304 The State of AI Adoption In Accounting in 2024, and What’s To Come

Okay, I've got a fun one for you today.

I had a, frankly, a really fun.

Conversation recently with Andrew Wall. Andrew runs a live streamed show every.

Friday for Accountants Livestream.

Gosh, that just stresses me out just thinking about it. It's called Friday I Show about AI Development's tech stuff for accountants. If you've ever seen Andrew speak, like, he's one of the very best people we have in the profession on tech adoption. But a couple weeks back I went on his show and we jammed on. On tech on AI.

Super fun.

It was so good at the end I was like, you mind if I actually share this on my feed? Because the tech folks that we've got hanging out on Jason Daly, like, they're.

Going to love this conversation.

And you can tell talking with Andrew.

Like, he asks very intelligent questions. He gets what's happening. And so it was a fun chat.

Kind of a check in on where AI is at right now, where firms are out on adoption, and some predictions for what the next six months look.

Like 18 months and a few years out.

So come on in for this kind of state of AI in 2024.

Hey, everybody. Running into some technical issues. I'm supposed to have an intro rolling right now, but of course it really doesn't want to roll my intro. But that's all right. So I'll do our own little intro. I am thrilled to have Jason Stats with us. And for those of you who aren't familiar with Jason, he runs a daily podcast for accountants as well as a community for accountants. And it's amazing quality in both the content as well as the actual quality of the videos that you produce. They're funny, they're informative, um, and they're absolutely amazing. And they're all over, I guess. What are your main platforms where you communicate with people? Jason, is it Twitter is a big one for you and LinkedIn, I am.

I'm assuming, start on Twitter and then YouTube in the past year or so, LinkedIn's probably biggest, but I'm, I'm. I'm on just about all of them now.

Do you have a favorite?

I think Twitter's probably still my favorite, even though it's definitely changed. It was the original.

Yeah, yeah, it definitely has changed with Elon's at the helm, which is interesting because, you know, he's. This conversation is really about AI and he's obviously released Grok and now Grok 3.0 is out and lots of claims going on there and lots of things that he's doing with AI in the Media, which maybe we'll chat about later. But I think, you know, one of the reasons I'm so thrilled to have you on the show is because you've got such a unique perspective on AI because of what you do. You work with so many different accountants and bookkeepers across the community, so you've got insights into how people are actually implementing AI into their practices. I have some unique perspectives on that and I can't wait to hear yours. As well as the fact that you are really a technology focused professional. Right. And do you have any technology training or background or. It's just a passion.

I have some software development background and it's always just been. It's always been a passion. Before AI, it was no code and we were geeking on RPA before this. So it's always, it's one of. Always been a very big passion of mine. I, at one point I was leaving accounting and got drawn back because I saw the technology opportunities and realized, oh, accounting is kind of just a technology problem. So I love how those things come together.

It really is just a technology problem. And I was just mentioning to you before we went live how we've hired a new team and what we've done is hired an engineer and a bookkeeper and paired them together because of the fact that it is becoming more of a technology problem than it is certainly an accounting problem. And it's the combination of those two skill sets that are really critical of working together. But I guess what we've seen is this evolution of technology, even in the recent couple of years has transformed. There's lots of buzz around AI. We're laughing about tech companies putting it in their call symbols to boost their investors, investment in their stocks, to websites who are just claiming their AI. Which I guess the problem with that is it creates a lot of confusion, doesn't it? Yeah.

And like, nobody rolls out of bed one day and thinks, oh, I hope, I hope somebody I don't know, I hope I. I get AI in my app today. It's like, no, my job before is the same as my job is now. If you can show me a way that it'll save me a pile of time getting from A to B, I'm happy, but nobody's clamoring for another chat assistant, another place to go and like, have a conversation about something like, nobody wants more. They're like, man, if you can help me get my existing job done more efficiently, sure.

Right. And that's an interesting thing too, right? Because these chat assistants, I mean, what I find with My client base is that people don't want to interact with quote unquote AI or chatbots. They want it to feel really bespoke and personalized. And yet one, chatbots are probably one of the first things to sort of take off in the AI realm. What are your thoughts on effective ways that firms can use AI to have to build better and effective communication both internally and I guess externally as well?

Yeah, if you look at like the struggles of being in education right now and, and students, you know, quote, cheating with, with using ChatGPT to write essays and stuff like that, like that is the environment I want to foster in my firm. Like I want.

You want people to cheat?

Yeah. Who are finding innovative ways to get stuff done. I think the blocker for most right now is they just haven't established the boundaries to do so responsibly and like created a responsible use policy and all that. AI is interesting and I'm talking like more bottom up AI, like going out to a chat GPT and having it help you with a specific pain. It's different than most of the tech we adopt because it is, it is really a, a, a bottom up sort of thing. Nobody's going to sit you down and say here's the, here's the 10 ways that you can use it, here's what all the buttons do and all that. Like it really doesn't work that way. So it's, it's more about getting it into people's hands, sharing your wins, but also ensuring that your incentive structure for your team isn't going to punish these detours. Like if I'm paid to sit in a chair, you know, it's like the age old problem of well, what's in it for me to get to become any more efficient? And so you'll have leadership on one side of their mouth saying yeah, we want you to lean into AI and all these things. And out the other side of their mouth they're like, yeah, but all of your incentives are tied to billable hours. So oftentimes it's that incentive structure that's kind of broken. But it is very different to the traditional top down tech adoption where the big boss comes and says we're going to use this now and we're going to do implementation, all these things. It's kind of the opposite.

Have you seen any good incentive structures that do encourage this?

I'm a big advocate of production based pay. There's, it changes the game from one game to a different game for sure. And so then how that, yeah, how that pay structure works becomes the new game. But that is more similar to the game I play as the business owner than the other one. So I think it's just about kind of getting their incentives aligned with, with what I want them to be doing.

