608 6 Claude Cowork Use-Cases for Accounting Firms [Accounting firms are actually doing this!]

My friends, we are in the thick of my workshop tour right now. I'm visiting 20 cities in the US, throwing half-day parties, continuing education, snacks, a dinner, an open bar. These are events that I have built with my team from scratch. We host them and we pull in 50 to 150 local accountants who all want to hang and hear a couple of talks from me and meet each other and just vibe, you know. These are unbelievably fun events. We are in the middle of running them right now. If you haven't checked out the tour schedule yet, please do head over to JasonOnFirms.com/tour. /events to see the list of all the cities. If you know somebody that's in one of those cities, give them the heads up. It's a super fun hang and I want to see you there. Okay, on to the show. Enough about your workshops, let's get on to the actual episode. So I've sourced several hundred AI use cases from actual accounting firms in the last week, and today I'm going to show you 6 of my favorites. These are things that accounting firms are actually doing in practice every week. Not hypothetical stuff like ways that they're using Claude to save themselves a pile of time. Plus, can Claude automate the headache that is fetching all the documents for a month-end close? Logging into this thing, logging into that thing, fist fighting this two-factor authentication. We're going to talk about it. Plus pricing struggles, knowing what you're worth, deciding when's the right time to niche your accounting firm and how to handle AI hallucinations inside an accounting firm. There's so much stuff. So much stuff we got to get to. Hurry up. Roll the music. How do you do, fellow accountants? My name is Jason Staats. Welcome to the Jason on Firms podcast, where we talk accounting firm stuff. Pretty cool. It's a hit at parties when people ask me what I do. But here's the thing. Right now I'm smack in the middle of a 20-city tour all over the United States of America. And in each city, I'm asking the attendees to submit the AI use case that they are most proud of. And the result is I have a treasure trove, hmm, a veritable treasure trove of real ways that firms are using this stuff. Today on the podcast, we are going to go through all of these. If you're new here, welcome along. This is a podcast. Most people listen to it with their ears. If you're watching on YouTube, I'm gonna explain these things. You're not gonna see 'em with your peepers. I've got another YouTube channel where we do that, where we make videos that are more peeper first. In fact, let me know. I'm thinking about just doing a series on the main channel, the YouTube channel, that where just every video is just like 4 to 6 use cases, building them in Claude, like use cases that I've seen from actual firms, just to kind of get you comfortable with throwing stuff together as you need it. Would that be helpful? Leave me a comment if so. Thinking about giving that a try. But enough foreplay, let's get tucked in, big boy. 6 AI use cases that I love. Number 1, talk to someone who takes a video, like records their screen, not a video of their screen, like uses a screen recording tool. They said they then have Claude have Claude watch it, and Claude creates a step-by-step SOP of all the clicks, all the entries they did in that video. In their case, they're doing some complex thing in NetSuite, which sounds pretty gross. NetSuite is the worst. But Claude, when they say they have Claude watch it, I don't know about you, in my head I think like Johnny 5 sitting in front of a screen. That's not what's happening. I imagine what Claude's actually doing when you give it a video is it is grabbing frames from the video and seeing what is happening frame to frame. But this one kind of blows my mind because in order for it to build an actual correct SOP, which by the way, they said it nailed it, like every detail. In order for it to correctly build an SOP, it's gotta look at so many frames, so many images in that video to understand, oh, this just got clicked or this just got typed there. Now Loom, which is what I use for recording most of my videos, Loom, L-O-O-M, they've got a cool built-in feature that will sort of build an SOP for anything you build a recording for, but it just works off the transcript. This, what they're describing here, is making a screen recording with no narration or anything like that and just chucking the screen recording in a cloud and saying, "Make an SOP." And it could do that with frames from the video even. Remember Three Ring Binder person? If you came up in the accounting profession, there was someone that was extremely proud of their three ring binders and they would make these massive like guides in Microsoft Word that have screenshots and words and all these things. Is Claude new Three Ring Binder person? Might be, hot take. Okay, second use case I thought was really cool. They're using Claude to consolidate 10+ entity financials into a unified reporting package. Consolidated reporting, long been the bane of accountants' existence. It means the chart of accounts have to be just so, and there's not really great tools for this stuff. Oh, you look at like a, you know, a CCH engagement and it's like, okay, cool. Like once you finally got that all set up, it kind of works. Claude though, the magic of Claude, is kind of just like chucking stuff into it and being like, do your best, champ. And then it comes back and you're like, holy geez, that's actually pretty good. So whether the chart of accounts matches or whether you even tell Claude like, hey, create like account groupings that make sense for these accounts because they're not going to line up perfectly, Claude can do that. Now, something we don't talk about much when it comes to Claude is the notion that it can like build entire apps. And because we're not software folk, I'm by no means an apps man. That's maybe a part of Cloud that we shy away from. But when it comes to financial reporting, Cloud is very good at making like little visuals to give you a balance sheet, a P&L, a trial balance like that's even drillable. Click on the number and it will give you the detailed reports. Cloud can build that stuff in its sleep. And so if you combine the fact that it'll do consolidations, you just chuck that information in. With the fact that it'll do reporting. Like, you can basically build a, a custom little one-page app to present that data however you want. You could even take this a step further and do like eliminating journal entries. In fact, Claude can probably suggest eliminating journal entries, so you may not even have to do them yourself, you big lazy bum. But this is, to me, this is a great use case. There are obviously dedicated tools for this stuff, but when people talk about Claude potentially replacing software and replacing apps, in my mind, the most likely things that it replaces are these sort of peripheral apps that do a very specific job. And like, when you got a few companies and you just have to roll them up, and you got a very basic app that does that, or you don't want to spring— like, cut a big check for a big complex app that does that, cloud makes total sense here, right? Okay, use case number 3: a dashboard for crypto clients— you ready for this?