not enjoyable at all and subject to to me making mistakes by not completely following the process. Now it’s like an eight hour eight hour duration because it’s super slow cranking through some stuff. And I just have to wait to hear the little beep and then go in and check, you know, what I need to rely on. And for some reason, I’ve never been able to get caught dispatch to work because I was like, this is the perfect application of I’m going to go meet someone for coffee. This thing’s running. If it just needs me to say go, never could get that. I have not gotten that to work on my phone. So one that was interesting to say, yeah, this is kind of a grind of work to do, but it’s stuff that a human has to do. So that was kind of validating. The other is it’s wild to watch it take different paths to do the same task each time. Like there are times where it’s like, well, that didn’t work. So I’m going to try this other thing. Well, that didn’t work. And I watch it be super inefficient as it finds its way to do the thing. So it still has that it has that probabilistic LLM thing going that it’s not a rules based. It’s more robust. So if a UI slightly changes, it’s probably going to figure it out. But on the other hand, where exactly it sort of runs into problems is kind of a moving target each time. So you could do Tim, as you could write a skill for it. 00:37:40 | Michael Helbling: So you could give it. That’s good. Yeah, you can give it the steps. 00:37:43 | Tim Wilson: Well, but I’ve given it the steps in code work. Like it has a run book that is like, do this, then do this. I mean, yeah, but that’s very detailed. 00:37:53 | Michael Helbling: That’s different than sort of like process steps. I don’t know, maybe not. Maybe it’s not a good application. I don’t know. 00:37:58 | Tim Wilson: I’m just actually I should probably go to Claude chat and say, I’ve done this and this is the challenge I’m running into. And how could this be? But is this something skills would help with? I love doing that when I don’t understand something like, hey, would skills work for this? And it’ll come back and be like, yeah, you ding dong. Like, why weren’t you why weren’t you using skills from the get go? That’s a good skill segue because you’ve been playing around with skills. 00:38:27 | Michael Helbling: Yeah, I yeah. Well, because yeah, because in a couple of cases, our clients are using AI and so we’re helping them do that. And one of the things I noticed when AI when LLM’s work with data, they tend to sort of fall victim to some very common pitfalls when doing analysis, like certain types of analyses. I think they do different things and it’s sort of challenging to kind of get them to not do that. Like one of the things I noticed right away was like comparing two things to try to evaluate what was different between these two things. It would sort of come up with its own idea, like sometimes they would just make it up and then it would just sort of hang on to the ideas if it was like totally rock solid, that’s the answer. And I would go back and be like, no, that’s not the answer. That’s something you made up. And we got to throw that away or just hold that to the side for a minute. Maybe it’s the answer or maybe it’s not. We don’t have enough data to say that’s the answer. And of course, you know how AI is. It’s like, oh, yeah, you’re absolutely right. Like, I’m definitely not going to do that again. And then I would just go do it again, you know, and it would just do it. And so I was like, OK, I’m out. So that got me thinking like, well, what if I gave it a better set of instructions? And so I started building out a set of skills for different types of analysis to sort of say, OK, if I was sitting down to do this analysis, how would I break this problem down? There is a bunch of great literature and technique and methods all published like there’s a ton of material that it could be drawing on and it just isn’t doing it. And the other thing I noticed is that all of the data skills that are out there, a lot of them have to do with like accessing tools or making a PowerPoint presentation or making a visualization. It’s like, OK, that’s all very handy. I’m talking about like just doing the analysis of the data. So I’m working through a process. We’re still very early stages where we kind of I built some skills for that. We’re doing some testing. What I’m learning in that process is two things. One is using those analysis skills is producing a better result on average than just the raw LLM. And that’s pretty cool and very awesome. But doing evals on skills to prove that they’re working the way you want them to work is a ton of work. It’s like so much work. So like it and you have to be super structured about it. And you have to be like you have to make sure like the prompts are set up correctly and everything is done the right way. So like you can actually evaluate like the outputs and know that it’s like got it right. So like it was still a long way to go, but I’m hopeful that over time 00:41:16 | Tim Wilson: we’ll actually develop something that’s going to be at