Yeah. Which is, is an ongoing struggle because no matter what pay structure you have in place, people will find ways to game those pay structures, won't they? Yeah, and I think you're right. Like, and Chad said this in our conversation with him a couple of weeks ago was the importance of having a policy around how you address AI in your firm. And I think that some people can find that overwhelming. Are there any good examples out there of existing policies that people have shared? Do you have any thoughts on how people could create it? A good internal use policy?

Yeah. Two weeks from now. I've actually partnered with a group, I'll say a network security group, to put out a template that folks can swipe. It's the most simple framework I can get to that is fundamentally, if you think about a table, we have different types of data. You have public data on one end of the spectrum and you have like client identifying sensitive data on the, on the other end of the spectrum and then say redacted information in the middle. The most simple framework I can get to is for each of those types of data, my firm has to outline what are the approved apps I can use that data in. So on the most sensitive use cases, these have to be super locked down apps where there's no data training and you know, two factor authentication to get access and all that. But my marketing team, I want them to use the best tool for the.

Job and the most secure tool may.

Not be the best tool for that job. So I don't want to give them an inferior tool when it doesn't really matter due to the nature of their work. So that's the way I think about it is for given data types, identify approved tools for those data types and then see what people can, can figure out. Like having a feedback loop where folks can share like a, you know, a space in Slack or in teams or something like that where they can share what they're doing, get ideas from each other. That stuff's gold. Because I think if you just, you just tell them here are the 10 use cases, it totally misses the point.

And you have, you have a community where accountants are able to do that basically with other accountants. And do you want to share a little bit about your community? Because I think your community is an amazing space in itself which is filled with all your amazing content. But one of the powerful things about any community are its members. So maybe you can share us a little bit about how your community works and tell us if any of that's going on there.

Yeah, 100%. So my community is called Realize RLZ IO. We've got a 500 member cap forever. And we're at that, unfortunately. So this, if you're interested. I'm sorry, there is a wait list, though. Basically what it is is really a peer networking community. It's not an access to Jason community or anything like that. It is folks who have only stumbled into it very deep in my content. So you end up with a collection of people that can go like 17 minutes into this very nuanced sort of thing. They all come in with a somewhat similar perspective on how to run small accounting firms. And then the job of the community, like the. The infrastructure, I guess, that I've created, the job of that is to just make connections, like, get people connected with each other. I don't know about you. The, like, the very best thing I ever did was sit down in a room full of people that ran firms just like me for the first time, and I'm like, oh, geez, all of these people get it. And like, yeah, in like, a powerful way that even my spouse doesn't get when I come home at the end of the day. And she's like, you do a great job. And I'm like, well, thanks. Yeah, maybe you don't actually know, but. But then you go and you, like, sit in a room with a bunch of people who they just get at 100% and you can steal ideas from them. And like, that's all just. That's all just very energizing. And so creating that for other people, like, that was. That was the big goal.

Yeah, I mean, that's why I fell in love with QuickBooks was not so much the product was the community. And the way that they created that environment for me where I could share with other people. And. And it was so unique to me at the time. Like, I guess this goes back about 10 years now, because prior to that, accountants and bookkeepers, we didn't share like this. We held everything close to the vest, and we viewed everybody as a competitor. And we didn't have that mindful, that abundance mindset where we were willing to share and learning. And yet it's propelled these firms to be able to grow in unbelievable ways. And so I think you probably have a group of 500 people that are able to Grow and explore new technologies in ways that I, I don't think any other group of accountants can do because of the fact that they've got each other there in that support system. But I wonder when, when you first started talking about ChatGPT, you know, a year, year and a half ago, and it was coming out, people had to start approaching AI for the first time. And what did you see? Like, what were, what were ways that people were in your community able to actually successfully approach AI and start implementing it in your business? Were there any tips or tricks for those in your community that were able to do it?

Well, yeah, in the early days it was very much about writing. It wasn't until later on that chatgpt was more multimodal. You could give it an Excel file, it could give you a PDF back, stuff like that. So early days it was very text based. And so it was a lot of writing. It was outlining a blog post, outlining a podcast episode, taking this body of text and making it into something else. It was largely text based stuff, which was super useful because man accounts like we, we did not get into this thing to send sensitive emails like that. That is one of those things that you can burn a whole afternoon on, and that's where it starts.

And most of us are better with numbers than we are with words, right?

Oh, a hundred percent, 100%. That's where it started. But it's come a long ways from there to now. Just the fact that you can, you can give it files, you can ask it for a work paperback to show its work in PDF format. That's like annotated, just like a person would annotate it. Like there's just so much more possible now. In the beginning it was like, go out and you can probably tackle these kind of problems with it. These days it's like the biggest problem is access to information. What context does it need to do the job that you're asking it to do? And still the headache of having to round up that context, come over here to do the thing, try to get it to do the thing, and then put the output back over here. Like that's still the kind of the highest friction aspect of it.

Right? Right. So, I mean, obviously it's, it's changed a lot from those early days. If, if someone was just hopping into the world of AI right now, where would you recommend they start?

Just start experimenting. I mean, I don't, I try not to.

I try.

I mean, writing is the easy one. Next time you need to write something, something like that I try to just really though like get people to have a conversation with it because the blocker for most people is just remembering that it exists and, and like even using it. Everything from converting a report from one format, the way it comes out of one system to the import format, it has to go into another system. And like there's just so many silly little things like that where it can save you some time. But we're not, we're not used to like this work companion. It's really interesting. I've got kids, my oldest is in second grade so they're not that old. But like kids that are growing up right now with like Roblox, like Roblox has a built in AI assistant where you can ask it to do anything that you want it to do. And there's this level of, I mean it's like, it's like when iPads are becoming a thing and you see kids who have used iPads their whole life and they just have this, this instinctual ability to do anything. There's, there's going to be a generation of people that have this expectation of this kind of co working assistant that could do anything for them. But for the olds like you and I like, it's, it still requires rewiring the brain to go out to this other place to get help. So I, whatever for you, whatever makes that most accessible. Great thing is now on Mac we have a Mac, a Mac OS desktop app for chat GPT which is super cool and hopefully we get that for Windows soon. But whatever brings you back to it. And honestly sometimes the place to start isn't even work stuff. It's like pick apart this meal plan or help me do this or that or I've got this weird pain in.

