032 What Does Post-AI Accounting Look Like?

Hey, it's Q and A Wednesday. We're gonna talk about why my robot broke yesterday. We're gonna talk about whether you can run a firm inside a firm, about what the future looks like for pros who aren't like putting out content and creating context in an age of AI. And what if you don't wanna do all that? What if you just wanna be the doer? What if you just wanna make stuff and prep returns and do accounting? Is there a place for that in the future? We're gonna talk about it today on Jason Daily. Okay, first order of business. If you watch on YouTube yesterday, you would have seen, if you watched in the first half of the day, you would have seen what I uploaded was an unedited version of my recording. So there's that. The immediate terror was when I made that realization. Was there embarrassing things in there? Was there nose picking? Was there clothing changing? Who knows? Sometimes these recordings get pretty sweaty. It didn't end up being that bad and it's fixed now and the podcast feed version was fine, but whatever. This stuff happens sometimes. But a couple people ask questions about that because the way these recordings work is if I need to like think about what I'm going to say next, I will pause and not make any noise so that when the editing team is working on it, when you have that sound visualization of what's happening, it's really clear that there's dead space there. Like if I'm like going, what am I going to say next? And like hemming and hawing, then when you see a 30 or 40 minute file, there isn't like a really obvious gap there so that there's no audio. But when people watch this back, I think the impression was like that I just had, would have like a 15 minute existential crisis where I would just sit there motionless, completely silent. Listen. That's, that's how these recordings usually go. So that happens every day. We just usually cut it out in post. But I'm doing fine. It's mainly just to make the jobs of my editors easier. So whatever, that stuff happens. You know what, like publishing a show like 30 times in a row. I think we're actually doing pretty good. I haven't embarrassed myself too much yet. How about this? An old timey accounting firm, what would your advice be? This was me at one point in time, like looking around at the firm I'm at and I got a little bit of perspective and realized, oh, okay, I'm living in like 1983 here. But I didn't Initially. Know that because when you're a fresh baby accountant, like the only perspective you have is where you are. Right? So I'll read this. I'll keep this anonymous. I'd love to hear what your advice is. I currently work at a mid sized firm and plan to buy into a tax practice in my small town later this summer. The practice consists of two overworked accountants and a few part time admin staff. So this is the firm they're talking about buying or buying into. Maybe they're not buying the whole thing. As far as technology goes, this place is at ground zero. No document management system, all paper. They don't keep time at all. No practice management system. No way to get stuff from the clients electronically besides through a single Gmail account that the office shares. Oh, run taxes on Intuit Pro Series. I could go on and on. The two full time owners are on their way out. They're not retiring right away though. And I don't want to run them off by trying to implement too much too quickly. Not to mention that the clientele is very used to things being run this way. Any ideas on the best way to approach this? Thank you so much. You got a. You got a project on your hands, my friend. Yeah, I mean, you hit on one thing there with client expectations too. So it's one thing to be the initiator of internal change, it's another thing to transition a client base to a completely different way of working. So, arguments for doing this? Arguments against doing this? I'll. I'll start with the arguments against doing this. Buying an accounting firm for someone in your situation right now is a sometimes a costly solution to what is a short term problem. Because it's never been easier to find accounting work. And I know it doesn't always feel that way when you're starting out, but if it's a 6 or 12 month burn until you're feeling comfortable, you gotta think about like, what am I going to sacrifice that could be like a forever sacrifice to get me through that six to 12 months? I do think oftentimes people give away maybe a little too much just to get over that hump. That will be uncomfortable and will be hard. But for the rest of your career, you will never have an issue having enough work again. So going in and buying a legacy practice, that's a tough one. Particularly when it comes to people who are quote unquote on their way out with no explicit agreement as to when they will be out. I've seen a lot of these deals fall apart ultimately because unless you have a hard and fast agreement as to the time frame of when they will go out, why would they leave? Steve's going to buy me out whenever I want to leave, whenever. What's my motivation to leave when I can take a paycheck every year until I go. It creates this really uncomfortable sort of hokey pokey where you end up being reliant upon them leaving, like either because they're afraid of you leaving or just like out of this kind of goodwill sort of thing. So if you did buy into this, I think the only way you would want to do that is if there was an upfront, agreed upon timeline of when those people