— to optimize tax loss harvesting. That was a sentence. A dashboard for crypto clients to optimize tax loss harvesting. Several things in here that I love. You can actually load in like, what is the client holding? Like, what do they have? But then also wire that up to like live pricing information. So when, when it has both, it can then like dynamically update in time to be able to tell the user what is the tax impact of this and identify times and, and then situations where it, it might be worth taking a certain action. And if you did, visually understanding what the effect would be. How cool is that? Imagine what an unbelievably premium thing that is. Most people are like, I would just be happy if my accountant called me back within 2 weeks. Imagine being the accountant that's like, hey, we know all the stuff that you got, we threw together this dashboard for you to help you understand how this tax loss harvesting stuff works. You do kind of scenario analysis, stuff like that. Give it a look, could help you get a better understanding of, of sort of your crypto holdings. Man, that is so cool. And this is one of the things that is maybe easiest for Claude to right now is sort of these dashboards, because dashboards are kind of, uh, they're absolutely everywhere. So it's something that Claude has seen a million examples of. The novelty is that dashboard being tailored to such a specific application and being something that's like just so useful for your situation, especially if you're the pro and you can layer in some cool logic to help them understand some larger, more complex topic and, and how it applies to their situation. I mean, there's a bunch of examples of how you could do this. When you think about like, uh, cost segregation, you think about QBI stuff. There's so many complex concepts that could be simplified with something like this. Use case number 4 is a big one. Tax checklists. This firm has a checklist that they always run through when reviewing a tax return. Now AI does it. Just, just whether you do tax or not, imagine 5 years ago if somebody came along and they're like, in 5 years you will have a magical machine that you can put any checklist in front of and it will go out and it will, it will do the checks. It will look into that stuff for you at practically no cost. You would have been like, phooey, what is this sorcery? I don't know about you. When I came up in this profession, checklists were all the rage. Accountants love their checklists. We had, we had PPC checklists for doing a test work. The guy that I worked for had this, this like 21-item tax checklist that he, that he came up with some catchy name for. And he would tell clients all about and all about it. And he's like, oh man, this, We got this proprietary process that saves people so much money. In reality behind the scenes we're like— We went through the checklist once and then every single time after that we're just like blah la la la la la la, writing your initials as fast as you can possibly write them. But if you let's say you are just a real whiz and you make this novel 50-item checklist where you're like these things are so important, we need everybody to sign off on them. Imagine putting a 50-item checklist in front of a human being in 2026. Imagine! Plopping that 50-item checklist in front of a millennial, or even worse, in front of a Zoomer. They're going to be like, you kidding me? I'm a fully actualized human being. I don't complete checklists. But an AI? That eager little beaver. You throw it whatever you can at it. I mean, put a 500-item checklist in front of an AI. That's actually a bad idea. I'll tell you why in a minute. But you can throw 500 items in front of an AI and it'll do it. In fact, if you're like me, sometimes you give AI these just painstakingly laborious things and you almost feel bad when you give it to them. And it's like, okay, I'm just happy to have something to do. I'm going to go burn up some drinking water and I'll report back when it's done. Now I say don't give it a 500-item checklist because something to be aware of in Claude right now is when you give it a task, Claude still has basically an allowance of compute that it's allowed to apply to any given task. So the powers that be won't let you spin up all their servers and melt the polar ice caps indefinitely. When you give it a job, it's only allowed to apply so much effort to doing that job. And if you've ever had a big meaty task, it maybe struggles to do, oftentimes this is why, because it's only allowed to work so long on a given problem. A rule of thumb right now is if you have a task that consistently takes 15 minutes or more, in all likelihood, you're losing some performance because it is bumping up towards the end of how long it is allowed to work on something. And that is where hallucinations become an even bigger issue and where the quality of what it gives you really drops. So what should you put a 500-item checklist in front of an AI? Probably not. You probably ought to actually chunk that out and optimize that for how much time it's going to take it to work through each of those items. So if they're very simple items, maybe that's splitting them in two. If they're very complex items, maybe that means you actually need, you know, 100 agents to run in sequence to actually do the deep, deep work of that. That would be— I mean, at that point, the 5 things you're asking it to do ought to be humongous tasks. But that's the power of AI agents is the fact that you can chunk work out and get a whole bunch of work done, not in sequence, but even in parallel. So you can have these agents kick off at the same time to do so much checking. Honestly, the limitation is your imagination. And boy, what all could we build into a checklist? Because we now have these incredibly eager little checklist machines that ought to basically ensure that like that mistake, that issue is never made. And I know there's a degree of worry of like, well, what if it doesn't properly review the checklist item? For sure, that's a risk. But have you met humans? Show me a human that's looking at a checklist for the 70th time and I will show you a speed reader. Okay. AI, honestly, like once you've seen the same checklist a certain number of times, I would almost take the side of an AI that it's going to do a more comprehensive job than the human who is ever so sick of your checklist. Hey, real quick, do you need to hire any humans inside of of your accounting firm. I wish you could see what I'm seeing through my own two eyes right now. I'm staring at a portal of qualified pre-vetted candidates that are champing at the bit to come and work for your accounting firm. These are real human beings. Aubrey, a bookkeeper with 3 years experience, $3,100 a month. Samuel, an executive assistant with almost 10 years experience, $3,000 a month. Maycol, a bookkeeper with 6 years experience. He's got a CPA experience with, with Bill.com Accounting CS. Deal, Power BI, QBO, NetSuite. Let's find a tax person. Mia, US tax accountant, experience with CCH Access, Drake, and UltraTax. Oh, bless her heart. $4,000 a month. Gang, these are all super talented accountants that are ready to come and work with your firm through Cloud Accountant Staffing, longtime sponsor of the channel. They have a portal where you can go around and see all the folks that they have pre-vetted and even book an interview directly with those people from the portal. You don't even have to hop on a call with Cloud Accountant Staffing. All these applicants have— they've got resumes, they've got intro videos. Can't tell you how many firms that I know who have been very happy with Cloud Accountant Staffing. Accounting, it is a universal language. Luca Pacioli, buckle up for this one. Not American, Italian. You know why? Because accounting knows no borders. Here accounting, there accounting, everywhere accounting, accounting. Okay, you need some human talent in your accounting firm? Talk with Cloud Accountant Staffing. Check out the link down in the show notes. If tax season just punched you right in the grief bone, I have good news, my friend. There's a better way. We talk a lot about AI on this podcast and the group that is using AI to automate a whole bunch of the 1040 process. It is, it is one of this episode's sponsors, Soroban. With every like new headline you see around new AI models and new AI this, new AI that, the companies that are moving fastest using AI, they get better with every new AI model. And especially when it comes to the 1040 prep process, like if you're a US tax firm, oh my gosh, is there so much tedium still associated with Collecting information from clients, getting it into the tax software, delivering it. It is a nightmare. If you are a firm doing like volume 1040 work, say 300, 500 1040s a year plus, this is a no-brainer. Take a look at Sorban. We've actually done some demos of them on my YouTube channel. You can even see how it works. Definitely worth checking out. More information down in the show notes. All right, use case number 5, renaming documents. And this doesn't sound that sexy, but their example, client drops off a huge pile of receipts, could be in person, could be digital. Either way, in the end, like it needs to be digital, like you need a scan of them. But then point Claude at the folder and have it rename the files in the naming convention that you want. So in their case, it was receipts and they would have it rename it long year, long month, and long date. So year, year, year, year, dash month, month, dash date, date. After that, the payee. After that, the amount. And Claude can actually do this like up to hundreds of files in a folder. Same logic applies here that I mentioned earlier. If there's so much that it can't do it within the like allowed compute that it's given, then expect it to try to find some