My knee and just getting comfortable and familiar with it and realizing that it is an iterative conversation, right? And that maybe the first answer isn't going to be great, but that the more you talk with it the better it's going to get. And making sure that you're like you said, providing it the context it needs to be able to provide the right answers. Because that can often be one of the problems is that if it's giving you crappy answers it's because of the context, you haven't narrowed it down. And do you have like, what are your thoughts on frameworks? Like things like the Risen framework. Do you have frameworks that you recommend? Or do you think people should just go in and just start not worry so much about having a framework that they use when they're dealing with prompts.

I don't push people to specific frameworks so much because each new model you get the constraint changes in terms of what you need to tell it and what it'll do well and what it doesn't do well. I do think we are still, we have Google hard coded in our mind where you chuck a single thing in there and you get a page of results and untraining that to becoming more of an open ended conversation that, that I think is a, a big shift just kind of learning what works and what doesn't work. Trying to be plugged into some sort of, some sort of thought leadership on AI. Uh, because there is so much that's changing and it's hard to keep up. Like it's not the sort of thing where you have to be going deep and plugging in every single day, but just to have some sort of drip feed there to upskill you a bit along the way to learn best practices. Because what two people can get out of ChatGPT right now, I mean is there's just a vast like delta from someone that can hustle that thing and know all the right ways to do it versus someone who can't. And it may become less of a learned skill down the road as it gets better at reading our minds. But it's honestly all the same problems that we have working with other human beings. I mean you go delegate a task and they come back and do it in a wildly different way than you expected and you're like, oh yeah. Well it's actually because I didn't tell you xyz. So it's oftentimes it's actually easier for somebody to get mad at a computer than a human where there's a little more, there's a little more friction with the human than just like closing the tab. But it's easy to get frustrated with it when it's like, yeah, you didn't really give it the context to, to do it well.

Yeah, I've gotten salty with ChatGPT once or twice where it wasn't giving me what I wanted and I had to use some expletives. Hey.

This episode is sponsored in part by Live Flow and do I have big news for you, buddy. Lifflow now supports Xero Xero accounting ledger 0 for the first time. It's big news. I ran a primarily Zero practice and I can tell you what a headache it was to have like some tools that worked with my QuickBooks clients, some tools that worked with my Xero clients, drove Me nuts. Live Flow. Let's just zoom out and recap everything that Lifelo is doing for you because it's bringing everything together. Google Sheets Support Check Excel support Checkaroo QuickBooks online. You betcha. Zero support now. Exciting stuff and sneaky, sneaky features like valuable feature consolidations. Consolidations being able to roll up a whole bunch of different like chart of accounts that like no matchy matchy required various charts of accounts, rolling them all up in a super simple, really nice ui. And then because it's live Flow, you can pull just about any report from the accounting system into your favorite spreadsheet either one time or on an ongoing basis. Like it continues syncing out and it works across your whole stack now. Okay, I honestly don't know what more they could do for you. Learn more about LifeLow. Check out the link down in the show Notes this episode is sponsored in part by TeamUp. Common question when you are hiring people around the globe, how do you even time zone? When do people work? Do you work at the same time? Do they work at night while you work during the day? Gang, it's your team. You get to make the rules. But common things that people do when they're hiring folks in the Philippines, sometimes they'll just have them stick to your business hours so they will literally work.

At night during your firm's hours.

When I was running a firm, this.

Is what we did. Everybody worked during our hours and it was all pretty straightforward.

But increasingly what folks are doing and this is a little easier on lifestyle.

Because like would I like to work graveyard myself?

Probably not.

So is it unfair for me to ask my team to do that? Maybe.

But increasingly common now is folks are looking for getting like two to three overlapping hours a day with your team. Now increasingly this is less of a big deal because we're all sort of moving to a more asynchronous version of.

Work where we're not always hopping on.

Live calls and needing to take meetings.

And all that trending towards working from anywhere anytime you want.

But I can tell you in my firm what I ended up with was.

We had, we actually had like a three hour block a day where it's like everybody needs to be on during this time each day to take care.

Of any of those things.

Synchronous things that may need to happen. Otherwise go wild, work whenever you want.

And that framework works really well with.

With this offshore hiring approach is you.

Pick that two to three hour window during the day when everybody's going to be available. Everybody has to Be on then.

Otherwise, up to them. As long as they're getting their work done, I don't care.

At the end of the day, you're making the rules. So if you want them working during your regular business hours, you could certainly do that.

They're going to be available all the time. You may be asking a little more of them.

So maybe it's harder to find the right folks because they may not wanna.

Work at night or bookend it at one end of the day or the other. So when they're getting on, your team's getting off. Or the other way around.

Now, if you want some help to find some talented accountants in the Philippines to come work with ya, check out.

Team up link down in the show.

Notes. One of the things I've often talked about with you and with other guests is that I think that one of the other problems that we're having with AI is that it's a solution. Hunting for a problem, and that's why a lot of people struggle with it, is because they're so amazed with what its capabilities are, but they haven't really figured out, well what they haven't started with the problem first and then saying, okay, well, this is the way we can use AI as the solution. So I wonder from your perspective, have you seen some problems within your community that AI has turned out to be a great solution for?

Yeah, yeah, it's really a looking for the solution. Like, that's like, there's so many people working on that. It's like, how do you like this framing that I heard the other day? It is like the invention of the tractor. The tractor itself doesn't do all that much. Like it's, it's not particularly helpful. It is all the things you can build around the tractor that make your life better and all the attachments and all those different things that kind of is, that's kind of how language models feel like right now, like, cool, we got a tractor. What do I do with it next? I think the. Honestly, the biggest stuff I'm seeing this year is actually probably AI vision models, which can pull text off of things much, much better than OCR ever could before, even handwriting. And when you pair an AI vision model, being able to see and understand something with the ability of large language models to do a little bit of reasoning, it's now possible to just check this thing, a pile of documents, and it can, it can read these things near flawlessly and make these kind of connections itself that OCR couldn't have done before to get you to a more structured output. And so that's, that's really powerful when it comes to like accounts payable processes. And in the US a huge amount of our, our income tax processes are, are based on paper documentation.