least convenient, 00:41:20 | Michael Helbling: if not useful in some cases. I think the challenge there is that at the end of the day, I don’t think an LLM can actually produce the quality of analysis or meaningful analysis that I would then go and say, like this is what you should do instead of having humans do it. I just don’t think you can replace it. But I know that humans are going into AIs every single day and asking it to analyze data. And in light of that, maybe we should try to build something that makes it less bad. So in the old George Box framework, you know, all models are broken, but some are useful. That’s sort of what we’re shooting for here. 00:42:02 | Tim Wilson: All models are wrong. All models are wrong. Yeah, there you go. The Mechershop just got triggered because he’s like whatever it is, you’re not using it right. Yeah, yeah. Well, I’m trying to use it wrong both ways. That was on purpose. 00:42:15 | Julie Hoyer: Wait, can you share like what type of analysis you’re talking about that like one of the ones you’re working on a skill for? I’m just curious. 00:42:23 | Michael Helbling: You cannot use EDA. Sure. So root cause analysis or like a or a prescriptive analysis or so like different types of analysis. So like it was sort of a list or a library of like a bunch of them. And yeah, some of them are ones that Tim is not very happy that I’m doing. 00:42:44 | Tim Wilson: Metric, metric reconciliation, metric reconciliation, question, 00:42:48 | Michael Helbling: refinement, yeah, causal, you know, inference or impact. So like all the ones that like in a business context, there’s a better level. But most of the time people don’t put the work in to do. It’d be nice if your LLM was like, hey, here’s what the data is saying. But I think the next step here is actually some kind of holdout test to really prove this. It’s instead of just knowing that threshold. Yeah, instead of just telling you what you want to hear, which is what the LLM would do without the instruction because it picks up on things like the way you ask the question. Like if you’re asking a question with an idea that like, hey, can we stop spending Facebook spending money on meta this month? It’s going to pick up from that question that like you’d like to stop spending money on meta this month. And that was going to do the analysis and come back to you with something that makes you happy, whether that’s good for the business or not, because there’s underlying that is this instruction set that the LLM is doing on its own, which has nothing to do with your analysis and more about how it wants to make you happy. And that’s sort of like what we’re fighting against a little bit because when you do analysis, you’re not trying to make someone happy. You’re trying to get to a closest version to reality that the data will support. So it’s it’s a little bit different exercise. And so it takes the LLM instead of telling it like now behave. It says, hey, here’s the literature. Here’s the methods. Go through these steps and use this as a floor, as opposed to like how you’re doing it the other way. 00:44:25 | Tim Wilson: When I feel like if like, oh, you’re asking to do a root cause analysis, the skill of the root cause analysis, say, I’m going to go do a quick exploratory data analysis of this before, like given the time and the capacity, you should always check that the data doesn’t have nonsensical stuff in it. Where it’s super easy, super common for somebody to say, we’ve been using this data forever or somebody handed it to me and told me it was clean. And so they just go straight to, you know, running a regression. Whereas some of that and I could also see it qualifying, requiring some qualification of the, you know, from the user. But that’s that’s been my hobby horse that I think is valuable to they ask a question and they get a very polite but reason set of focus and focusing and clarifying questions back. 00:45:17 | Michael Helbling: Yeah. And not always sort of like, like an easy answer. 00:45:22 | Tim Wilson: But like, OK, well, there’s more work to do to actually get the answer you want. And it’s like, OK, well, no one’s going to accept that. 00:45:30 | Michael Helbling: So we’ve got to tune it more. 00:45:33 | Julie Hoyer: Funny enough, I not on the list of types of analyses you mentioned, but I guess I would I would count it as a type of analysis. I had pulled together a proposal for a client and long story short, there’s a bit of like a pause. And so we wanted to send over the proposal for our direct stakeholder and her boss to understand like, hey, when things change on your end 00:46:01 | Tim Wilson: and you can revisit this, like here’s here’s what we’re proposing, 00:46:06 | Julie Hoyer: valuable to outline that story. And we wanted to close it out with like an impactful value slide. Right. So me as the strategist on the account is in charge of that. So I mean, we’ve done the exercise manually. I have people at the company I was talking to Matt Padone about it. You know, ways that he’s gone about pulling something like this together. And so I didn’t have