shortcuts. And so do be aware of like, if I put 5,000 things in front of this, no way in the world it's going to knock all those out like in a single task correctly in all likelihood. I say that, I have seen it do like some shockingly large numbers, but this is not just a receipt thing. It is in any type of documentation thing. I, man, I got to do some really cool projects earlier in my career around like litigation work and sort of discovery and putting together different data and Holy shamoly, would Claude have been unbelievably valuable for that stuff. Working through big volumes of information and then writing scripts to make that stuff easier. Whether this is 500 invoices, 500,000 invoices, remember Claude can see into that stuff very quickly, so efficiently and make things possible that you may not have even considered. In fact, you could take it further than this and like build this stuff into a folder structure. Like maybe you actually want all these organized into subfolders by payee. Maybe this is something that you're already doing for a client and you just, you want Claude to manage it going forward. In fact, in a, in a video in the past, we showed basically using Claude as a file system Roomba. The notion that you can give it a sort of protocol for here is the naming convention for all these different file types inside of our accounting firm. This is basically the only way that files can be named. Here's the folder structure. And then you can let Claude play the police. It runs around tattling on any, any bad files. Maybe you want to automatically fixing things. Now that use case admittedly is trickier because if you're cutting Claude loose on a folder right now, One way or another, you want to make sure that you have a backup of it. Claude, Cowork, the way that most of us are using it, it will still generally ask if it's making like a permanent deletion of a file, but that doesn't mean that it couldn't accidentally make a mistake. And this is exceedingly unlikely, especially if you don't have computer use enabled, which if you're on a business plan, it's not even available yet. But out of an abundance of caution right now, I probably wouldn't point Claude at like the root folder in my entire client network drive. If I'm going to have it do a task, I'm probably going to have it do a task on a copy of that data. Like just in case somehow something goes sideways, or I'm going to use a service that would let me roll back to a, you know, a what, what did, what did this file system look like an hour ago? That sort of thing, which honestly is something that we should all be using. That's incredibly valuable because probably more likely than Claude making this mistake is one of your humans making the same mistake. But I digress. This is one of those things where it can do like a shockingly large volume of things really fast. Last use case, this accounting firm, they're going after chiropractor clinics. They used Perplexity Computer, which is kind of a new AI agent, to go out and source a list of prospects, chiropractor clinics of a certain size. So they have like very specific metrics. They're looking for the type of chiropractor clinics they want to work with, doing sort of CFO stuff. Perplexity Computer went off and worked for hours, came back with 85 practices that it said they would be a fit for. That is pretty cool if you are looking to summon your inner Yukon Cornelius and do some prospecting. Now, the way they were using this was on a totally separate machine. Perplexity Computer basically takes over and controls that entire computer. So the last place you want it working is probably on your computer where you've got admin privileges to all these different things, right? In a world where you're going to cut something loose like that, you want to make sure it doesn't have privileged information that it could accidentally put on the web, but also that on that computer, if it has free rein, you just don't have apps where it can do problematic stuff. And so his solution was a Mac Mini, which you can get as cheap as $400. I've got several that I now use for all sorts of AI stuff. He put Perplexity Computer on that. All it had was a web browser, didn't have any other information about his firm. Beyond what was required for like trying to find these chiropractor clinics that he was looking for. I shared an example in our Open Claw video. Open Claw is kind of the most like free range, this AI agent that can control your entire machine and go out and do stuff on the web with no rules. I shared an example of where I gave it a name, Tick, like, like a tick mark in an accounting firm, right? And it signed up for this service that I asked it to sign up for. And the service said, what's your first name? And it said Tick. And it said, what's your last name? And it said Staats. I'm like, dang it, I didn't give him a last name, but he has disclosed who the Wizard of Oz is behind the curtain here. So an agent doing stuff out on the web for you, I mean, it will surprise you the situations that it ends up in that you just can't think of ahead of time. So if you are going to have an agent doing stuff on the web, do it from a sandboxed environment where it doesn't have access to privileged information. And if, and if it can control the computer, the apps on that computer, then like you don't want your Outlook on there. You don't want stuff that it could get into and potentially create some issues, right? Pretty cool use cases. Here's the thing I'm gonna do though, is at the end of my 20-city tour, I'm going to send all of my favorite use cases from the tour. There will probably be over 100 at that point to the people on my email newsletter. Are you on my email newsletter? Let me tell you what a hoot it is. Every Thursday at 2:00 PM Eastern, I send this email out and it is the greatest hits of stuff that I saw in the wild that week, of stuff that we put out that week, of AI announcements that I need you to see. We've done so many things on that newsletter. We once gave out a list of 350 accountants that opted into sharing that they were looking for a new job. We once put out a list of 300 accounting firms for whom the owner said, hey, I'm, I'm ready to sell my firm. Actually, at least one sale came out of that when we put that, that, that accountant list out. I know a number of hires ended up happening from that. We're doing all sorts of weird stuff on our newsletter each week. It is worth checking out. Sign up at jasononfirms.com/. Can you guess? Newsletter. Try to keep things simple. But the people on that list, they are going to get my mega roundup of cool use cases at the end of the tour. And if I'm not going to see you on tour, one, I'm mad. Two, it might just be because you haven't looked at the dates yet. Go to jasononfirms.com/events. You can see my full tour. We are 2 weeks in now, but it is 5 weeks long. We are having the time of our lives. I'm so tired, but I still want to see you. Okay. Okay, I got a great question about a kind of fringe AI use case that is the bane of so many of our existences. Let's get into it. It's Nerd Corner. Nerd Corner. Let's nerd it out. Nerd Corner. Welcome to Nerd Corner. As if the entire episode isn't nerdy stuff, but this corner, it is particularly nerdy. Here's a question. Have you figured out an AI approach to pull down documents for month-end closes? This might be the worst task that happens inside of an accounting firm is every month, the first of the month ticks over and every month-end close requires a unique combination of things to be fetched from different websites and different humans. And all these different things have to be pulled together in order to get the month-end close done. And if you are like most accountants, you're like, okay, what day is the month-end close on? Okay, the 15th. Well, all right, it's the 14th. Let's hop in and grab all that stuff now. Second thing you log into, you're like, boo, this password has changed. Something has happened. I now can't get that thing. This is such a problem in my bookkeeping practice. It got to the point where 40% of the time we were spending on month-end closes was spent on just this, gathering documents from all the different places. That's ridiculous. 