left. And it wouldn't have to be, you know, a single year nailed down. It could be a sliding scale of, you know, what happens if it happens on any one of five years or something like that. But I would not go into a deal like that without a clear agreement as to when they will go. Because if you don't have that, there's virtually no incentive for them to ever go. Even if they're not working in the firm, even if they're not doing that much client facing stuff, you're relying on them just getting out of the way and not taking that paycheck anymore just to do you a service, which puts everybody in a tough spot. I do think when we are starting out new firms, something we often overlook is the fact that buying an existing practice, even if it's not a perfect fit practice, can be a way to get you over that cash flow hump short term. So like an example of somebody who's done that is Logan Graff, who's talked about how much his firm has made the first couple of years online. And he bought a small book of business in the beginning and has cycled out a lot of those clients. But those clients paid the bills from day one. If the alternative is to go out and not have a client base or not have much. I know I said sometimes we can make sacrifices in the short term that don't make sense long term, but I do think there's scenarios where you go out and buy a small book and finance it and the whole thing, cash flows, and those clients basically pay for pay for you to go out and have a little more time and space to find your perfect client that you build, where that, that client, that practice that you're building organically that maybe is better aligned with what it is that you want to do. So is that an option here? Like, is buying this better than starting from Nothing, maybe. And it doesn't mean that you don't ultimately cycle all those clients out. I know that can be hard to do or seem nonsensical because you just paid for them, but if those are the clients that will pay your bills in the short term, that's also totally fine. I think a lot of us can be guilty the more we're. Especially the more we're plugged into think boy ism and like, kind of the, I don't know, the pressures of building a cool modern practice. A lot of us are feel this urgency to change a lot fast. And you put a lot of pressure on yourself to make that happen when that's usually not the best thing for a firm. So I recognize that can be a struggle, especially if you're coming into a more traditional firm. I would say, here's permission to run a traditional firm in a traditional way. You have paying clients and you gotta think about, like, what does the version of this firm look like that's built just around me? And if the stepping stone to get you to the firm you're really proud of is doing some traditional work for some traditional people along the way, that's totally fine. Like, there's nothing wrong with that. But biggest red flag here is you need to have a really explicit agreement with the existing folks so that you know when they're going away. Otherwise they have like, no incentive to go away. But you're. I think you're barking up the right tree, getting out of regional and doing your own thing. Like, I think that's. You just. You gain so much more control over your life and, and freedom in how you approach this stuff. And one way or another, you're doing the right thing. If you got different advice, be happy to see it. Drop that in the comments or even like, if you've been through a similar sort of transaction. I've been through a couple. I've been through two, and both of them there were major issues around when is that person actually going to go. Even if I did want to serve their traditional clients in a traditional way, financially, what got complex was what incentive do they have to go? John Sanchez dropped a YouTube comment. I think this was in regards to AI and developing AI around your own proprietary context. He said the larger context around this discussion, what. What does the future look like for pros who have nothing to say, create or contribute to the ether sphere? Thinning of the herd, coming hard and fast. I do think there's been a trend, like, for as long as I've been in the profession. I think people have been saying there's less and less space for the number crunchers and that job will continue getting smaller and smaller. I think the biggest thing that Covid kicked off was not just distributed work, but the globalization of work. I think that will be the longest, the biggest long term impact of COVID as people were like, oh, if I can work remotely just fine with people across town from me, why can't those people be in the Philippines or in India or something like that? And the globalization of work also I think reduces the pool of people who stateside can just sit there and do the technical stuff so edit like it is a, I don't know, it's one of those things where there, there will be lots of exceptions to that rule, I'm sure but like thought leadership, launching a firm, running a firm, all of that stuff. I think ultimately longer term is going to sit on top of your own sort of proprietary context. And that does start with creating and documenting. I, I think sometimes we, we look at like quote unquote creators or quote unquote influencers and that's our, maybe the version of creating that we see and that probably doesn't resonate with a lot of people, like just kind of all of the sort of slimy things that come with that. And the fact that you know, most people that