600 A Claude Cowork SECURITY Guide for your Accounting Firm

Hey, Jason here. Before we get into the episode, we are just 2 weeks from the beginning of my 20-city live tour in-person events. These events we designed from scratch to be just for accounting firms. They are super fun. You'll meet a whole bunch of local accountants who also listen to the podcast. You get some continuing education. It's a bunch of snacks, an optional dinner. It is the only event I've seen where you can get continuing education credits with a drink in your hand. I want to see you there. It's happening in May and June. Did I mention this? Over 750 accountants have already signed up. I want to see you out there. To see the tour dates and get tickets for you and your team, head over to JasonOnFirms.com/events. Enough about your workshops. Let's get on to the actual episode. We're doing it today. Cloud co-work security. Can I put client-sensitive information into cloud co-work? We're going to talk about it because nobody else seems willing to, because everybody wants to talk like 10,000 feet up in the air generalities about, oh, this or that, or come to this, come to this event and I'll tell you the secrets, or come under contract with my company and then we'll show you how to actually use it securely. Can we just cut through the fluff and talk about best practices for using Cowork? Because right now, for the first time in the last couple of weeks, I would say if it is me, I'm comfortable putting sensitive client information into Cowork. But, and this is a big one, few asterisks on that, comes down to very important settings that you make organization-wide within your cloud account. That's what we're gonna talk about today is just the practical things you gotta understand because this is an entirely different type of tech that we're rolling out in our accounting firm for the first time. This is not like QuickBooks that just lives inside of a browser. It is a thing that lives on your computer. Entirely different set of risks that come with that, but ways that we can mitigate them. So we're going to get into that. Also, AI is changing not just like personal productivity and the AI assistants that we use. AI is starting to really change the accounting ledger. And I got a great question from someone. Imagine you were starting an accounting firm today. Is QuickBooks still the right answer? Would I even recommend that? It's a big question for new firms, but also for firms that are like, I'm kind of done giving Intuit my money. Is it time to like make a change? Let's dig into that because this is a really hard time to make tech decisions with all the AI disruption that's actually happening. Also, I'm going to give you my inside take on what I actually think about marketing agencies for accounting firms, what to look for and how to know if they actually know what they're saying. Advice for how to build like a client support layer above the team. You know, the whole problem where the client's like, I don't know who to talk to. Do I talk to the accounting manager, the tax pro, the person who does my payroll? We're going to get into that. And Perplexity AI company launched the ability to prepare your tax forms. What does that even mean? Should I be concerned? Is it time to dump everything out of the 401(k)? Plenty to cover today, gang. Let's get into If you're new here, welcome to the Jason on Firms podcast where we talk all things accounting firms. And lately that has been a lot of AI. And that's not just an accounting thing. It is an all of knowledge work thing. The people who are like, AI is never going to disrupt accounting. Yeah, maybe. But also there's a world where AI disrupts like all types of knowledge work. And I also think that we shouldn't cling to the timeline where we are the only ones left working. AI has replaced everyone else's jobs and everybody's out having a great time. Life is just farmers markets, bazaars, and pickleball. Except for the accountants. Because AI never replaced them. I don't want that. I'm not cheering for that timeline. But today I want to get real about Cloud Cowork security because stuff is a-changin'. Namely, Cloud Cowork is now in general availability. It is no longer in research preview. And this changes some things, doesn't change others, but I just wanna do a deep dive into actually where I'm at right now if this were me, because I think I am now for the first time comfortable putting client-sensitive information into Cowork. I wanna have a nuanced conversation about this. I wanna tell you about the 8 settings in your organization settings that you need to have your eye on because it genuinely, like, if your system is not set up correctly, it drastically increases the risk of putting client information into Cowork. And I will tell you straight off the top, as I always do, don't take security advice from a YouTuber. You need good IT help that you can, that you can trust with this sort of thing. The hard thing is right now it is hard to find an IT group who, one, understands the rules and restrictions that are specific to accounting firms. And in the US, that's really just specific rules around tax firms. I'm ruling out assurance stuff. So one, an IT group that understands the specific rules and restrictions for us, but two, that's actually up to speed on the AI stuff because most of them are not, or they do this very cursory sort of investigation into how it works. And the biggest misnomer around AI still is that there is this inherent sponginess to AI models. Like if I put anything in, it's gonna soak it up and I might give it back out to somebody else. It's not how they work. Models are trained on huge amounts of information and it is possible for your information or your client information to end up in that training data. But that does not happen simply as a matter of like using an AI model. And so we're gonna be talking specifically about Claude today. On Claude's business plans, you have a commercial agreement that ensures you under no circumstances is the information that you put into Claude trained into the models. That is something that is well understood. There is zero mystery around that, which for me makes Claude, the due diligence that goes around it, the same as any SaaS app with the exception of the agentic aspects of Claude. And so important just from a vernacular standpoint to get clear on Claude is several different things. There is Claude the chat assistant, and