40%. And this was actually because we were very efficient about the bookkeeping part. Like we automated as much as we could in the bookkeeping, which means of the time that was spent on the month-end close, almost half was going into just logging in and pulling down all these things. Now it is— it's, it's banking stuff, but it's also the health insurance invoice so that we can do the payroll accrual right. It's also a screenshot of the loan activity because they don't issue a monthly statement for this loan. It is asking Jim to send the report from that system that's on-premise for the client and we don't have access to it. Every client has a unique combination of these things. And every tech person out there, they hear this and they're like, oh, you could automate all these things. And the truth is, like, you could, like, through a combination of, like, browser automation and RPA and all these things. The trouble is, like, in my firm, there's probably 1,500 details that need to be fetched across 100 clients every single month. And the fetching of every single one of them is different. And so building an automation for every single one is a humongous undertaking. It is like death by a thousand cuts. So then building automation for every single one of those cuts becomes death by 100. I don't know what's worse than a cut. I'll leave it there so the video doesn't get demonetized. Now, the reality is AI, it is very good at navigating the web and navigating logins. Like, it's not amazing. This is probably the element of AI agents that is still least performant, its ability to navigate a browser consistently. And that's just because that's, that's a hard problem to solve. Actually, ChatGPT is probably the best at this right now, but Claude is better at so many other things. The problem is today I'm not ultimately comfortable with an AI agent banging around on the web trying to grab information for my client. We, we touched on some of these reasons in the last section, but the whole security, like net new security dangers around AI agents is the unholy trinity of the agent having access to privileged information, stuff that you don't want getting out in the wild, the agent having access to communicating with like the outside world, and the fact that it can do these things in an unattended way. So there's a world where you can ask it to do a job and 16 minutes later it reports back that it's done, but you don't know that along the way it interacted with a human there. It put something into a web form that was like, whoops, we probably shouldn't be sharing that publicly. And so these things can happen. Without you even realizing it. Now, inside of Claude right now, I'm just gonna talk Claude Cowork because that's the AI agent that most of us are going hardest on right now. Inside of Claude Cowork, you may have heard me say this, there's basically 3 ways that it can get stuff from the web. One is web search, that is summarized web search results. Two is web fetch, that is the entirety of the contents of a webpage. Method 3 is Claude for Chrome, and this is where it takes over a Chrome window and goes to town, starts banging around on the, on the World Wide Web. And you will notice that the first two there are read-only. The third one is read-write. And this introduces the risk of potentially Claude putting something out on the web as it's navigating around that maybe it didn't intend, or it going to an unsafe website that might have some sort of prompt injection to trick it into doing something that it shouldn't do. Now, these things are all exceedingly unlikely, But for that reason, in my organization's Claude co-work settings, I do not recommend enabling Claude for Chrome because that is the biggest way in which Claude would like go out into the external world trying to get stuff done on your behalf, but introducing the risk of prompt injection and it's, uh, you know, either knowingly or unknowingly disclosing a detail like Tiki sharing that his last name was Staats. I mean, this, in this Open Claude video, it was signing up for a service where I didn't want them to know that it that I was signing up for it literally put stats in there. This is an accounting software. It's like, oh boy, who are they gonna think is doing this? So agents can do this stuff and, and maybe their greatest quality is they try ever so hard, right? I mean, you could have an agent that try to go out and grab one thing and it could work on it for hours on end, like depending on the agent system and, and how long it is allowed to work on something. But I'd say the bigger issue right now is simply trusting an agent to work on the web with sort of sensitive information at its disposal. Now that will change one day. Like, this will be table stakes soon. But right now, like, that is how I advise accounting firms to be thinking about it. You probably don't want an agent doing stuff on the web unattended until we just have more trust here. Now you still have the problem. What do I do with all these things that I got to fetch every single month? The greatest app in the history of accountingdom that we've had for this particular for particular purpose is called Hubdoc. But one day it was acquired by Xero, and soon thereafter Xero stripped it of what made it magic. That was that for a long time Hubdoc was doing fetching from all these vendor sites and all these bank sites that was not like fetching via API, like programmatically getting that information from the vendor. They were building bots to like crawl the website and log in and with screen scraping and stuff like that, download that stuff pretending as if it was a human. And so they built out support for all these different vendors and they had all of these people that were constantly managing these connections and keeping them from breaking. And if you remember Hubdoc back then, it broke a lot, but it was one of those situations where it working two-thirds of the time was a lot better than it never working or having no automated way to do this. And so it made Hubdoc an amazing place to centralize all that information gathering, but it got neutered. After Xero acquired it, they stripped it back to only the information that could be fetched via API, which is vastly more durable than bots going out on the web and trying to get things for you, presumably because they didn't want to have a gymungous team of people sitting there constantly fixing these bot connections. And since then, we haven't really had a good solution for this. But I would argue whether we have a 100% solution, a 50% solution, a 10% solution, the first thing you have to have which most accounting firms don't have, is a perfect record of exactly what needs to be fetched. And so what we did after learning that 40% of our team was— our time in month-end closes was being spent on this— was we first de-skilled this work. So we took it out of the professionals' hands and had admin do all that fetching instead. But we also centralized all of that work of gathering that, and we centralized that into what we called our pre-accounting system. A humongous spreadsheet of every single detail that needed to be collected in order to complete the month-end close. We'd have an admin team oversee that, even if it meant reaching out to the client, chasing somebody down for something so that the professional staff didn't touch that project until all the info was in. And it was happening sooner than the 14th of the month. It was happening sooner than one day before the month-end close was done. We don't think about enough just how damaging it is to the client relationship to have to be constantly following up with the client for documentation. If 80% of the emails that you send to your client are administrative, they will think that 80% of the work that you do is administrative. They get another notification from you and they're like, oh, here we go. Yo, ball and chain accountant. Got to go rifle through the file cabinet for them. That's not what you want them thinking. But if some of that work still does exist, have it be overseen by someone who is not the advisor, not the professional staff, so that the perception of them doesn't be eroded down to like, oh yeah, they just gather up all the documents for me because I don't want to have to file my tax return or do my books. So I have them gather the documents and us accountants are like, oh, that's what you think I do? But if that's the only time they hear from us and that's what we're hounding them about, whether the client is conscious of it or not, the perception becomes like, oh, they do, they just, they do a lot of document gathering. They're really helping me out with gathering all those documents. So my opinion here is even if you don't have a way to automate it, the first thing you need for it is a system, is to centralize all this stuff, to get it out of the hands of your professional staff and de-skill it down to