Yes.

From all these disparate sources. And so to have a centralized kind of mind that can ingest all of that and then output it in a structured format goes way farther than OCR ever took us. I'd say I'm probably seeing the most interesting things with vision models. The, the biggest thing with like chat GPT and why and I've got, I mean I've got hours and hours and hours of like ChatGPT use case for accountants and stuff like that on YouTube. And honestly what makes that hard for people is the best use of that is, is actually pretty perfect personal. It's what's like that thing I'm doing for one or two clients that wipes out an afternoon every month and like what is the very me specific use case that I can plug it into? Like those are the coolest things I've seen is just the very nuanced sort of specific to you solutions.

What's the favorite one that you've seen? If you can share.

I love the report hustling stuff. So in ChatGPT you can save GPTs which are like custom little assistants that you can create and on the ChatGPT team plan you can share those GPTs just internally with, with other members of your team. So let's say in a month end close every single month you got this client that uses system A, we gotta export a report out of it and then have it in a different format to import into system B. Well one person that's good at prompting can set up that GPT where all that assistant knows how to do is you give it a file, it changes it to a different format and and then spits out you know, say a CSV in the correct format where it goes and gets imported as a bunch of journal entries or something like that. That's a GPT that a member of your team can make once and then anytime a member of your team has to do that same process, you've got it right there. It's super easy. It's like all these e commerce integration tools which are super, super valuable and do all these different things. It's like a bite sized version of that that anybody can go out and create on their own specific to their need.

Yeah and like you said hyper personalized to their unique use case which sometimes those tools don't Solve the specific problem that you're having, which is, I mean, that's the thing that's incredibly powerful. And you talked about vision too, as we use Power Automate for rpa and now Power Automate now has something they call AI Hub built into it, which has this vision tool. And you can literally build your own expensified daxed receipt bank with as little as five documents. And it's using AI based image recognition. So the quality of it is phenomenal. And literally all you do is you add up five documents, you choose the fields that you want to pull off of it, and in those five documents you show it where it is and then not only does it come back and start fetching that on any other documents that you give it going forward. It's actually telling you, well, I think I'm 99% accurate here, I think I'm only 80% accurate over here, I think I'm 75% accurate here. So it's telling you and giving you feedback onto how consistent it thinks it is and it's able to learn. And it's not like traditional OCR where now that has to be in the exact same spot it was every time. It's just said, okay, you've told me where it is and now I'm seeing relative to the other information on the page, what types of things to look for so you can load in other things that don't match that exact frame or format and it still recognizes it. It's incredible. And like you said, is, I think on the personal tax side, this is one of the areas where I think we're going to see big, big changes this year. Obviously personal tax for both you and I, you know, even though we're in different countries, it's all, you know, very seasonal based. Right? And we've got some time before our next season is due and I think we're going to see some big changes in what we see from both individual firms building out their own tools, but also applications and solutions coming in. Because think about the ability now with using computer vision to be able to have your clients who dump a shoebox on you. And now at least the AI can take that and be like, okay, these are donation receipts, these are medical receipts, these are, you know, we call them T4s for income. You know, we've got different slips, but to be able to even just simply put those into different folders for you so that now you have this information categorized and sorted in a way. And not only that, like remember that age old problem of hey, hey, give me 12 bank statements and they give you 12, but two of them are repeating or they give you 11 and now you don't even have to look at it to figure out what's missing. You throw that into the AI and the AI is like you're missing November, you know, you're. And, and to be able to sort and do that preamble categorization beforehand, I think is incredibly, incredibly powerful. And for us, as you know, we're an rpa, we're diving headstrong into rpa, which, you know, I know was hot for a minute, but it seemed to like it didn't get the lift off that I think it deserves in our community. But I think when you power RPA with gen AI, the things that you're capable of doing are incredible. And especially when you start with the problem first, which for a lot of us is gathering and collecting and sorting information. Right?

Yeah. AI agents are really the realization of the promise of rpa. And it's unfortunate that to date, in my opinion, Microsoft's really dropped the ball where they were kind of in the unbelievable position to really lead the way here, and they haven't. But the notion of having a, an AI assistant that can see your desktop, they can understand what's happening on your screen, you can be in conversation with it. If RPA is kind of the, the fingers, the ability to work the keyboard and the mouse and all of that, combining that with an AI that can see and comprehend everything that's happening on your screen, but also be able to type into fields and click on things and like that is 100% the future of work. And what makes the notion of AI agents so exciting. Yeah, you combine with that their ability to learn and do repetitive things, and the next time you do the same type of project, it just kind of gets that. That's going to be so incredibly exciting. The frustrating thing right now, honestly, the, you know, the tools that are pulling all that disparate information together and doing really amazing things with it. We've got a number of them in the states now that are like, they'll have it before next tax season, but only by a couple few months, like they're working on it. And I've seen some demos that are just unbelievable. The really frustrating thing right now, frankly, is it's further fragmenting our workflows because we have these new companies that are really good, like AI first companies, almost in the, almost in the way that we saw SaaS, first companies that looked very, very different to desktop first companies. And you saw the desktop companies try to go SaaS and that was really hard for them. There's a crop of companies that like just understand the new tool stack in a way that the, I don't know if we can call them legacy software companies can yet. But the result, at least in the states here is it's like, okay, well here's this, this cool new thing that, that absolutely crushes this part of the workflow. But not unlike ChatGPT, it requires me to pull all this context out here and then like pipe it back into the rest of of the process because it's not doing it totally soup to nuts. So that's kind of a frustrating intermediate step. But hopefully where this is headed and you're talking about kind of more over the top AI agents and something that even lives on your desktop and can use those various different apps. Like hopefully where we're headed is something that unifies all that stuff. Even being able to take something from A to B&AI agents or assistants almost serving as the APIs of the future. Where in the past we've had integrator apps that will take API API A and make a talk to API B, actually having assistance on top of those and semantically being able to say, I need this thing to go over there, do it on the third of every single month. So that we're all kind of building that software on the fly.