are running quote unquote creator businesses, like ultimately there's a thing they're trying to sell you on. And so I know that leaves a bad taste in a lot of people's mouth in this case when it comes to creating context that AI can see into and whether that's enabling, you know, chatbots that your clients can, can talk with or enabling an AI agent that can see the sops you've developed and then go be able to do those tasks autonomously because you've built that context and documentation. I think that future version of creating context is like more pedestrian than somebody you would look at today that has a million subscriber YouTube channel and like it's not like really influencery type stuff as much as just simply being well documented will be beneficial. And I do think there are folks who are like the back of house, the people that just want to do the number crunching who can be documented because that documentation, it doesn't have to be a highly produced YouTube video, can literally just be a text text, it can be a voice memo, stuff like that. So I would say, I would say the future looks harder for pros who yeah, don't have maybe domain specific things to Contribute because the mass market AI tools like your ChatGPTs, they will do the general stuff very well. Like the accessibility and connectedness of information for those general situations will never be better. Like doing bookkeeping, doing basic tax, all that stuff will continue to get easier and more accessible. So it's probably the folks who maybe don't have an interest or are not positioning themselves to have that specialized expertise. Like if you, if you work with Chat GPT right now and you don't know anything that Chat GPT doesn't know, then you're kind of in a bad spot. Right. Like it's, and it's not the first time we've had conversations like this. If you didn't do tax returns in the past that TurboTax would do, I said that the right way. If all you did in the past was tax returns, the TurboTax could do for you. And this is going way Back to when TurboTax, you know, desktop software came out. Then you just weren't positioning yourself in a place that was going to be sustainable longer term. So yeah, I do think we have to be mindful of the expertise that we're developing. The more specific it is, the better because like kind of that groundswell of how good general AI tooling will be will keep getting better. And the context that we need to then layer on top of that to still have value to our clients is something that looks more specific. Derek Foot would be interesting to hear or find more about the costs of having our own custom AI or the proliferation of these things. The financial cost to a practice and the energy consumption, all the computing power requires. What sort of impact does that have, if any? Yeah, the, the cost of developing your own stuff internally, like that's a hard, it ends up being a hard yardstick because there's so many different versions of, of what AI could be. Right. I think the systems that like see into everything and are really helpful to us and see across our documents and our emails and all that stuff, I don't think anybody's ultimately going to be building that themselves. I think that's something we're going to rely on our practice management systems to do because to build something that will see into all of those sources of context that are dynamically changing, like meeting transcripts and projects and tasks and all these different things, free to build that yourself probably isn't really realistic today. So that's where your practice management system is going to help you out. The more expertise, specific applications, like building a chatbot right now, like that's pretty Cheap and accessible. The two two tools that I shared on that Chatbot episode was ask my AI or myaskai.com and then chatt thing IO. What I've learned with more testing, my ask AI is built on GPT 3.5 chat thing at the middle tier, which is 50 bucks a month. Everything ask my ask AI is cheaper than 50 bucks a month. Chat thing at that middle 50amonth tier, it's on GPT 4. And the big difference there is GPT 4 will reliably only answer questions with a given set of contexts, whereas GPT 3.5 will not. So if I give GPT 3.5 the transcript from this podcast episode and I ask it how do I make ice cream? It'll tell me how to make ice cream. If I give GPT4 just the transcript from this episode and ask how to make ice cream, it'll say, sorry, I don't have that information. And that's a big threshold because we're all afraid of AI giving unexpected answers. And that's where ultimately we can't just blindly trust it to go do its own thing. But we're there with GPT4. So to build a, like a client facing or an internal team facing AI that is just sits on top of the context you provide it that's there right now. The best version of it that I found is Chat Thing IO. Like if I were to build something tomorrow, that's what I would use. That's 50 bucks a month for the GPT4 version. There's some limitations there in terms of the volume of messaging that can happen because Chat thing is paying OpenAI for every GPT4 prompt that a user, you know, submits to it. So it has some limitations in terms of like the volume of messaging that it can do. But that flavor of building your own AI today, man, that actually got super accessible super fast. From a cost standpoint, when you think about the fact that six months ago we didn't have any of this stuff and now you can build your own chatbot on top of