something an admin can do. Then when we do have ways that we feel like we can responsibly automate it, you can tackle those things one by one. You can, you can point an agent literally to that spreadsheet and say, go out and grab these things for us. Access, like managed access for agents will get better. One thing OpenClaw has, which is still like the most freewheeling Agent, and I don't recommend using it in an accounting firm, but one thing that it has is built-in support for 1Password, where you can give it a service account and it can manage its own credentials there for logging into stuff and all that so that a human doesn't have to step in and, and log the agent into every little thing, right? So we will get there, but either way, first thing we need is a system for organizing all of that because even when the day comes when we maybe do trust an agent with going out and doing this stuff, if it fails 20% of the time, we need to know where it worked and where it didn't work, right? And so we still need that same system of managing what items have been gathered and what items haven't been gathered. Spring is here. It is spring cleaning season. The perfect time to clean out the dusty corners of your accounting firm and just get rid of the stuff that you don't want to be doing still. And for most of us, one of those cobweb-laden services that is just stuffed into the corner of the garage It is bill pay. Well, good news, my friends. Artificial intelligence is here to help. You might have heard of it. The platform that's crushing it right now, this video's sponsor, Maker's Hub. Listen, nobody likes bill pay, okay? It sucks. But almost every business has to do it, and oftentimes it's high stakes. It's messy. It impacts inventory you're going after, like savings. The details matter. You miss a miss a zero or a comma here or there. Oh my, oh my, oh my gosh, that's a big deal on bill pay, right? We are getting wildly good new AI models seemingly by the week right now. And honestly, Maker's Hub are the folks that I trust to like put this work— Maker's Hub have been like arguably making the best use of AI models over the last 3 years to solve a very, very specific problem. The problem of bill pay. Nobody likes it. So if you do this stuff for your clients or you have clients who deal with project tracking, receiving, like construction people and manufacturing folk, HVACsmen, super recommend you check out MakersHub, a great example of AI done well. To learn more, check out the link down in the show notes. Hey, real quick, the most important question accounting firms are asking right now in 2026 is, how do I use AI securely in my production workflows? I've said this many times, don't take security advice from a YouTuber. What you need is actual, like, proper IT help, like this episode sponsor, Farodil. If I had to shoot from the hip with what I've seen from folks, Farodil might be the fastest growing IT solution for small accounting firms. I say small, I mean that like 100 employees and under. You need an IT group that just works with accounting professionals. Otherwise, you are paying this IT group to learn accounting firm things. Like, I had a local IT guy that I once brought over and spent 2 days watching him learn CCH engagement, like how the, like, database worked on my server. Oh my gosh, please don't do that. Work with an IT group that only supports accounting firms, particularly right now, as it is so important that we are beginning to implement AI effectively inside of our firms. You want to do that responsibly too, right? You need better help with this stuff? Look no further than Verito. They do hosting. They'll be your entire IT IT group, they'll do your whisp. Can't recommend them enough. Check out the link down in the show notes to learn more. One app to call out here, um, is called LedgerSync, L-E-D-G-E-R-S-Y-N-C. They don't have support for every bank, but they have built support for a whole bunch of banks. The idea being that like an admin can go in and connect LedgerSync to a bunch of banks, but then give your team access to LedgerSync without giving them access to the bank. So I'm not sure that it's a problem that needs solving 100% of the time, but the whole idea is if you work with 100 clients, then you connect the bank accounts of your 100 clients to LedgerSync, and then you can divvy out access to your team to LedgerSync instead of to the client's bank. Now, other ways of managing this is using an SSO platform like Okta, like PracticeProtect, where your team has credential access, but they can't see the credentials. That is, they can use a credential to log in to, say, US Bank, but at no point are they able to see the password, for example, because if a staff person leaves and they can see all your clients' passwords, that's a real issue. In fact, like, there shouldn't be a situation where your team can see client passwords. Basic password managers, they don't give you a way to avoid that. That's why SSO is better than traditional password management. The, the leader in our space, it's Practice Protect, was acquired by Rightworks a year or two ago. It's what I used in my firm, and so we could give out access to the credentials, but the people on my team who could use those credentials could not see the credentials themselves. So fetching stuff for month-end closes, still a big pain in my heiny, but build a system for it. And I do think AI agents will be helping with this stuff soon. But as I say that, I think I hear— hang on, mailman's here. Mailbag, mailbag, this segment is called Mailbag. The mailbag is bursting, overflowing with thoughts, questions, opinions. These are submitted from, uh, social media DMs, from comments on the podcast, from texts, from voicemails, 1-516-980-4968. First up, I know I can charge $4,000, but it feels like a $3,000 return, and I can't bring myself to say it's worth the higher price, even though that's better for my firm. Any advice on that mindset? So you yourself internally, you're saying this is worth that, but the client will happily pay That. Here's the secret. Nobody cares what you think it's worth. They really don't. The people who are on the other side of that, you're going to charge them $3,000. You think it's worth $3,000, but they're only willing to pay $2,000. Are they wrong? What do we, what do we do with those people? It doesn't matter. That's what they're willing to pay. And the sooner we can come to grips with the fact that pricing is relative, the only price that matters is kind of in the, in the eye of the beholder, in the eye of the client. Then we can start like strategically thinking about like, ooh, Who are the people that value me more? Because people value you very differently. And maybe you got a client today that really, really values you, and you may not even understand why yet. Dig into that. That is a golden nugget. It is an opportunity, my friend, because if one person is feeling that, there are more out there. And if you think about your entire client list and your A clients up here who pay you a big old sum of money with a giant smile on their face, imagine if you had that client all the way down. Imagine if you had a whole client list of those clients. Your life would be completely different. That all comes from understanding what is the job that I'm doing for them and why is it so valuable? That's the game, man. That is positioning. That is finding a rare pain that is an expensive pain, a pain that if a, if a business gets it wrong, it costs them a whole bunch of money. But it's so specific that there aren't many providers out there solving it. That is an opportunity for you to step in and be like, oh yeah, we got that. And then frame your fee through the lens of the opportunity for the person you're providing that service for. If you're selling a service, then they will compare the price of your service to other service providers. If you are selling a solution that gives them $2 for the cost of $1, they're like, amazing, I love that. Can I give you more dollars so that you can then in return give me more dollars? When we frame what we do around an, like, an expensive problem, then the price has no ceiling. The price is only relative to the upside of what you can unlock for them. And, and to be clear, That's hard. Like, there aren't an abundance of those things out there. But doing bookkeeping is not framing what you do in terms of how it can make them money. And for us, like, we can make 6 connections back to how reconciling a bank account can make them money, but the client can't. Tax planning can save lots of people