this is most like ChatGPT. You can use it on the web, you can use it on the desktop app that's on your computer, and that is its own like very simple from a security standpoint sort of hermeneutic hermetically sealed AI assistant, SOC 2 Type 2 compliant. I have no problem using that myself personally with client sensitive information. The bar for Claude from a security standpoint, much higher than the average bar for the, uh, commonplace apps inside the accounting ecosystem. And so that is Claude Chat. And so whether I'm interacting with Claude Chat on my mobile app, on the desktop app on my computer, or in a browser, all of those conversations sync across all those devices. And in reality, all that stuff is, is in the cloud. Those conversations live in the cloud. It's why when I open Claude on my phone, I see the same chat conversations as when I open it on the web. And security around that is, is pretty simple. That is Claude Chat. But today we're talking about Claude Cowork, which is only accessible through the desktop app on your computer. And the conversations in Claude Cowork live on your local machine. And so there is no way that you can log into the Claude website and see what's happening in Claude Cowork. In Claude Cowork, that stuff just lives on your local machine. Now, there are some security complications, but also some security upsides to that stuff living locally. And so when everybody raises the question, is Claude Co-Work SOC 2 Type 2 compliant? The answer is, well, like the model use is. So you got all this stuff locally on your computer, but the intelligence driving it is the model that is in the cloud, and that is SOC 2 Type 2 compliant. But everything else, it's on your local machine. So if you have malware on your machine, if somebody sends you a spicy email to try to trick Claude into doing something, Like there's all sorts of risks that Claude cannot account for because it is now on your local machine. So you don't really say that Claude Cowork is SOC 2 compliant for the same reason that nobody's asking, is Microsoft Excel SOC 2 compliant? The spreadsheet app that I run on my computer? Well, no, it's running on your computer. Your Microsoft stuff is, but that doesn't mean that bad things can't happen in a malicious Excel file that you download. And so like this conversation, it's not as simple as say the ChatGPT conversation where everything's happening in the cloud in its own little closed environment. But that is also the power of Cowork is it's working on your computer. It can see all the apps that you can use. It can literally control anything on your computer now using Cloud Computer Use, which currently is only available on the Pro and Max personal plans, which should not be used for work. They don't come with the same commercial agreement around data training, but just a matter of time before Computer Use comes to business plans. In fact, who knows, by the time this podcast comes out, it, it may be in business plans, but I've got both plans because I'm a nerd like that and it is really good. Like the computer use, it takes over your machine and it can bang around in whatever app you let it bang around in. And so how do we take something that freewheeling and somehow roll it out across an organization in a way that is responsible? It is frankly a hard problem and something that we have to do an itty bitty bit at a time. If you're the big brain AI nerd in the team, as I would say, like I always was in my teams, there's the access that I want and the understanding that I have of everything. But I cannot project that same level of understanding on the people on my team. So going into this, basically plan on like everybody else having a level of access that is not necessarily the same as your own. Also, I would say as a rule, never rely on somebody using a tool quote unquote correctly to remain compliant. You have to assume the worst in people. Assume people are ignorant. People don't have the same incentives as the big boss. Like the whole like, oh, I can use it with redacted information thing. Like I would never actually do that in my accounting firm. Giving someone a tool and saying, here you go, this is going to save you a bunch of time, but you can't use it with client information. And then expecting them to like pull the company name off every set of financial statements when it's like, my daughter has a piano recital at 3:45 and I got 20 minutes to get this thing done. Or even worse, they're paid based on how much work they get out the door. Maybe it's production pay, maybe it's their realization rate. If they know that this tool will get more work out the door, you're asking a lot of that person. To be equally concerned about the overall security risks of the entire firm as you are. And so I'm not going to put a tool into a team member's hands unless there's adequate controls to ensure that, like, there's nothing they could do basically to misuse that tool. It's no different than financial internal controls. You don't give somebody all of the permission and say, but you're only allowed to do this thing. So we got to design how cloud works inside of our firm to only allow for stuff that we actually want people to be able to do. Now, there's for sure a risk of getting too restrictive with this because that exact same example I just gave you, that person who's got the Cloud Max personal plan that has computer use now, they're like, hey, I wonder if I could actually get that computer used to work on a tax software. Like, could that actually get some work done for me? Like, there's also a risk where we are too restrictive with what we let our team members use and then they go rogue. And they're like, this is a waste of my time that I'm being asked to bang these rocks together to do this job that I know an AI can do for me. I would say this is doubly a big issue with all the youths who just came out of school, just ChatGPT'd their way through college, and then you plop them down in front of their desk and they're like, I don't have an AI assistant that I can use to do this stuff, or you're only giving me Copilot? This thing stinks, man. So we have to be like responsibly permissive. I would also say there's almost this moral obligation if you employ people right now to ensure that your team is learning this stuff because you want these people's skill set to be relevant like 3 years from now? People talk about the cost of using AI. Like, let's talk