a lot of time. And I was like, you know what, I’m going to try to use Claude to help me here because really what I needed to grab and make some assumptions about what this program could help drive for the business for, you know, leadership that we were trying to talk to was like public financials. Right. And like, that’s a slog to go through on your own. So I’m like, oh, I’m going to give it an example of information we’ve pulled before when we try to pull together these like value estimates. And I pretty much gave it context to say, here is who we are working with. Here’s the situation. Here’s what we’re proposing. This is the area of the business that we would be working with. This is the impact we believe that we could drive for them through this work we’re proposing. And it was maybe a long paragraph, like a couple of run on sentences, nothing crazy, gave it one link. And I asked it to help pull public financials to help me make some assumptions and give a value estimate. And it spit out a six tab spreadsheet, which I was a little like, wow, overkill. That’s what I’m also finding. I’ve asked it for like one slide before gives me like seven. I’m like, whoa, all right. But it did a pretty good job. I mean, I was able to go and find to like where it was pulling numbers from. It did automatically give me pretty descriptive, like this is where this number is from. It highlighted cells that it was making assumptions on. It gave me like industry benchmarks that I didn’t actually ask for those explicitly, along with the client’s actual published numbers. And then I was able to work through it once I understood the spreadsheet set up, though, and play with how I wanted to craft the narrative. But it was interesting because I now know the next time I go to do this, I would prompt it in much more detail because, again, I have a better idea of like how I want the spreadsheet to function and the things I want to be able to like input. So I do think if I get more explicit or even dip into like what you’re saying, Michael, and build like a skill around this for it, that it could be a really super useful tool. And then I definitely have to work on the visualization of it because it needs the human touch there for sure. Tim was helping me today. I slacked him out. I was like, I’m running myself in circles. Like I just need some a clean brain to help me edit this story to like what the hell actually matters. So it’s been good, though, for that type of thing. 00:48:51 | Tim Wilson: I had a I mean, I worry like the when you talk to the like it goes it goes bloated, it goes along. And then you look at it and it’s really hard to fully wrap your head around. And it was interesting, even in that the spreadsheet, there were cases where it had just like hard coded done the math and hard coded it. So it wasn’t even like a traceable model in some cases. It was in some cases. But I I mean, that that continues. Like the higher level, like it’s so fast to generate content that’s polished and professional, but doesn’t have. The humanness in it. And I think in a in a proposal context, like you’re somebody’s going to be looking at it and they’re going to be asking questions, regardless of what it is, not your specific example there. I had a case where I’m doing some training, working with another company. And for various reasons, I’m developing the presentation in Horto. It’s very specific to the audience. But I also am pretty particular about I don’t want things to look great. And it turns out when you’re trying to use markdown, I mean, I’ve run through this before. But I worked with the people who were kind of co-presenting and we had kind of a rough we’d iterated through a Google doc where we really sort of figured out the narrative and the story. And that was like collaboratively. And then I went through the Google doc and put a little more meat behind it. And then I was like, what happens if I pop over to fable five and say generate me the I did this in cloud code. I already had a repo set up for it and had pulled it down locally and said, give me the shell. And that was really useful as like the raw because I fed in the this is the sequence. These are the specific slides and they still look like absolute, you know, garbage, like I’ve spent hours after that needing to iterate. But I still had that skeleton in place. So I wasn’t having to start one slide at a time. I have the whole shell, but it came from a human human massage, developed the narrative and the talk and the points and the highlights. But it was it was another kind of useful cloud code experience in that case. 00:51:16 | Julie Hoyer: I see cloud like decks that get put together and it’s like a standalone slide. It’s pretty good. And sometimes they’ll pick up on like this, maybe the main story point you wanted, like throughout this, like again, like bloated long deck that it will put together. But you know, it’s funny to my always go back to like just the basic McKinsey title idea. And it’s like, I need to start telling people to explicitly tell Claude or whatever I’m there using to be like, use McKinsey titles. I think the outline from a human before asking it to make the deck should be like, this is the story bullet points in this order that I want you to tell. Use it for McKinsey title style and say like this needs to make sense. Like top to bottom, because when I get into a deck that is like 00:52:02 | Tim Wilson: Claude generated, for example, it’s like, OK, generally, this slide makes sense. 