Micah Wheeler — How a CPA Uses AI to Make Better Decisions Faster

Transcript

Micah Wheeler: I am not personally, like a techie, like a doer, a coder, or anything along those lines, but it's been about a year now that I've been going to the LA-AI meetups on essentially the premise that AI is the biggest thing going in the world. Why would I not be in the room where everyone's talking about it.

Narrator: This is AI with Kai. Practical AI for real business leaders. We break down what matters, talk with people using AI in real businesses and share one idea you can use right away.

Kai: As always, today is another great day to talk about AI. I am your host, Kai Gray. I am here with producer Tim Scott. How you doing, Tim?

Tim: I'm doing great, man. I'm excited to learn some new stuff today.

Kai: Awesome. Yeah, a lot of stuff going on. We're here to help you and your business understand what's going on with AI and answer any questions that are plaguing you, and that's a common thing for businesses right now.

Narrator: The Rundown. Five AI stories you need to know right now.

Kai: Recently Anthropic released their new model. They released it to a set of companies. The new model's called Mythos and this is a very powerful model, and it's so powerful that it was able to find vulnerabilities in existing software that no other model has been able to find. And so what they've done is they have brought together a consortium of 40 companies to give them a preview of it so those companies can work on patching their own software. And so this gets into new territory for us,

Tim: like Skynet,

Kai: well, hopefully not Skynet, but maybe closer to Skynet than we've had before. You know, I think Anthropic is a little worried about it. A lot of skeptics look at this and say, well, you know, this sounds like marketing hype. When ChatGPT first came out, ChatGPT 2 had this same messaging behind it. OpenAI said, oh, it's too advanced for us to release to the general public. And in fact, it wasn't, we seem to have done fine with it. But this one seems real. There's a number of stories around Mythos, that it was able to build its own software to get out of its own sandbox and find vulnerabilities in software in seconds that people had been working on for years. And the real concern here is a lot of the software is being used to run critical infrastructure. The bigger point is that this is a bit of a threshold that we've now crossed, right? We're now into dealing with models that are so advanced that we have to really think about whether they get released to the general public for fear that bad actors are gonna use these to do bad things. In other news, if you're a user of Slack, you just got a bunch of AI tools included. And Slack is the tool that Tim and I use to communicate. Now, unfortunately, this is only for paid Slack plans. So Tim, I think we might have to pony up and start paying for our Slack. Because right now we use the free version. But if you're on a paid Slack plan, you just got 30 new AI tools, including things like daily recaps and a number of other things. So if that's you and your business, congrats, you now have that. Visa just announced their agentic commerce. What this means is that Visa is getting prepared for AI agents to make purchases on behalf of consumers. So if you sell something online, your buyer may not be a human anymore. It may be an AI agent with the power to buy things.

Tim: Wow.

Kai: It's pretty crazy.

Tim: I was just thinking about the time I called Amazon and I was like, hey, you know, my kid bought this. Can I get a refund? And now it's gonna be, hey, my AI agent bought this.

Kai: That's right. You know, I imagine that's gonna have to have just tons of guardrails around it. But yeah, I think we're quickly getting to the point where you can direct an AI agent to go out and purchase something for you. If you are a Microsoft 365 user, the new Copilot came out with a new pricing tier. So I think that this is important for you and your business, particularly on the enterprise side. Things just got a little bit more expensive for you when you go to renew. What you wanna think about is, is the AI upsell worth it for your business, right? Are you using Copilot enough to pony up more money for it? Next up, Google Vids AI video creation upgrades are coming. AI generated avatars and video creation inside of Google Workspace is going to make it much easier for business owners who are doing their own marketing. This is a big upgrade for Google. Google's had really good products on this side. Certainly the image creation stuff like Nano Banana is, I think, the leading image creation model right now. A lot of people are using it. Google also has something called Veo and Veo 2, which are their video tools, and those were standalone things, and they're starting to bundle those in Google Workspace. So if your company's a Google Workspace user, go look for those things. You can create film quality video with Veo. And so if you're interested in making videos for your company for marketing purposes, this is definitely the way to go.