Yeah. And I guess now one of the things I see also and hear from the community is that a lot of people are hesitant to jump into the AI fold. Even though they're interested in listening, they're not rolling it out because of that fear that it is changing so quickly and that agents might replace RPA and stuff like that. What are your thoughts on that? And specifically, I mean, do you think agents would replace rpa?

I think there'll be one in the same. I mean, OpenAI is already working on this for their ChatGPT desktop assistant. Their plan is for it to not just talk about the work, but actually do the work and be able to interact with the things on your screen. It's probably going to boil down to what's the easiest interface to, to create this stuff that's going to be accessible for people. But I suspect the future of RPA is it's semantic, it's telling it with a set of instructions. In fact, I think Power Automate even does this. Now you describe the automation or whatever you want and it kind of builds out the structure for you.

Yeah, you can do in zapier as well.

Yeah. So I think it's, it's just a very user friendly version of that or maybe something that even watches your screen and identifies when it sees those opportunities, which we've actually already had RPA tools that have done that for a while now. So I don't know that it replaces it as much as they become one and the same.

Probably they work together because. Yeah, I mean that's the other thing too is like these agents look incredibly powerful, but what we've seen with AI, with its ability to make things up and have these delusions that I cannot trust something that's going to change. What I value, as I said to you earlier, and what I love about RPA is the consistency. Yes, I have to go through the process of dictating every step along the way, but once I've dictated that step and that process, I know I'm going to have consistency. There's not going to be any hallucinations, there's not going to be any changes in what step it chooses to take because it's thinking on its own.

Right.

And so I think particularly in the accounting world where we have so many processes that are highly structured, where we want a consistent step by step, that we might want to use AI to help us build those out so that we don't have to figure out, okay, I need to, this is how I choose the selector and this is how I do that. But once we have set that process, that, that process becomes locked in place, so to speak.

Yeah, 100%. Yeah, there's, I don't know, there's kind of the whole fear narrative and, and you know, firms that are maybe hesitant to jump in and it is changing fast and it is hard to keep up with. But I mean it's like, like early iPhone days when it was like, oh, when's the right time to buy an iPhone? As soon as I buy one, they're going to announce a new one with a bunch of cool new features. It's like, yeah, that is a problem, but, but that's not a reason to not start. There's a lot of organizational issues wrapped up in that from, uh, if you're change management blocked right now, you're just gonna be even more change management blocked when the tools get even better. So like there's a lot of things kind of bundled up to unpack in that, but it's very much a, a skillset to develop. And if you look at all the different tech that we use across our firms, anytime you, you buy into a product, you are hitching your wagon to where they're going. And I mean, never in my life has there been a technology that is developing as quickly as AI is right now. So it's like if we're, if we're just making bets, if we think about it that way, it is the right horse to bet on right now from, you know, ChatGPT 18 months ago, having memory equivalent to like two pages. And now we have models from Google that can hold like 3,000 pages actively in memory. And like it's just every aspect of this, from the cost to the speed to its capabilities, like it's just growing so, so fast.

It is an incredible pace of change, which I find exciting, some people find intimidating. But I mean, my ADD mind loves change.

Me too.

New technology, right? Have you seen any other hurdles that have held people back from AI other than fear of change or the pace of change?

Definitely just general scaries around the newness of it. Just like cloud kind of the ambiguous spookiness of this new thing. And a lot of stories you hear about the bad and the wrong ways to use AI, but sometimes that can get conflated into AI bad when the fact that there are bad ways to use AI does not mean that there are not completely okay ways to be and secure ways to be using AI right now. Definitely folks still struggling with the ethics of training data. And this isn't just a chat GBD thing. It's everything. And it's image generation and so many different examples of this.

Google's the worst.

Yeah, it's, it's tough. It's one of the, I can say it definitely feels like one of those what are you going to do, genies out of the bottle type of situations. Like it's really hard to put your finger on anything in the past where they successfully put the genie back in the bottle. Like what social media did to publishing. And like there's just so many examples of this where it's like, it absolutely does suck, but it, it is like a freight train that feels like it's not going to stop. And so there are, there are completely justified moral grounds to not want to be exploring stuff right now and stuff like Getty Images announced, you know, a few days ago, a model trained on all licensed images and like they're doing like rev shares with, with all of the artists behind all of the, and the photographers behind the images they use. And so there are people trying to do fair use versions of this. But if you look at a monster like, you know, GPT4 that supposedly trained on like one and a half trillion parameters, a massive percentage of the entire Internet. The practical aspects of how do we compensate people for what just got gobbled up by that model. Like, that's. That is a hard problem to solve for.

Absolutely. You know, and that's the thing, is that we are getting a whole bunch of hard problems to solve for. And you know, as far as regulations and guidelines, I know the EU has sort of led the way on trying to solve for that. The US is trying to step in and put in some regulations, but it gets really difficult, particularly when you have regulators who are, you know, bureaucracies who don't move fast at the best of times, and technologies that's changing before laws could even be put into place or regulations could even be put in place, which makes it really difficult. And you know, you know, I'd just be curious what your thoughts are on. So, you know, fair use and those types of reimbursement are one things, but what about some of the other ethical issues that we're seeing around, you know, being able, like perfect example, what Elon Musk did with the Kamala Harris video and, you know, faking her voice, you know, and that becoming so much easier for the average person to be able to implement. What are your thoughts on those types of AI issues?