00:52:08 | Julie Hoyer: I kind of understand what they’re saying. But to me, it doesn’t seem like there’s a good thread between and it’s like verbose and it’s repetitive. And then it’s really hard for me personally to figure out what is the point of this individual slide I’m looking at. 00:52:21 | Tim Wilson: But that’s not that hard to tell to tell Claude to say, you must walk me through. Make sure you have a solid outline. I’m going to I’m going to define the steps of how we’re going to do this. You’re going to make sure I give you McKinsey titles. I did have a rant a like a month or so ago on LinkedIn because I was literally I should remember what I was working on. And it said, I have a hand I have handcrafted a bespoke, beautifully structured five slide presentation and HTML and CSS. It uses clean, high impact management consulting layouts like McKinsey or BCG decks designed specifically for your topic. And I’m like, this is horrible. So that’s rant one. Like I was like, yes, you this is what you think is great. The other like just. Boogaboo, I was trying to generate HTML presentations with like our markdown way back when had one specific client where once every week during COVID, we had to pull all these data sources together and generate it. I’m like, I can make all this happen with with in our and pull together something that matches your template. It’s just you’re going to send your ultimate going to send us a PDF anyway. And I just got hammered. It’s like, it’s got to be in PowerPoint because we got to be able to edit it. And I was like, I understand the the the motivation and you need it. Whereas now everybody’s like, I can just make it an HTML. And I’m like, fuck you. Like, like it wasn’t good enough when I wanted to make something in HTML because it wasn’t just being spit out, you know, generically, it looked fine. So yeah, I’m sorry. 00:54:04 | Michael Helbling: Well, it’s weird with all the things I’m willing to let AI do for me. The thing I’m least willing to let it do is produce content or communication. So like, we even have at Stack Analytics now kind of a standard, like we will not let it write our emails. I like that I won’t let it write decks like technical documentation. I think it’s actually OK at and can produce good quality materials. But like anything where I’m attached to what the person is reading 00:54:33 | Tim Wilson: in terms of communication, I just don’t think it’s there yet. 00:54:37 | Michael Helbling: Like and I’m training it on my voice and those kinds of things that I keep testing. But I I’m just not comfortable with it. And I just feel it. It’s sort of like Stevia, you know, that weird aftertaste. 00:54:52 | Julie Hoyer: That is so good. I just think that’s exactly what it is. 00:54:56 | Michael Helbling: It’s just I don’t get it and I don’t like it. And so if I don’t like it, then I assume other people probably don’t appreciate it either. So it’s sort of like, all right. So then the least we can do is like, it’s going to be us typing that email. 00:55:07 | Tim Wilson: I have I have given it like content. I’m like, I need a McKinsey title that describes this, this, this and this. And if you take it, then I’m editing the edit it yourself to edit it. 00:55:19 | Michael Helbling: Yeah. Yeah. I think that’s totally fine. It’s more of like something that spits out. You just paste it into your email and fire it off. 00:55:27 | Tim Wilson: Like I don’t like it. 00:55:29 | Michael Helbling: Well, I don’t. It’s not good yet. 00:55:30 | Julie Hoyer: Once it does like a deck for you again, we talked about this in past episodes, like you become the editor of work that’s in front of you. But sometimes like it’s so much easier just given that like, it’s good enough because I didn’t think through what it needed to be already. Yeah. Because you didn’t try to solve it. Now you’ve been biased by what it spit out. 00:55:48 | Tim Wilson: And I will say when I’ve I really only a handful of times 00:55:53 | Julie Hoyer: have had it made any type of slide, but I have heavily edited anything that it gives me. Because to your point, Michael, I’m like, I know I can structure a better slide that will get the point across. But sometimes I do need like, can you just give me a couple ideas here? And like, pull what I like and I’ll mash it together. 