money, but there is not a— but there are very few tax pros who would say they don't do tax planning. And so then you're in this race to like be like, well, how do I How do I convince them that I'm better than anybody else at this stuff? Don't be the best, be the only. It is way easier to be the only than it is to be the best. So be the only person doing tax planning for taxidermists, for 3 clinic vet clinics. What? 3 location vet clinics. Talked with someone at an event in Minnesota the other day where all they do is work with Minnesota gambling companies. And there's all this, as you can imagine, all this reporting around it. They have to basically issue a tax return every single month. And they're like the guru for gaming in Minnesota. And I, she's explaining this to me and I'm like, oh my gosh, this is such a cool business because there's so much specific knowledge around how this stuff has to work. And she is the hub. She is the one doing it for people. How cool is that? That is what we're looking for. A rare expensive pain where if the company gets this wrong, it's high stakes, but we know how to do it. Once you get to that point, They value what you do tremendously, and I'll let you in on a secret. The number they're willing to pay will be uncomfortable to you. That doesn't make it wrong. That is human nature. Your internal identity will always lag behind kind of your external reality. It's why oftentimes other people will see something in you, something you're capable of before you can even believe it yourself. You better believe this is the case with pricing too. You won't believe you are worth a number until a certain number of folks are willing to pay you that number. And if you don't believe me, look in the rearview mirror 18 months because that person would see the prices you're charging right now and be like, holy, people are saying yes to that. Next up, how do the videos and podcasts go out while you're traveling so much? Ooh, I can let you in on a secret of, uh, how the sausage is made here at Jason NonFirm's HQ is there's a lot of buttons that I don't push. We've actually got a whole team of talented button pushers that are editing videos, that are publishing podcasts, that are telling me, hey, so you're gonna be gone for looks like 5 weeks out of the the next 6, uh, here are the days that I'm pretty sure you're gonna be at home. And if I work backwards, that means you have to shoot X number of podcasts. And the, the videos and podcasts are going out because I have a tremendous amount of help from a team, a team that we've built in-house, no agencies, just a whole crew of full-time employees whose job is to basically make me look good. So these podcasts, they're, they are right now that I'm having to shoot them in batch, but also like sitting here and talking, like a record is about 90 minutes, sitting here and talking for 90 minutes and then trimming that down to a 60-minute episode. Like this boy's voice can only handle so many of those, especially when you're coming off the back of this live event tour where the reality of these live workshops that we're doing, because they're in back-to-back cities day to day, you're up at 6:00 AM. Uh, the latest we're on the road is like 7:00 AM. You're flying to the next city, getting in at about 11:30. People start arriving at 12:30 and you're usually, usually leaving the venue by about 9 PM. And then the next day you do the whole thing again in a different city. And it honestly, like, it's by design that it's that way. It's the only way we could get through 20 cities without me being away from the family way too much. Like now, even in the heat of the tour, I'm home probably half time. But yeah, the video and the podcast shooting takes a bunch of pre-planning, takes thinking strategically about what is timely content versus what is evergreen content. Evergreen content you kind of shoot whenever, right? But then you still have to leave space for this cool Claude announcement just came out and we have to do a podcast on it. Oh my gosh, making videos about Claude. The Claude desktop app changes like every single day. And so, I mean, there's never a world where we can like shoot something a month ahead of time or 2 months ahead of time or something like that. Too much changes. But yeah, there's definitely a lot of planning that has to go into keeping up the production schedule while I'm away. Which is 2 podcasts a week and a YouTube video about every week and a half. Next up, this was a comment around specificity and kind of niching down your firm. Totally agree with everything being said here, but right now I'm still in a massive growth phase and taking on pretty much everything that comes my way, trying to scale as quickly as possible. The alternative would be slowing my growth rate, being more selective about who I take on, and accepting a longer path to my revenue goal in exchange for less heartache once I get there. How do you think about optimizing growth rate against that trade-off? Pain now versus pain later, basically. Curious whether you lean towards pushing hard through the messy phase or pacing yourself. I can tell you that my, um, most painful window in running my firm was my biggest growth period. And you know why I was doing it? Revenue goals. And what I found at the end of that rainbow was not a pot of gold, but a really unprofitable practice. Along the way, I was making trade-offs on who I would hire because I was bringing in clients. I mean, at this, in this phase, I, I built a CAS practice to like $1.5 million ARR in 18 months from zero. And we could not keep up on hiring. It meant we ended up fudging some hires of folks that maybe weren't at the standard that we would normally, uh, let 'em in at. It meant we were saying yes to clients that maybe we shouldn't have said yes to, and we kind of hit about 1.5 and we're like, I think I'm ready to hit pause here because profitability wasn't there and the fun was very much not there. And we took about a year or two to sort of iterate through the client list, get paid better for the work that we already have, but also cycle through some team members. And after a couple years' time, it was a much more lean team. It was about 3 times as profitable as it was originally. And I actually enjoyed it. Like I went to work with people I really enjoyed working with at that point, as opposed to this manic growth phase to— for me, it was at a period in my career where I was sort of up against some other partners and had to kind of be the alpha and show that I could do this thing. And it's easy for me to say like, no, that's not a thing that you should do. But there are absolutely unlearnable lessons in life where it doesn't matter how many people tell you you shouldn't do a thing, you have to do it to prove to yourself that you can do it and then be a more reasonable person on the other side of it. So there might be a degree of this in what I'm saying, like, no, don't do it because I did it and it was a mistake. There are some things you just have to do to learn that lesson. I would say the biggest thing here that I would caution against is the sentence having to accept a longer path to my revenue goal. That is worth digging into because revenue, as you know, means nothing. Revenue really doesn't mean anything. Maybe there's some other argument like you got to sell in 2 years or something like that. But by and large, revenue doesn't mean anything. In fact, when I settled into ownership in my firm, I owned it with a partner and basically our agreement was we had sort of a profit envelope to fill, not a revenue target, but a profit target. And the agreement was we want to find the most efficient way to fill that profit bucket. And it meant when we brought in a really profitable project, it gave us permission to let a couple legacy projects go. And that is ultimately what matters more than revenue, right? And especially during a period of fast growth, it's very hard to grow fast and do that work very profitably. Maybe there's an argument to be made that this needs to go in cycles. You go through a growth cycle, you hit pause for a bit, you go through another growth cycle. I haven't done that myself. Never really talked with anybody that actually swore by that approach. But in the, in this question itself, recognize that revenue goal is said as an absolute where it's like, oh, well, well, yeah, like let's bend the universe around this revenue goal when obviously that's something that is negotiable and may be totally wrong, right? And that is the way all of our thinking is, is like there's