thousands of pages of your own context for $50 a month, that's pretty cool. The flip side of this, specifically energy consumption and, you know, computing power that it requires and all that stuff. It is a really big important question with AI. Ultimately it became a massive problem with digital currency and like the inefficiency of Bitcoin and like the outrageous sums of energy that all that stuff consumed. And that got better over time. Like Ethereum, you know, they did updates that made that more efficient and stuff like that. The big risk with AI one one school of thought right now is that you will have this quote unquote frontier AI model that is a single model that is more powerful than all the other models because it has more parameters, because it's bigger and because it's better than all the other models. Everyone will use that model and keep spending money to use that model. And over time that's like this self reinforcing loop that you end up having this single frontier AI model that just gets bigger, bigger and bigger and bigger and bigger and more powerful than anything else because it's the best and everybody's investing back into that one and nobody will ever catch it because ultimately like everything on the planet will run off of this one AI model. That's one school of thought. So like an AI company called Anthropic, that's, that's kind of what they've said. They've raised a few billion dollars towards building like a GPT4 alternative. OpenAI. On the other hand, Sam Altman, their CEO, believes that the, he said that the age of the large language model is over and that there's diminishing returns for a bigger, larger parameter model and that the most meaningful things going forward will just be like tailored applications of GPT4 type models. So depending on which of those two timelines happen, there's that first timeline could definitely create this cascading energy consumption problem with this model that gets bigger and bigger and bigger and more powerful. The latter isn't like, the latter scenario, if that comes to bear, isn't going to involve like the same environmental impact in terms of the impact of like making, of requesting things from the model. Because that's like, that's the environmental impact of creating AI models. The impact of requesting things from the models. Right now it's people don't like have a great handle on this right now they estimate the cost to OpenAI is about half of the cost that you would pay to make an API call to GPT 4 or 3.5 or something like that. And so like there are vastly different efficiencies in AI models, just like there are in different, you know, blockchain digital currency technologies. Right now GPT4 costs about 10x what GPT GPT 3.5 does for, for using that model. So that's a harder one to measure right now. I mean they released an update for 3.5 last year that made it like literally 10 times or it costs a tenth of what it did previously. They made it way faster, way more efficient. So it's kind of thing that you ought to expect would get better over time. The last thing I'd say on that is we're also seeing really fast development right now of open source AI models that are pretty darn good and you can run locally, which means they're wildly efficient, much more efficient and consuming less energy. And a lot of this has interestingly been enabled by training these new open source AI models on the state of the art models. So like using GPT4 to, to train an inferior model to get it up to a level that is something closer to 3.5 or something along those lines. And there's versions of those now that you can just download and run on your computer, which 12 months ago like would have been ridiculous to think you could get something that powerful and run it, run it locally. So it is a super important consideration like what are the long term impacts of this? And it's one of the reasons why like, like regulation is going to have to come in and get involved in AI. You can't have this, like there can't be something that is so good that you just don't even stop to consider what are the negatives of this thing. So it'll be interesting to see how that develops. That being said, they have got wildly more efficient in the last year. So it'll be, it's probably come down to like whether we end up just using these more and more absurdly powerful models or whether we consolidate around like a language model that's like good enough to do everything that we need. CJ, Are you creating your own program to make your personalized chatbot or using one of the programs released? I've done both now. So I built a tax research product and I'm not sure if I'll ever ship it or not, but it was kind of a proof of concept where we fed it all the IRS publications and it was basically a much, much better version of Control F in a PDF. So you could type into the application a question, it would search across all the IRS publications, it would give you an answer, it would give you a list of exceptions to that answer, it would give you another tab of quote, unquote, other considerations, things that are kind of adjacent to the answer that it gave you. And for every single statement that it gave you, it gave you a deep link into the exact line of the source publication PDF so you can see where that information came from. So ultimately it gave you an answer. But the idea was can an AI based tax research tool just make it Faster to get to all of the source material in a more intelligent way, rather than just like, text matching