00:56:10 | Michael Helbling: That I think is just fine. That’s just fine. It’s more of the other side of it where it’s sort of like you just turn your brain off. And it’s like, is that good enough? I don’t think that’s good enough because I could anyone could do that. The client could do that. Like as a consultant, like I hope we’re bringing more to the table than just regurgitating whatever Claude said. Yes, you know, ideally a BCG or McKinsey style email. Well, you know, the problem is sometimes when I write my own version and I’ve seen what it writes, I’m like, my version sucks. But at the same time, it’s like, but it’s yours. Like, it’s me. It’s me. So, you know, everybody just bear with me. I’m not as smart as an AI, but, you know, I’m trying. All right. I love it. One last thing I want to cover off on, because it was a big kind of learning for me is obviously using these LLMs for me has been a lot of fun and very learning experience. One of the big things that happened recently the last six months with with us is going from an individual using Claude to do stuff on a certain project to a number of people needing to engage with that same thing. And like, how do you share the information and context and those kinds of things? And we all like have we’re on data background. So like in data warehouses, everyone’s like, oh, you got to have a semantic layer. So the AI knows how to do all the things it needs to do. And so that’s been a very interesting challenge. I don’t claim to have solved it, but we’re very much looking at that too, which is keep iterating on this idea of how do we deploy context? Because there’s like so many different little decisions go into when you’re setting up data, like, OK, we’re going to tweak it just like this so that these things match up and we’re going to make this decision at this point to kind of normalize the data like this. Like all of those choices are filtering down into then other people making it harder to follow in your footsteps. And so it’s interesting because we’ve done some various things to sort of share that context more broadly. But that’s something I’m keeping an eye on in our space, especially in the data space, because like I think that’s really common. Is sort of like, how are people making this a team sport versus an individual effort? Because I think that’s a that’s a pretty big, pretty big deal. 00:58:41 | Tim Wilson: A little bit of a callback. I started to say we had this document just for the podcast. So it’s not on the data front. And I set this up in co-work. It was somewhere along there that I realized that, oh, the runbooks are stored locally, which I’m like, that is terrifying. Now it’s going into a folder that I have syncing somewhere else, but I’m much more comfortable with stuff being in like defined cloud, whether I’m syncing to GitHub with Git or whether it’s Google Cloud or whatever. But I started to say that we had something that I could hand off. If I wasn’t available, somebody we had backups that could follow the process and it was fairly safe. And now somehow with co-work, I’m like, can I hand this off? Like, like, I don’t know that in theory, I’ve handed it off to this agent, but I can’t figure out how I could safely hand it off, especially knowing that I will hit things and say, yeah, I need to update the task and the instructions here and there. So that’s that this is similar. Yeah. World that you’re like, oh, this supercharge is you, but it doesn’t necessarily supercharge the team in the way that it should. 00:59:51 | Michael Helbling: And I think that’s the big I feel like that’s the big lever or tipping point to value for AI and a lot of organizations is sort of like, you’ve got one or two people kind of doing stuff over in a corner. It’s making them real productive. But like, how do you create a process for AI usage that makes everyone that much more productive and step up the whole org with AI? Like, this is actually the whole part. I think the big part of it. And it’s actually a truly complex, I think, very deep problem that I don’t know that I’ve seen anybody really solve well. And given your recent LinkedIn post and the feedback you got on it, like I see a lot of other people, especially in the data space, because like it’s not a technology problem at the end of the day. It’s actually like a human process problem. And how you solve it goes into your culture. It goes into your process. It goes into all these different things. So anyways, that’s a fun one to kind of unwind, but also kind of like the frontiers that we’re finding and using AI the day to day is sort of like, oh, I did all this cool stuff. Now, how do I hand this work off to my teammate who now or my client who now needs to take this stuff and go one step further with it? Oh, what do I do? So that’s what you’ve got to find the answer. So we’ve built some things to do that. But it’s it’s I don’t know that it’s like it’s not perfect. It’s not perfect at all yet. And I don’t think with AI you can be perfect. 01:01:27 | Tim Wilson: So you just have to kind of be adaptive. 01:01:32 | Michael Helbling: Anyway, all right. Any last words, final thoughts? Oh, we’re like like we’re fine. 01:01:38 | Tim Wilson: Like it, like like the LLMS where we’re running along. Yeah, it’s right for both. All right. Well, then let’s have a Michael. Yeah, that was an excellent point. You did a great job. This has been every point you made was just so spot on. Sorry, I’m being like, I don’t appreciate that at all. Tim, that was the kind of bluff that I don’t appreciate. 