all these assumptions built into it. And we think we have to go down this path until somebody's like, well, why are you doing that? And you're like, ah, you know, to be honest, I have absolutely no idea. Sometimes, sometimes there's good reasons for that and, and they're justified. Other times we don't even think, like we just, we end up in this tunnel vision and we're totally blind to those assumptions that are baked into our approach, right? If it's me, I still prefer the approach to organic growth that is profit-centric. And I would actually argue, even if you're trying to grow fast, this more controlled approach actually probably yields more profit in the same time because you're focusing on your efforts on getting in front of the right people, the people who will pay you top dollar. And so this more controlled approach, this more focused approach might be just as profitable, if not more profitable in total, than the take, take everything in generalist approach, right? I can tell you that was the case with me. Like profitability was not good when we were growing fast, but once we focused and kind of trimmed the fat out, it was very good. Thanks for the question. I actually just talked with this person in person at my Chicago event. They had 4 members of their team out. It's fun. It's fun meeting people in real life. Hey, I know you love Cloud Cowork as much as I do, but have you heard the news about Canopy Coworker? Subtle. It's a new AI assistant they launched inside of Canopy, one of this episode's sponsors. Canopy, a practice management system that is a feature nerd's dream. They're trying to cover as much of everything you need need in an accounting firm in a single place. But as we talk about on this podcast often, it is much easier to roll out an AI agent inside of a closed system than it is to roll out an AI agent that works across, like, all of my systems, right? That's the trade-off. I want the AI that works across all my systems, but we're struggling now with, like, what's the most secure way to do that? The AI agent that works inside my all-in-one system, buddy, that is easy. Flip the switch, count me Strap in, buckle up. I want it. Canopy is so convinced that you're going to love what they got to show you, they're going to cut you a check for 75 big ones. They'll give you a $75 gift card. Gift card to Bed Bath Beyond. What's that? Okay, I'm not sure where the gift card is for, but if you book a demo with Canopy via getcanopy.com/jason, I'll put that link down in the show notes. They give you a $75 gift card just for coming and demoing. Do it. What do you got to lose? Hey, just real quick, if you're using Outlook or Gmail to manage email inside of your accounting firm, Brother, those were tools that were built, uh, 20+ years ago and not for accounting firms like yours. If you haven't heard of them, let me introduce you to one of today's sponsors, Missive. Missive, M-I-S-S-I-V-E. It is exactly what I needed. I actually used a bunch of like off-the-shelf help desk sort of email tools in my accounting firm. Man, man, do I wish I had Missive. It's a vastly more efficient way to manage email across across your entire team. Because email is not a hill that anybody should be dying on alone. We did a demo on the main channel of it. If you want to see it in action, just search my name and Missive. Really cool, doing some cool AI stuff as well. Worth checking out. To learn more about just what you're missing by using Outlook or Gmail, check out the Missive info down in the show notes. Next up, could Digits, the kind of AI accounting ledger, be used for personal finance? Say I just want to drag and drop statements in there and run what-if scenarios and look at different budgeting Some accounting firms still do personal budgeting for their clients. Could you use Digits for this? Here's my take on tech selection of any kind is you always start with what is the most simple way to get that job done. And the, the starting point for me here is probably not Digits. So to give you an example, in the accounts payable domain, if a company needs a way to like pay bills more efficiently, I'm gonna start with, well, what does the banking do? Does the banking give us a way to do this bill pay that has enough features for us? Traditional banking's really bad in this regard, but there are some new banks that have very good bill pay. If not, what am I gonna look at next? In the, in the US, I'm probably gonna look at Ramp, which has a totally free bill pay platform that's really good. But if that's not enough, say you have very complicated needs, very detailed invoices, or a very complex integration with the accounting system, then usually right now the next best answer is Maker's Hub. They're kind of the most advanced AI platform, but I'm never gonna start with Maker's Hub, right? I kind of go through that progression only if I need it. And so for personal budgeting here, there's all sorts of budgeting apps out there like Monarch and stuff like that that people can go and use, but they're not really suited for an accounting firm that might do personal budgeting for 100 clients, especially if you're getting up into sort of high net wealth family office stuff. This is totally a valuable service, but still, again, I wouldn't start with Digits, to be totally honest. Today I would probably start with Cloud. And we got a video coming on this that I'm so excited for you to see. But Claude itself, just straight out of the box, is an unbelievably good accounting system. Like with no spreadsheets, no backing or anything like that. I can give it 6,700 transactions across 12 months and it will do the bookkeeping pretty much flawlessly. Like the only arguments you can make is maybe you would classify something this way rather than that way. But it is stuff like, well, there's an Amazon purchase and they put it in supplies. Do you want it in office expense or in COGS? It's all that normal stuff where if it doesn't have the context, it'll make its best guess. And right now its best guess is pretty much what you'd expect from a staff person. That is 6,700 transactions off of 12 months of bank statements. It can do that in about 10 minutes, just putting that in Claude, not asking it to prepare it in a spreadsheet, no other backing, nothing. The problem with that is while it will give you a really impressive response that is as correct as it can be, there's still no like backend to that. There's no database. Where that stuff lives long-term. And so if somebody else wants to continue working on that file, like, doesn't exist because it's a conversation in Claude. But there are open-source libraries that solve for this. The one I featured in a video that we shot recently is called Bean Count. That is Bean Count. And it is a double entry, like, full proper accounting system where it stores all the data in a text file. And like, I know that sounds silly, but this thing works like lightning fast up to about 50,000 transactions. And you can do like class tracking, you can do all sorts of other stuff on top of that. And so you can take the power, and this is what the video's about, you can take the power of Claude where you can throw in check images, bank statements, uh, bank transaction exports, throw in literally whatever you want and it will make sense of it. But that can be backed by, in this case we used Bean Count. And so the output is this, uh, structured file where someone else could then take that text file and, and continue working on the accounting file in a new Claude conversation or something like that. You put that, that file in maybe a shared network location. It's just like, say, a QuickBooks desktop file where anybody can open it up and work on it. But you have the unbelievable power of Claude Cowork as sort of the translator of anything you want to put into it that is then converted by Claude into like accounting entries. And it's crazy cool. So I don't think right now that we're talking enough about just how far large language models can get us in doing actual accounting work. I've been saying for years, there's all this debate over who's gonna be the next QuickBooks, who's gonna be the next Xero. Don't sleep on large language models right now. If anything, the trend is in the elimination of apps. The fact you don't even need an app for it anymore because the AI agent can do it. And on simple accounting use cases, I think that is 100% the case. So for this personal finance application, In order to need an app to do it, you need to come up with a reason why Claude can't do it first. And if you dig in there, you