search, which is what most of our research tools do right now. And it's pretty darn good. But that's where I learned about context and, like, chunking and conditioning the source material. Like IRS publications have a bunch of visual tables. Well, obviously that's not searchable. So you have to then process that to create, like, a semantic explanation of every possible path through this, you know, worksheet or table, that sort of thing. And then when you search for it, it'll actually find something semantic that it can pull into the search result, as opposed to just a picture, which is meaningless to that technology. Right. So that required learning everything that goes into that, from embeddings to LangChain and chunking and stuffing and all of that stuff. And, like, a lot of fiddling with it to optimize it. And having done all of that, I would use one off the shelf. So, like, if I built my own chat bot, if I built Jason Bot right now, I think I would build it with chat thing IO. There's probably a whole bunch of investment and work that could go into dialing something in that would work really good for your use case. But the biggest difference I see right now between AI chatbots that work really well and the ones that stink is actually the conditioning of the underlying context. When you upload a PDF file to this chatbot service and we'll use an IRS publication, for example, those publications have split columns, they have page breaks. And the way that those programs ingest that information is usually they'll be chunked on those column breaks or on those paragraph breaks. Because the IRS documents, believe it or not, are not structured in a very helpful way. And oftentimes that'll give you the right. The wrong answer. Because, for example, like, those page breaks and column breaks happen smack in the middle of a sentence. So what do you do if. If the answer is, like, kind of straddling those things? So I think the way to make most chatbots better right now is having better conditioned underlying data. So, like, a better version of an IRS publication is just a continuous stream of text, text, a text file with semantic text added for tables and worksheets that you put in there that'll make the chat bot work really well. How it's chunked, how many results it takes into account in the answer are smaller issues, I think, than how the underlying, like, source material is conditioned. Derek, how do you reconcile the idea that our expertise has value that should be paid to access or utilize with the idea of having that expertise inside an accessible AI rather than the person. Matt's crypto consulting example shouldn't come for free if it's sitting in a robot. And Derek did follow up, because I did later say I would want to pay Matt for access to that bot. Yeah. So we've been talking a lot about what is your. What is your proprietary context that you. The stuff that, you know, the stuff that you're able to do that ChatGPT can't do, basically, and how do we better document that context and give people access to it? Because the future of communication, I think, is probably not just sitting down and talking with me across the table or nothing. I think AI enables a version of that that's like in the middle, like. And this, I mean, this shows a great example of something that anybody can sit down and watch, but it's not answering your specific questions. And so now I have all this context from transcripts of shows that you could build into a bot. And being able to chat with that bot that has all my transcripts, that's a heck of a lot easier than going back and watching every single thing I've ever made to find a very specific answer to a question. And that's an example of where AI is creating this version of communication that still isn't the real deal. It still isn't us sitting down eating sub sandwiches together, but it's a whole lot better than anything else. Right. And so I do think there's a version of professional services where clients are able to use this chatbot and it'll answer 50% of their questions without them having to bug you. They can use it 24 7, but you have to charge for it. And like, Derek, this is. I'm, like, paralyzed by this right now. And there will probably be a timeline where I have something like this, where I release something like this. It's a very hard thing to be, like the first, the whole, like, the notion of releasing a chatbot of yourself right now is so sci fi. It's doable and it would be valuable. But, I mean, you look like a, like a total, I don't know, idiot or narcissist or something like that. To be like, hey, you can now pay for access to my wisdom. Like, I absolutely think that's the future that we're going to. But if you rolled that out tomorrow, I think people would laugh at you, like, right or wrong. Uh, but like, what do you charge for that? That's another paralyzing thing. About like making that decision, should that be? I was so using Matt as an example. Matt Mattress is like a crypto tax expert and he does. People pay him to come do talks all over the country. And like he's really, really, really knows his stuff. So access to matbot, should that be astronomically expensive? Almost like a fractional employee where maybe you're paying every single time you prompt it, or is it like the sawdust of the stuff that you're already doing? So it's like, well, I've already got all this stuff, so like it should be pretty cheap, right? Like it's somewhere in that spectrum, but. But I don't know where. The risk of