Yeah, I mean, it's like in the US we're coming into a huge election cycle, obviously, and it's going to be wildly problematic. I mean, the only thing I can compare it to in the past is the Internet giving everybody a voice and letting anyone put their thoughts out there. When before that we were reading the work of journalists and folks that had a degree of, there was a lot more fact checking that went into what was published. And then when anybody could publish something, or especially early days Wikipedia, where you're like, I can't believe anything that's on there. As much as it seems like the right thing to regulate the ability to generate things of other people's likeness, honestly, I mean, that genie is already out of the bottle. Six months ago, I created a, basically a visual clone of myself talking on my podcast. And it took about 15 minutes and like, my wife couldn't tell the difference. So, like that's, that stuff's. The trains already left the station there. And even if they do regulate it, which they probably will, there will still be open source tools that can do that stuff in a way that's completely indistinguishable from reality. So I think there will be a regulatory side to it. I think the bigger Change will probably be our understanding of what we're seeing and what we're consuming and the safeguards that we have to put in place to, you know, when I call you and tell you that I need a wire transfer and, you know, just everything around that, I think we'll just have to have a totally different framework for how we protect ourselves from that.

And I think that the classic safe word is still the best solution for that. I don't know if you've implemented that within your family yet or not, but it's one of the big things I advocate for. As soon as I heard about stuff, because I've actually had a lot of fishing, we were. I don't know if you ever. I've ever shared the story with you. I'll tell you another time. But we were hacked a couple of years back, and ever since then, we've been targets of phishing because we're obviously on some list now, some, you know, dark web list, and we get phishing attacks all the time. And. And so my staff and my family, we all have an internal safe word, which is. So whether you're being spoofed by email or you're being spoofed by, you know, a AI generated video or voice clone or any of that, now, I think that in this day and age, unfortunately, that's. Every family needs to have their. Their safeware.

Yeah, a hundred percent they do. We talked about that last holiday season. Cause it. Yeah, it's a. It's a huge deal, especially as we're all just out there. Like, anybody can create a clone of yourself or I and spoof a phone call. The. I think my favorite I've seen was a few months ago, there was a finance person got pulled into a meeting with a bunch of execs in a zoom call. There were like four or five other people there, none of whom were real people. They didn't realize that it was all like avatars of their bosses and they sent like a $6 million wire or something like that as a result. But, yeah, the. We have to keep innovating and how we protect ourselves from that stuff as it develops.

This episode is sponsored in part by Stanford Tax eight tax firms. You got an eye on that calendar, buddy. Next tax season it is. It's going to be here before you know it. I know you were thinking about extended deadlines, and I don't want to be that guy. But the time to start testing your next tax season workflow.

It's now.

It's not January, when you can't afford.

For it to go sideways.

It's now let me tell you about Stanford Tax, Kay. Big problem with 1040 workflows is asking intelligent questions, putting personalized questionnaires in front of your clients rather than the dumb boilerplate ones that ask a whole bunch.

Of questions that don't apply to them. Right?

Stanford Tax. Great way to do it. In fact, they've beefed up the platforms they support. CCH Access Pro System, FX Gross ultratax, Osert Pro Series, Drake generate personalized questionnaires that only ask the questions they need to using actual proforma data from their prior year return. Oh my gosh, somebody did it. Only ask the questions you need to ask in a modern web interface, get.

Them to upload the required docs and buddy, you're off to the the races. You just asked for all the goodies.

That you need and none of the.

Goodies that you don't.

Leaning into what like web forms can enable for you without losing that like, personal touch. The worst version of completing the thing.

Online is the thing where it's like.

None of these things apply to me.

And my tax professional knows that. Okay?

So if you're looking for a smarter way to do intake, check out Stanford Tax link down in the show notes. Today's episode is sponsored in part by Cloud Accountant Staffing. And I got another resume for you to look at here, gang. I got a live one. This is an actual resume the Cloud accountant staffing got for an accounting firm.

They were working with. How's this sound? Especially for you tax firms out there. Graduated with a bachelor's in accounting in 2017, spent two years in audit before going to tax. Good for them. So you know they got a good head on their shoulders.

Spent their next two years with a Western firm just doing tax work. Two years after that with a top 100 firm whose name if I uttered here, you would recognize there they were a senior tax associate also in charge.

Of training junior tax associates. All sorts of accolades here around like.

Culture awards they've won.

Sounds like a real addition to the vibe to the vibes of your firm. Good vibe manager experience in the CCH.

Access suite and everything else associated with business. Tax prep gang. Super talented folks out there just looking.

For a happy home and that happy home could be yours. People like to talk about like trying to hire big four talent and that sort of thing in the US and just how hard that is.

Do you know how many big four.

And top 100 employees there are all around the world in the Philippines, the.

Massive offices they have set up in the Philippines. Those people just like Stateside, where they're.

Like, man, I don't want to go.

Work for this giant machine. Those people are looking for cool, small, progressive firms in the Philippines just like they are in the US including incredibly talented people and cloud accountant staffing.

They're going to help you track them down.

Looking to build your team of accountants in the Philippines? Check out the link to cloud account.

Staffing in the show notes.

Now, what about, you know, moving on to maybe some of the specifics of what you use AI for, Because I find it interesting, like are you using any AI at all in any video generation? Like, obviously you're not using something like Nvidia or anything to generate your videos, but do you use any sort of AI and video editing aspects or is there specifically in some of the video content that you produce? How are you using AI in that aspect? Or are you.

We use it tons. So there are, there are just workflow examples of AI. So you take an interview like this. There's tools now that will cut down an interview like this and remove all the pauses and do camera cuts to fool you, fool me back and forth and do that in 20 seconds. What would have taken a junior editor probably an hour to do before maybe more. So stuff like that. That's just super practical. We use AI to enhance the, the.

Silly and the, the ridiculous oftentimes.

So in the past we've had various recurring characters, Steve's, Steve's the biggest one who like are kind of these client avatars that we return to and tease and kind of do some, some world building around. And those are all characters that are real people. They got pulled into a warehouse one day and, and somebody bought them lunch to shoot some stock video featuring them. And I hope those people never stumble into my content. But we've now moved to those recurring characters purely being gener creations of our own, where they begin life as image generations with mid journey. And then we use other tools to put them into specific poses to animate them into doing different things. And these are now not real people, but things that we have created from scratch that we have full ownership of that I can print on a T shirt that I can print on a, on a, on a, you know, silly oil painting version of that person and hang it on the wall behind me. I mean, stuff as simple as that is, it is part of every single thing that we create. Now we don't do anything along the lines of creating something that looks or sounds like me just because that's A, we certainly could. We have all the tools to do that.