will be shocked at just how capable it is. Now, Digit specifically, I don't think they do what-if scenarios or budgeting. I could, I could be totally wrong there, but for added functionality like that, have a conversation with Claude. Like, how can we layer this stuff in? Let's make some custom reporting. You'll be surprised at how far that can take you. 3 more to go. Next up, Google's research team recently found that the best models give false information on 1/3 of responses. OpenAI found that their latest model makes up info on 60% of responses. Any tips for dealing with the possibility of false info? Just that whole framing is very misleading. That does not mean 60% of responses have hallucinations. What it means is one raw model response is very likely to make something up. But the way models work right now is they have built-in reasoning and fact-checking to ensure that what gets back to you, it's much less likely to have something made up. Now, to give you a very basic example, in Cloud Cowork right now, if you go to Cloud Cowork and you say, what are the IRA deduction limits in 2025? That is something that has a correct answer, and it is, it is complex, right? There are different levels to it, but there's absolutely a correct answer. Right now, if you go, go to Cloud Cowork and you simply give it that question, it will instantly give you a response without going out and doing any research. But if you ask the question in the, in the framing of go look up what the IR contribution limits are, it will then go out and pull a bunch of sources and give you a much better answer. So hallucination, will it give you a correct response or not, still is very much down to how it is prompted. And every day you see examples of like AI giving a dumb response, right, on social media. And it's like, well, usually they're using the free version. Usually they're not prompting it right. It is like taking a video of a hammer breaking a window and being like, hammers are dumb, man. These hammers are the worst. It's like, well, the hammer is in your hand. You could do— like our jobs is to find the way to use it to where it can help. It doesn't matter how many ways it can't help, but they do a lot of things to prevent hallucinations from like going out to the web to confirm before it gets a response. And how we have AI do work for us, we have to request that work back in a way that is sort of reviewable. Reviewable. So like if you, if you give a research project to an intern and they come back and they just give you an answer, that's not enough, right? Like they need to build like the case for here's what I think the answer is and references for why that is. We need AI to do the same, like not just a black box answer, but the proof so that it is reviewable. And in that way it is, it's, I would say it's actually very similar to working with humans. Uh, you're talking about AI hallucinations. If you've ever worked with an intern inside of an accounting firm. You want to talk about hallucinations, the things that they will send across your desk. It is about getting the best out of both the human and out of the AI. Next up, Ping. This is wild. What sort of software even is this? It's like a CRM/PM/AI. I don't even know where does this fit on the app pyramid. We recently did a, uh, a sponsored demo day on Ping, which is a sort of a meeting assistant, but not really because it like centralizes your emails and basically all communication with the client into this sort of intelligence. Intelligence layer where it's, where it can see into all the interactions you have with the client and do things like give you a single place to ask, like, hey, has anybody talked about XYZ with the client ever? Or have it automatically detect like revenue opportunities where they're expressing a need for something. You can even integrate it with your proposal platform so that the AI understands the context of what is scoped in the work versus what is being discussed. It's a super unique product. I, I don't know where it fits in, in the app pyramid, but we have talked on the podcast more abstractly about how AI agents, um, are almost this entirely new layer of tech. Like if you think about all the apps that you use inside of your accounting firm, each of these apps individually could build their own AI agents that work inside of those apps. And, and that's fine and maybe that's useful, but the app that's incredibly useful for us is the agent that works across all of our apps because so many, so much of the accounting work is just like integration problems, getting information from system A to system B. And so the agent that works across all of your systems, that's very exciting. But who's going to build it? Intuit's going to build agents inside of QuickBooks, maybe an agent that works across, you know, QuickBooks and ProConnect and Mailchimp. But Intuit is not going to build an agent that works across the ledger and the practice management system and your Outlook and all of these different things, right? right? And so this, this new layer of software almost has to come from a new type of vendor. I would say Ping is an example of this, where Ping is very integrations first. They want to integrate into as much of your tech stack as possible to centralize all that client context, but it exists on almost this new layer of software that is working across all of our apps. Another example of this, actually, so Filed. Filed does automation around like pushing data into your tax software, a very, very hard problem that nobody's ever really nailed. But Filed has evolved from being like this app that you log in and use to almost no longer having an AI, to almost no longer having a UI, a user interface. So if you use, uh, Carbon, if you use Canopy, use the platforms that Filed integrates with, Filed now is almost more of this background sync where you point it to your tax work papers and then it syncs what it sees in the tax work papers to the tax software. Because what they got to is like, we don't all want another app to use. We, we really don't. We just want the data to come out of our workpapers and go into the tax software. And so like, where does that live? It is an agent that's working across our tools and that's super cool. That's a, a really exciting aspect of AI agents specifically for accounting work because we do so much manual human integrating of all these different systems. Last question, is it better to try to learn all this AI stuff or master the AI in Excel, PowerPoint, QBO, bill.com, et cetera? Et cetera. So every single tool has their own AI. Do you learn that or do you learn the AI that works across all this stuff? I say you learn the AI that works across all those apps, especially in accounting, because right now when the perception of accounting is like, oh, you do repetitive stuff, which, how has this not been automated yet? That is, that's not our reality. Like we've already automated the repetitive stuff. What accountants do is track down all the details, chase down all the people, get the system here to talk with the system there. And understand all those systems to get to a month-end close, to get to a tax return. And the most exciting AI for us is the AI that works across all those apps. It's not to use Excel better. It's not to use QBO better. Within each of those tools are ways to automate the work that we do. What's left is how to get all the work across all those apps to the right place. And that is the opportunity that, and that is the opportunity of AI agents that will now work on your computer or will connect to all sorts of different apps and kind of become these centralized workers that is doing stuff in those different apps and getting information to where it needs to be. So if you're gonna learn one AI agent right now, Claude, man, Claude Cowork. If you haven't seen how it works yet, if a lot of this is going over your head, we got some Claude Cowork videos on my YouTube channel, not the podcast video, but the main YouTube channel that will blow your mind just how good and how ready this thing is for doing like actual production work in an accounting firm. Very exciting, but also overwhelming, right? Coming so fast and we're talking about it every day here on the podcast. All right. Right? People tuning into this podcast, you are the most plugged into AI of anybody in the accounting profession. You really are. You're not gonna be surprised by anything. And it's why, it's why I'm gonna see you back here in the next one, right? We're gonna do this again. You bet we are. Okay, I'll see you there. Toodaloo.