giving access to that context away for too cheap is that ultimately that long term, like, that is our only value. Like, what are the things that we know that AI doesn't know and what are the things we can advise people on that AI can't advise people on because of this specific expertise that we have? So if you give that away for so cheap, like, that's like one of the last things that we have when technology gets better and better and will handle more and more for us. So all that is to say it's too early to call. I think I'm excited for somebody to ship theirs. I think I wouldn't be surprised if influencery types start leaning into this more as a way for. I don't know, it depends on the type of influencer. Like there has to be some subject matter expertise there rather than just entertaining them to give people access to that chat experience. I suspect we'll see it sooner than later. Personally, what excites me about that, what I think is really cool, is there are things I would talk about in this show and then there were. There are things that I would not talk about in this show that would still be valuable context. So like there's a version of it that's just rolling up all of your public stuff that you have. And that's useful because it's easier to find than going back and watching, you know, every video. But there's also a version that goes deeper into more technical stuff around certain domains where you probably would never actually, like do a podcast episode about it. And the only place that context exists is like under that chatbot. Like that's the only way that you can get to that context. So, like a funny example, me and Chad Davis are giving a talk at AICPA engaged. It's like, steal these 10 automations to save yourself a pile of time. Or it's Got a title, Something like that. And we're going to run through 10 automations that anybody can do that are good timesavers. The Z AICPA engage in June. I'll spoil something here because most of y' all won't be there live. We're building a bot around that talk. So the whole talk we will pre record a transcript and some supporting documentation around all the Automations that we show off. So we'll have all this context and there will be a chatbot sitting on top of that context that people can ask questions of either during the session or after the session. So there's a chatbot that can like answer a bunch of specific questions about how to do the things that we're talking about. People can chat with that during the session. They can go home and try to build this thing and chat with this thing further to troubleshoot it. But we can also put behind that chatbot like her product information and like any sort of like Easter eggy or, or other like adjacent information that you wanted to, you could build into that chat bot and the only place that it would exist is if it is queried through the chatbot, which I don't know, I like. I just, I think that's fun. It's really interesting. I'm excited to see how all that stuff develops. Mo have you ever seen a firm within a firm work where you can grow your practice and use the resources but not feel like an employee and haven't an exit strategy? I've tried to do this. This was my accounting practice that I built at the firm that I ultimately bought. And the biggest things you need are independence and decision making and how you build the team and financial independence. So you really need, if you're going to build this sort of new practice within a practice, it can be done, but it requires the commitment of everybody to really give you that independence on your decision making and, and in how that business is managed financially. It can't be coming out of a shared pot. You have to be just as accountable as, you know, the rest of the business. Where that gets a little bit stickier is when you have shared clients. So for example, if you've got a tax practice, but you're going to start an accounting practice within that firm, the folks that serve those clients in the tax capacity will be opinionated on the right way to do accounting for them. So that gets a little dodgy. Also, how you allocate revenue between the tax practice and the accounting practice can get a little problematic, especially if you are, you know, the most common cross sell from tax to full back office and accounting work for a small business is like hey, you already pay us, you know, 7k a year for all your tax work. But it's because it's a mess, we do a whole bunch of cleanup for you, we'll do the whole back office for you for 12K. And that tax return will take, you know, a sixth of the time or something like that. When you have what are kind of two businesses operating independently, how do you allocate revenue for the tax part of the work to the accounting part of the work so it can be done there ends up being a lot of sticky things you also need to be mindful of. If you're not an owner, what does it look like to then buy this thing that you just built? So like in my case it was build an accounting practice but I still didn't own it. So like you're building an asset for somebody else and if you're ever going to own that asset yourself, are you then buying it at full price from that owner? So it can kind of be done and there's upsides to it too. Like for me, building an accounting practice when we had a 1500 client tax practice was like shooting fish in a barrel when it came to finding clients. Like we could hand pick all the people that we wanted to do that stuff for. So we never had to do