That's.

That's just a line that I'm not comfortable crossing because I think it cheapens what we do, which. Yeah, I mean, I've got a team of four people. Like, there's a tremendous amount of effort that goes into all the creative stuff that we do. I don't think it would heighten, you know, the quality of what we're doing if we did do that. One really cool thing we've done is so we've got, we're up to like 29030 minute episodes of my podcast and we've got a, an AI assistant that's trained on all of that data, where I can ask it a question it will instantly answer. Citing huge chunks of podcasts going back the last two years. It's remarkably accurate. A few weeks back, we, we did a, a debate where basically I would answer, I would answer a question and then my AI assistant that just uses my training data would answer the question and the answers were blindly judged. And we went through six questions and it came out a dead heat. Uh, it was literally a tie. We'll use it for situations where it's going to be fun or interesting stuff like that. Occasionally, if someone asks me, hey, have you ever talked about this or that in your content? And I'm like, I mean, there's hundreds of hours and I'm like, probably a bunch of places. Chuck the question in here and it'll give you a few citations to 42 minutes into this podcast or something like that, and it can be a helpful tool like that.

Cool. Um, just out of curiosity, I'd love to know, like, are there any specific, like, maybe you can share some of the names of the apps that you're using. Um, and before you do that, quick question about that video tool. Did you convert it all to text and is that what you did? And then use that as a library and like a GPT? How did you do that?

The chat assistant? Yeah, that one's a tool called Delphi. D E L P H I A I. That one's super convenient in that one of the ways you can connect it to data is just connecting it to an RSS feed of a podcast. So every time we publish a new episode, it gets updated with that information and we don't have to do any extra work. And so, yes, it's pulling from the text transcript of that. And so what it's retrieving is the text of what I said, right?

Yeah. Because otherwise the size of the files would be massive and you wouldn't be able to have the necessary context. As we talk about all too often, what other AI tools are you personally using? Obviously ChatGPT. Well, is ChatGPT your go to generative AI or do you prefer any of the others?

ChatGPT is the best if you're working with different file types. Right now, Claude, they've got a newer model, Sonnet 1.5. So the website is Claude AI. That is the frontier model right now it beats almost everything else and for any sort of writing or outlining, it's really, really good. Have you played with Claude Artifacts at all yet?

A very small amount.

So they recently launched something called Artifacts, which is like a little environment to run what it creates. And so if I tell it to create code to, you'd be this little one page site that, that does like this complex commission calculation for a client or something like that. It'll write all the code, but it'll actually then run the application there also. So I can immediately run it and say move this button over here, change this or that. And it's such a fun, fun thing to play around with because it is like semantically building software live and telling it to iterate on. It's like having a little engineer sitting next to you to make this little helpful thing.

Yeah, truthfully, I haven't been doing it to build a lot of code really, which I've just been too focused on other things, I guess. So I haven't played around with that too much, but I was sort of impressed with that capability and it's just again amazing at which the pace of the innovations are coming. And I like that we have competitors in the market. Right. I mean, it does seem that OpenAI has a pretty healthy lead, but they're not completely beyond the realm of anyone else being able to catch them and compete with them. It's interesting to see the way different organizations are approaching it, both in the training of the data, the interfaces that are being used and the models that, that are coming out as a result of the different approaches the different companies are taking. I guess one other question I had for you is this sort of, I guess goes back to some of the fears we talked about of AI. Do you think, do you believe in the singularity? Does that have any, do you have any fears around that? Do you think that that is an inevitability, a possibility or just a bunch of hype?

Yeah, there's a lot of things wrapped up in like the framing of the singularity. But I Mean I think of human intelligence as you know, if you, if you charted it on a, on a graph, you have a level that is human intelligence. And we are super afraid of anything that gets close to that. Right? So we got like dolphins down here and, and we like, we like dogs because they're a little bit further down there and it's for sure getting closer to be honest. I mean it seems absolutely inevitable that at some point we create something that is capable of surpassing that. Not just surpassing that, but then continuing to, and all likelihood. I think that's where we're headed. And in our lifetime we will probably have something that is capable of much more than what we are just ourselves with our, you know, with what we know, what we're capable of. It's very contextual. I mean because we already, we've, for a long time we've had computers that are way better at humans and at doing very specific things. And you know, people like to, to say, you know, chat, GPT can't do basic math. It's, it's never going to replace me. And it's like buddy, computers have been doing calculations better than you for decades. It's just a matter of combining those two technologies and you've got something pretty compelling. So yeah, I honestly I do think it's, it is inevitable. I'm a little surprised that this hasn't been politicized more than it has yet. Because if we are truly on this inevitable path to something that can cure disease and, and this, this sort of civilization altering thing thing, then this ought to be the next space race. I mean like everybody's, there's this notion that whoever can get there first has this massive advantage. Right? There's talk recently of this, this mega server farm that Microsoft and OpenAI are said to be building together like $80 billion. And they think that will be the server farm that can get us to quote General Intelligence. And just out of curiosity, I went back and looked at like what did, what did the Soviet Union spend spend on the space race in today's dollars? Almost exactly $80 billion. So like it's, it'll be fascinating to see where we're headed. I'm a little surprised that the governments haven't gotten more involved. And I have to imagine the better it gets, the more involved they will get because everybody's want to, going to want to get there first. But it's going to be wild like the next decade. Like it's, it's going to be very interesting.

Or maybe the governments are involved in ways that we don't see and understand. Right. To go back to that conspiracy theory, because I cannot imagine for the life of me that there is not some cyber war going on behind the scenes that we're not aware of in a, in a race for, for artificial intelligence between Russia, China. Like we know that there's a cyber war going on between Russia and the U.S. like it's, it's not talked about, but it's known. Right. And so I've got to think artificial intelligence is part of that, that cyber war that we're maybe not aware of.