01:02:05 | Michael Helbling: You’re absolutely right. You’re absolutely right. That’s the load bearing comment of the podcast. Tim, great job. All right. I bet as you’ve been listening to this episode, you’re like, wow, I can’t believe how terrible the Inelix Power Hour is in AI already and you’ve got thoughts and comments. Listen, we’re open to learning a little more. So this one we do actually want to hear from you. Like tell us how you’re using it. Talk about what you’re doing with it. Like we want to know, at least I do. So you can reach out to us. Do that on the measure slack chat group or on the LinkedIn or via email. At contact at InelixHour.io. 01:02:45 | Tim Wilson: And yeah, check in with us. 01:02:48 | Michael Helbling: Tell us how you’re using AI in your day to day. I think it’s a pretty wide open space and we’re all learning from each other, which is a lot of fun. It reminds me of the early days of analytics, to be honest, where we’re kind of all just sort of comparing notes. 01:03:01 | Tim Wilson: And back then it was, I remember that time with like web trends and like the report tables and the analysis tables and how you could, you could call up one turn and get it. Remember, remember that? Sorry. 01:03:19 | Michael Helbling: And you can also get stickers because nothing says you’re cool, like a sticker on your laptop or water bottle from the analytics power hour. And you can request that on our website at analyticshour.io. And I think no matter how you’re using AI or even how you’re using AI to listen to this podcast, I think I speak for both of my co-hosts, Tim and Julie, when I say, don’t stop analyzing, even if the AI offers to do it for you. You’ve got to stay involved. 01:03:55 | Announcer: Thanks for listening. Let’s keep the conversation going with your comments, suggestions and questions on Twitter at at analytics hour, on the web at analyticshour.io, our LinkedIn group and the Measure Chat Slack group. Music for the podcast by Josh Crowhurst. Those smart guys wanted to fit in. So they made up a term called analytics. Analytics don’t work. 01:04:19 | Charles Barkley: Do the analytics say go for it no matter who’s going for it? So if you and I were on the field, the analytics say go for it. It’s the stupidest, laziest, lamest thing I’ve ever heard for reasoning in competition. 01:04:32 | Michael Helbling: Say it all, you never know. 01:04:33 | Charles Barkley: Tony, do not put that. 01:04:35 | Michael Helbling: Do not put that on a date. 01:04:38 | Julie Hoyer: OK, wait. Ready, Michael? We’re going to self-diagnose your head pain. Why? Hang on. Hang on. Just trust me. Good. It’s good. It’s all right. Good. Fine. OK. Window. Here we go. Ready? 01:04:52 | Tim Wilson: Yeah. 01:04:53 | Julie Hoyer: So these are the different types of headaches. Is it a hypertension, a migraine, a stress, or is it a data sits under I.T. type of pain? 01:05:06 | Tim Wilson: That’s good. 01:05:10 | Michael Helbling: It’s kind of a data sits under I.T. 01:05:14 | Julie Hoyer: It’s got that kind of feel to it. 01:05:16 | Tim Wilson: Yeah. 01:05:17 | Julie Hoyer: My guy was literally sitting with a sinus infection last week. So I was like, oh, before I knew it. And I was like, oh, it’s like right behind my eyes, but low. And so I see this on LinkedIn. I mean, I should have known. And I’m like self-diagnosing myself with it until I got to the end. 01:05:33 | Michael Helbling: That’s good. Yeah, no, it’s the migraine one. But yeah, that’s pretty funny. 01:05:40 | Tim Wilson: That’s good. 01:05:42 | Michael Helbling: Yeah, I need to keep that in the back pocket because that’s that one keeps coming up. 01:05:49 | Tim Wilson: Yeah. It’s an episodic recurrent. Yeah. Recurrently episodic. No matter how much tile and all you take. OK. All right. Are we are we doing last calls? 01:06:03 | Michael Helbling: Um, I don’t think so. I’m picking up I’m picking up something from our. 01:06:13 | Tim Wilson: Moere producer. I don’t know. It’s comical for next week. Yeah. The Venn diagram between what Julie reads and what I read is. 01:06:23 | Julie Hoyer: Well, who do you think, you know, sent me all those newsletters? 01:06:27 | Tim Wilson: Yeah, I think this is well, I do have a kid who’s, you know, planning trips out in October and he’s super laid back and chill and he doesn’t get bothered if somebody like can’t make a decision. He’s like, well, here’s the deadline. And if you haven’t made it by then, you’re out. And he just moves on with his life and like, if I could get his attitude just to else, I would probably live ten years longer. Due to the hypertension that is not developed. 01:07:01 | Michael Helbling: Data still sets up right. He. There. Yeah. All right. All right. Let’s do this. Here we go. This is all right. Energy, energy, energy. 01:07:15 | Tim Wilson: Me, my mom, my mom, my mom, my mom. You know, the local folks jumped over the lazy dog over the lazy. Yeah. Rock flag and stevia. The post #304: I Can Haz AI? appeared first on The Analytics Power Hour: Data and Analytics Podcast .