any marketing, anything like that. That was really nice. But downsides to that approach also Priya, is there any way to remove pii that is was that personally identifiable information or something personally identifiable? If I restaurant. Yep, I got it right on the first try from bank statements. Sometimes transfer accounts are listed. Some small banks do it before pasting it in chat GPT. So I shared a an automation something I do with Chat GPT where if somebody gives you a bank statement but not a transaction download, you can copy paste the transaction area of the statement into chat GPT and chatGPT will give you a CSV download. Like it is super easy and super reliable and usually what I'm copying is just the transactions and so some of the statements like Priya said, especially with the older or smaller banks, they'll put like a full account number on the statement. Most of the bigger banks don't do that but generally the transactions themselves are just the merchant descriptions which are the same for everybody. So it's not really a big deal to copy it. But Priya saying here, sometimes there are transfer accounts listed there so you have the full number of the transfer account. What I like about this, like, ability of GPT to grab this information from the bank statements is I can see 100% of what it can see. It's just what I'm copying and pasting. Like, there is no ambiguity around exactly what information it has. So if you run into those situations, like, for me, it's probably just an added step where maybe I first paste it into a text document or Excel or something like that, remove or obfuscate the stuff that I don't want it to have. Like, maybe I change it so it just shows the last four or something like that, and then drop it into ChatGPT. So is there. So ultimately, to get back to the question, is there a way to easily remove PII from the bank statements? No, not to my knowledge. Because if there was, it would involve, like, giving that full information to another service, which is what we're trying to avoid to begin with. Right. So for me, when I pull that stuff into ChatGPT, it's just a matter of paying attention to what I'm pasting in and pulling out any stuff that isn't sensitive. I get a ton of ChatGPT security questions. So a couple things to remember. Anonymized information generally isn't pii. So under gdpr, for example, even under hipaa, under California's data privacy rules, if that information is anonymized and not able to be tied back to a person or a business, it's not pii. So medical information, for example, like that can be shared freely if it is not associated with a person. Like that is according to the hipaa rules. So 99% of the time, the solution is just pull out anything that's identifying and you're good. That being said, ultimately the coolest AI stuff that we will do will contain client info. Like, that's, that's just the reality. And so an example of that is Carbon's new AI stuff, where you can now you generate a email to reply. You can use AI to generate a reply to an email. And so that means what they're having to do then is send the original thread to the language model to inform how it suggests that reply. It's really cool stuff. We did an episode on Carbon's AI stuff last week, but it involves sending client information to an AI model. And this is the first time we've crossed this hurdle. And it's a big deal because ultimately the most useful applications will do this. Now, one thing Carbon's doing differently and how they work with OpenAI's models is they're using Microsoft Azure's OpenAI models and with that rather than just sending it to OpenAI's API, with Azure you have a little bit more control over how you can work with that model. And it's a little bit more secure of an environment than just making API calls. The normal way to open AI. But all that being said, like there's, there are tons of ways that AI is being used with sensitive data right now with medical information. The most sensitive applications right now are using Azure rather than just OpenAI's API. So if you see your tech partners like building stuff on top of AI that uses client data, it's fair to ask like how does what's happening behind the scenes there? But there's absolutely perfectly secure ways to do it that are compliant with like HIPAA rules and all sorts of stuff. Like not like anonymization aside, if you anonymize the data, it's not PII anymore and that gets way simpler. But even with stuff that isn't anonymized, there's absolutely secure ways to do it right now, which is awesome because like it's going to make all of us way more effective. Okay, I teased last week a mind blowing AI episode. We're gonna do it tomorrow. It's happening and along with that is a kind of release that a practice management system did. But they did. It was like marketing, but they haven't actually shipped it. So I don't know that I can actually fully give them credit for that thing yet. But it is on another level to anything else that we've seen yet in terms of AI releases in the accounting ecosystem. Tomorrow we're gonna do deep dive on it. I'm trying to limit the AI nerdery to like one or two episodes a week because I know it can be a lot and we still have to be grounded and hey bud, that's really cool but I still can't respond to all these emails and my clients don't pay me enough. So like, as exciting as AI is, like we also still have to do our jobs every day. So we're gonna do that tomorrow. Thanks for coming and hanging and I'll see you then.