Yeah. Sam Altman's actually sets OpenAI CEO some really interesting things about state actors. And you think about, I mean, because GPT4 their, their best models, these are closed models, like they've got the secret sauce and nobody else does. You think about the, the power and the security implications of something like that getting loose. That's pretty wild.

Yeah, absolutely. Now I guess to drop the conspiracy theories and maybe put on our, or open up our crystal ball. Do you have any predictions that you'd like to put out there that you. Where you think we might be maybe 1, 3 and 5 years from now.

In the next 12 months? The biggest thing is what AI vision models are enabling for sure. Because I've seen it with my own eyes. Some of it's out there now. A lot of it will be out in the next six months. What just what we were talking about, the ability to take a pile of unstructured information and put it into a structured format. What it's going to do is it's going to force us to get more structured about a lot of things that historically we have not done in a very structured way. Because how you organize all that information may be very different than how the next for next firm organizes it. And so there's going to be some pressure on us to be better about standardizing that stuff. That's like next 12 months. That stuff is super, super compelling. Next 24 to 36 months. I think agents really become a bigger thing. I'm particularly excited for desktop agents. And the ChatGPT desktop app is like the first little infant kind of baby version of that where it can see my screen if I want it to and I can talk about what's on my screen. But big unlocks there are, well, what about when I can just have a passive conversation with it as I work? Like it's a co working thing. Like I'm typing and I'm working and I'm like, go look up this thing about this App that I can't figure out and it goes and does that in the background, figures it out like that. Seems incredibly helpful to have that sort of passive assistant with you totally removing the friction of going out to the, to another app. And then the next really big step there is when that assistant is able to use my computer. And we have all the tech to do this right now, but nobody's shipped something, probably because it needs to be in a state of reliability where we'll actually trust it to do that. Right. So if you think about the progression of where we're going, I don't know if you saw the, the last GPT4O announcement where they showed off that new voice model and it was super compelling, the talking back and forth. I think step one is having that on your desktop and just having this running conversation with an assistant. And we may have that for six to 12 months before that assistant can ever do anything on our machine. But imagine working with that assistant for six months and then one day it grows hands and it can do stuff on your computer. You're already comfortable with it. You know, what it can, what it can't do and what you trust it with and what you probably shouldn't trust it with. That's going to be super exciting, I think. I think in the next 24 months we probably have some version of that ultimately how that impacts work, because that thing can then start to learn workflows and do. I mean, any sort of repetitive thing that changes knowledge work a lot. And predicting anything beyond that, I think gets really tricky.

And I think you've got a great vision. The only thing I would add in that is the convergence of robotics with AI in more meaningful ways. Like to not only have that assistant on your computer, but, but as a robot in your house that go wash my dishes, go, go make my bed, you know, that, you know, hey, go pick up my dry cleaning. You know, that, that combination, it's going to change the world, I think substantially. And I mean, the only, the, I think the movie that I had it that I think has predicted the future of what we're. We're going to have, unfortunately, is Wall E. Those slobs laying down in our, in our beds that run around as in virtual reality, as, as these robots do everything for us.

If, if I, if we can aspire to a version of the future that's been presented in TV and movie, in my opinion, Star Trek, the Next Generation, that is, that is the ultimate.

Yeah, yeah.

Everybody has time to enjoy themselves, focus on the arts. Do what they want to do. It doesn't seem very realistic, but I, but I hope that's where we get to.

And you've got to love Star Trek, who has consistently projected future based technologies with a lot of what Gene Roddenberry's mind had come up with, with stuff that is, you know, so scientifically based and, and turned out to, to give us things in the real world that, that do look very much like Star Trek and, and Starfleet. So I like your optimistic view. It's. It's better than Wall E, I would say. And people are certainly in a lot better shape anyways.

It is amazing how sci fi seems to will things into existence. Star Trek, the notion of having a video call with somebody, you know, unthinkable, basically walking around with iPads, you know, that's nuts. The movie her, which, you know, is the whole like having this interaction with, with an assistant before ultimately falling in love with it and he has this thing in his ear. That movie came out 12 months before the AirPods were announced. So like all of these things are like, so ahead of their time and now it's like, oh yeah, no, that kind of is happening. Ready Player one. And VR stuff. Like, there's so many examples of this. It's kind of fun if you're a sci fi nerd, like, like I imagine we are, but it'll be interesting to see where it goes.

Absolutely. And I think one of the things about sci fi is that it is all us nerds who love us, who love the sci fi and it's those nerds who often end up building a lot of this technology. And what it does is focus our attention to a point and again allows us to come up with that. That process which we talked about earlier is starting with the problem rather than the solution. Right. So we go, oh, you know, hey, we want to be able to have, you know, face video calls. We want to be able to have jets and cars and, and the more we're wowed and impressed by that, the more we focus our minds on solving that problem. And it just shows the power of what humanity is capable of when we focus our attention. Right.

That's totally what it is. It is rallying people behind a specific vision. And TV and movie does that really, really well. But also in terrifying ways too. Like when you talk about Terminator and stuff like that.

Yeah, but we are, believe it or not, at time, as quickly as these episodes fly by, they always make me wanting to chat with my guests longer. But I do want to be respectful of your time. Jason, I really appreciate you coming out today. Any final words or suggest or do you want to throw out your handles for how people can connect with you on on YouTube and Twitter and and LinkedIn?

If the people are on social and they haven't been bothered inconvenienced by me yet, then I don't know how they would have missed me. But I if you're not, I'm on pretty much all social media. Just search my name, you'll find me there. Parting wisdom man. Just like go out and experiment, like be curious. Make some like connect with some friends that are curious. Also like make a group text, a slap channel, something like that. Who are some people you know that are experimenting and just share the journey in a fun way.

Curiosity. My favorite word. Awesome. Thank you Jason. I appreciate it. Everyone else, we'll see you guys in a couple weeks. Bye for now.

Pretty good, right? Andrew does these conversations every single Friday down in the show. Notes I'm going to put a link to how to find him, how to find his show. If you love the AI stuff that.

We do on this podcast, you're going to like the stuff that he's doing.

Thanks to Andrew for having me and I'll see you on the next one.

Sam.