Narrator: The Spotlight. How real businesses are using AI.

Kai: Today we're sitting down with Micah Wheeler, partner at Wilkins Miller. Wilkins Miller is an award-winning CPA firm on the Gulf Coast. Welcome to the show, Micah. Thanks for joining us.

Micah Wheeler: Thanks for having me, Kai.

Kai: Before we get started, Micah would like to give a quick shout out to his wife.

Micah Wheeler: Yeah, Hannah. I hope that you're listening. If you're not, I can't believe you're not listening. This is for you.

Kai: That's priceless. That is great. I'm guessing that's a long running joke you have with your wife.

Micah Wheeler: I don't think it's a joke.

Kai: Just on your end. That's great. So let's start things off with a little bit of background. Tell us about yourself, what you do, and a little bit about Wilkins Miller.

Micah Wheeler: Yeah, my name's Micah. I've been with Wilkins Miller since I graduated college or shortly thereafter. There's a little background of where I was in a rock and roll band. I'll go ahead and throw that in off the start just because, you know, the glory days. You gotta start with the glory days. And so the firm has evolved a lot over the years. These days we're somewhere around a hundred strong. The work that we do is aimed at being, you know, full service for any needs of a business with regards to their accounting. Also have wealth management, so try to be a one-stop shop.

Kai: Cool. What do you see happening with businesses or the accounting field at large that's notable to you?

Micah Wheeler: Yeah, so over the last year and just going and seeing how other people are using it, that's sort of been the education that I've wanted to understand. And just listening, seeing what's happening in the world. I'm also, you know, researching softwares from the accounting industry on a regular basis and seeing what they're doing. And combining those things together, just trying to understand, okay, what can be done? And when I have an understanding of what could be done within the world, ask the next question of should I build that or should I buy it? And then with the buy process, sort of being patient because we've all seen how fast everything's moving, how quickly the software world is moving. And so why would I be, unless there's something broken today, I'm a little bit hesitant to be in too much of a rush to buy because if I lock myself into some sort of agreement, is that product gonna be the best product six months from now?

Kai: Do you think that sort of the fundamentals have changed at all?

Micah Wheeler: Businesses and their owners still have to make decisions. That's really, you know, the businesses that win make good decisions. The thing that's changed, I would say, has been the tools that are available. And you know, different people make decisions in different ways, but I'm a believer in having good data and good reporting and information that can be used to help those decisions. And so a question that I ask businesses all the time is, is your data ready for the software of tomorrow?

Kai: Assuming you mean AI,

Micah Wheeler: right? And I mean, just what is it gonna be? It's hard to say, but I know that these large language models are proving that they're pretty good at, you can ask them a question about any subject and they can give you some good information. And a year ago and just being plugged into watching how these things are used, there was a lot of hallucination. It still exists. Man, it's a lot better these days. And so as those machines get better and they can analyze and they can see all of your records and see how everything's tied together and see your agreements, and then tie that to the industry that you're in, and understand how your transactions with vendor A and vendor B compared to, you know, the industry, and put it all together, in essence, it's hard to articulate, but if you have your data organized in such a way that the machines can read it, then I think you're gonna be better off than the organization that doesn't.

Kai: Yeah, totally agree.

Micah Wheeler: I'm very pro-human in all of this.

Kai: Yeah.

Micah Wheeler: But the reality is that if I go back, you talk about how the accounting industry has changed. If I go back just a few years, if I had some sort of scenario that was running through my head and I needed to think about where do I start in thinking about this, what I would do is I would walk over to one of my partner's office and we'd start brainstorming. That's not what I do anymore. I brainstorm with the machine.

Kai: Yeah. We're gonna get into sort of how you use AI. So this is a good segue, but you're using it sort of as a thinking partner,

Micah Wheeler: right. If I think about research, if I have a question that comes across, certainly there's all the codification and all of the rules that exist within there, but that's not normally where I'm gonna start. Back in the day, I would just start with Google and I would, you know, put some things out there and I would try to get a broad landscape of where things were going. I knew that my question would live somewhere in this nuanced piece of the question. And so in order to get there, what I would do is I would probably read 10 different articles that said 90% of the same thing until I found that 10% that I was looking for that was different. And then I would take that starting point and I would probably then go into the codification and get deeper into the hole from where that was. Today, I don't have to read those 10 articles because I just have a conversation with the machine. And the machine, once it is telling me the thing that I want it to say, I say, give me the source information of where that goes from. And I find myself directly. So the path from the initial question to the answer is much more of a straight line. Mind you, I'm not relying on the large language model in the way that we've heard the horror stories, right? Like I'm just using the large language model to get me to the point where then I can kick that off into the direction because that model will tell you things that you didn't know exist. I can think of scenarios back in the day where I would go and I would find the answer, and the answer would make absolute sense, and I would come to find out later that there was some little nuance section somewhere that was relevant to this particular scenario. The conversation with large language model, for the most part, is gonna pick those things up and avoid that danger. An example of the difference between how research was before, where research is now and where it's going is, as I said, I used to have the Google, read 10 articles. Today, I will go have a conversation with large language model, but I won't be able to rely on the answer that it gets. I'll get to the source information. Well, where this is going is similar to that question of whenever you add up some numbers in Excel, do you go use a 10 key and then add those numbers up and make sure you get the same thing? You don't, right? You trust the machine. And over time, I believe that it will get there pretty quickly. And an example of just like a test that I did recently on that, I'd gone through a research on a question that a client had put to me and found the answer. You know, got a large response and put it together on my own and then kind of whittled it down to just the very key information that the client wanted to have. And so after I had completed that project, I ran a test and took the information that was the final source, which was this big thousand page document that, you know, had whittled in. I put that in NotebookLM, and I basically asked NotebookLM the same question that was presented.

Kai: Mm-hmm.

Micah Wheeler: And then asked it to whittle down to a very succinct answer. And it nailed it a hundred percent.

Kai: That's a great how-to as well, like for anyone doing that. That is very cool. You know, you talk about the sort of where we were a year ago and hallucinations. It reminded me, this week I was having a conversation, I was at a company and I hear this a lot where people said, oh, you know, I tried AI once and it was really bad at math. That was the exact argument. The question was, when was the last time you used this, right? Like, I remember two and a half years ago it was really bad at math. That was the big joke. To your point about the calculator analogy, it's like almost people who said, well, you know, I'm not gonna use a calculator anymore because that one time I tried it 75 years ago, it just didn't work that well. And we've clearly made no progress, so I'm not going back.

Micah Wheeler: Well, there's a few things there, right. For one, there's a lot of pushback on the technology of it's gonna take our jobs. Whenever I think about the theoretical capabilities of business and finance, there could be a whole lot of people out there that are pushing back and that they're fearful of that. But history has said that the people that don't adopt the new technology probably don't win. Think about the Luddites. You know, are we gonna go and destroy the machinery that's gonna take our job, or are we gonna learn how to operate the machinery?

Kai: Shout out to my wife, the Luddite. This is the wife episode. That's great. So I think it'd be interesting to hear your take on sort of the gap between the capabilities and the realities.

Micah Wheeler: The theory is you got rules, you got principles, and you apply to the data, right? And so I hear about in different softwares as I look at the software market and the different options. I see some really flashy, some really cool, futuristic ideas of how the processes will work. But I also think there's a lot of hesitation. There's a lot of security concern. We gotta make sure that anything that we know that we're dealing with people's financial information and we know that there's a lot of fear surrounding putting the information in, and there's a lot of basis for that fear of allowing large language models to look at people's data, right? And so that is very much so a limiting factor, I believe, in getting the thing really going, to the extent that it's real. I would lean on my security team with Wilkins Miller Information Technologies to give the advice and I listen to them and follow the rules that they establish and what it is that we can use. But I do think that that's a limiting factor right now in how fast this thing's kicking up. Because I think about the theory of where it can be. And I do think about where we are today. And I do think there is a significant gap between those things. And I see examples and I feel like it can work. But how quickly can you just run into that? You gotta be very mindful of the data that you're working with.

Kai: Yeah. On a personal front, you mentioned NotebookLM. What other tools do you use? What's your favorite large language model?

Micah Wheeler: I bounce around. Sometimes I'm using ChatGPT, sometimes I'm using Claude. Sometimes I'm using Perplexity. Sometimes I'm using Copilot just to, you know, just to get a different flavor. They all have different personalities and so I'm interested to see how they're advancing. And we also know that it's sort of a race where different models nose ahead. And so I'm not committed to anybody.

Kai: That's great. Well, Micah, thank you so much for coming on AI with Kai. It's been a pleasure.

Micah Wheeler: Hey, thank you for having me.

Narrator: The Playbook. One practical way to use AI right away.

Tim: Alright, Kai, you know, this is one of my favorite parts of the show. A practical use for AI in my business.

Kai: It's also one of my favorite parts of the show too, Tim. So today we're gonna talk about ditching the slide deck. For you out there that are making slides, PowerPoints, keynotes, things like that, I encourage you to try a new system. Rather than having AI make your slides or you hand making your slides, ask AI to make your slide deck as a website. Okay. Now you may say, well, I don't have any background as a web developer. I've never done this before. I say give it a try. It does a very, very good job. And that is because you can do a lot more with web technology in terms of display and making it dynamic than you can with traditional slides such as PowerPoint and Keynote. So it'll end up looking the same, but it will be much more interactive. In fact, the people who are viewing them don't even realize that what they're looking at is a webpage. And you can do these, whether it's in ChatGPT, they have Canvas. Claude has Artifacts. Basically, it'll create it. You can share it the same way. You can add a lot of functionality such as videos and transitions and things that you're used to from slides, but in a much more dynamic way. What I would encourage you to do is fire up your favorite AI tool and just describe what you're trying to do, right? So if you have a document that's an outline of a deck, you have data, whatever you're trying to display, just upload that into your desktop client and say, I would like a slide deck, but as web pages. And I want to be able to move back and forth through my deck with the left and right keys, things like that. And here's all of the things I want to do. And you can even add to your prompt, say, hey, I wanna make this interactive in some way, or I want a lot of great design elements. You can prompt the AI to create a really, really cool looking presentation. It's a lot easier than trying to manipulate PowerPoint slides.

Tim: Is it hosted on the LLM that you're using?

Kai: Yeah, it is. So if you ask Claude, for instance, to do this, it'll create what's called an Artifact. So that's when it sort of opens up a new screen and it will give you a preview of what it just built. And at the top there's a button that says Publish. And this is the same in ChatGPT for Canvas. So you can publish these and it will give you a link that you can share with people. And they will click that, it'll open it up in a browser and they'll be looking at the same thing you're looking at. And so they can interact with it as well, right? Or you can just simply share your screen like you would in a typical situation with slides, but you have a lot more control and you're not fighting with the apps such as PowerPoint or Keynote.

Tim: Okay. And does it expire?

Kai: You have to do that manually.

Tim: Okay.

Kai: Right. So you basically unpublish it if you want to take it down. Well, that's it for another episode of AI with Kai. I wanna take a moment and thank everyone. This has been a great experience, having a lot of new listeners. Really appreciate the support from everyone. If you know of people that you think would like the podcast, please take a moment to share it with them, and we look forward to talking to you next week.

Narrator: You are listening to AI with Kai. For more episodes, practical tools and business insights, visit aiwithkai.com. Follow the show wherever you listen to podcasts and share it with someone building a business on the Gulf Coast.