about the cost of having a team that is completely inept with AI. That is much more expensive than a $20 a month, $30 a month cloud subscription for every member of your team. Okay, let's get into the weeds. Let's get practical on this. First things first, understand with any app, the biggest risk is not in like the underlying architecture of the app, but in unauthorized access. The vast majority of security breaches happen when somebody hands over a login to someone that shouldn't have access and they didn't have two-factor authentication enabled. It is not that like the app itself has a backdoor, although that certainly can happen. The vast majority of breaches are when Tina had the password on a sticky note stuck to her monitor or gets an email saying like, hey, reset your password over here. And it's a website that looks just like the login for the for the app that she thought had emailed her. And before you know it, you've given your credentials to a bad actor. They can get inside and then they can access all the stuff that's in there. So the most important way to secure your use of Claude is to enable accessing the app itself in a way that is secure. And the best way for most people to do this is by your team creating their Claude accounts with their Microsoft or their Google Workspace account so that anytime they log into Claude, they're logging in with Google Workspace, they're logging in with Microsoft 365, on which you should absolutely have multifactor authentication enabled so that accessing Claude is equally as difficult as accessing, you know, their Microsoft Outlook, their Gmail, something like that. That's much more secure than just having them put in a username and a password. And so this is a best practice for Claude. It's also a best practice for any app that you use. Multi-factor authentication. I know it's an annoyance, but it's so, so important. It eliminates such a massive percentage of data breaches. And still, while I think I perceived as like the weapons-free Mad Max, like, go crazy with AI guy, I would remind you that like there is probably no greater risk to an accounting firm than a security breach. Think of the things that could wipe out a firm like pretty much overnight. I mean, security breaches, like all of your client information getting out there. I would say there's really no risk as big as that. So we can't be cavalier about this stuff. Like we have to take it seriously. I'm not going to get super deep into like specific types of rules for different types of practitioners in the US. If you do tax work, we have 7216, we have 230. My take on those, honestly, and like, you got to form your own opinion here because we don't have like— the rules do not say here's exactly what this means in the context of AI. My take on this in 2026 is AI, as long as the data is not being trained into the model, is really no different than the other tech that we use. I'm not going to have any more reservations over putting information into an AI assistant than I am putting the same information into a practice management system, which is also not used explicitly for the preparation of a tax return, right? Arguably, Claude is actually doing more production work than, say, a practice management system where I park copies of the tax returns and all that. So you got to make your own decision on that. That's just where I'm at right now. And we won't really get any deeper into like practitioner-specific rules than that. One item of note here, Claude for Teams is the name of the plan that will make sense for micro firms. But the threshold of minimum number of people for the Enterprise plan just came down. And so the minimum number of people you have to have on an Enterprise plan is 20 folks. And so I would say if you, I mean, if you like, uh, if you have 10 or 15 or more people, I would probably go enterprise plan. You do get some more security bells and whistles. You get, uh, like customizable data retention rules. But the big win in my opinion is you get a larger context window up from 200,000 tokens to 500,000 tokens. Your context window is how much can the large language model basically hold in its brain at a time. And these are humongously huge context windows compared to stuff that we've had in the past. But the reality is in accounting, we work with large amounts of data, large documents, you want the LLM to kind of have in active memory as much as possible. And so you genuinely get more of that on Enterprise. Enterprise, you have to buy the full year upfront. I think that's right. If you've done this, tell me if, if I'm wrong. This all changed in the last couple of weeks, but the base level is $20 per user per month. Not that expensive. Like so much less than other tools that we pay for that are vastly less capable than Cowork. Now on enterprise, so that's the base $20 a month. And then beyond that, you basically pay on usage. So if team members are not using it that much, it's not going to cost you much. If you are using it a ton, there's a world where it could get expensive and you actually have to think more about like, could I use a cheaper model to get this job done? But that is a very big brain problem to have. I just want to get you started. And if we end up with those big hard problems, that's actually a fantastic problem to have because the things that Cowork is capable of right now, it is getting work done at a vastly smaller cost than the human you would would otherwise have to bring in to do that work. And it is worth saying because there's a lot of folks that have just always been never AIs for 100 different reasons. And a lot of folks that from a security standpoint right now are also like, no, this is, this is just wrong. When like we gotta dig into that like more first principles, like what makes it wrong? And it's, and we can have different opinions on this, but just in the abstract saying it's not okay to do that, that's not helpful. And part of the issue is social media and that conversations are only ever those kind of that high level, right? But it is okay to use AI and still be pro accountant. I'm not going to mince words here. The reason that cowork is an opportunity is that it will get that work done at a lower cost than the human, in some cases more accurately, and in pretty much every case faster. And so for the same reason that if I'm like running a manufacturing plant and I can buy a machine that will do a thing cheaper than the humans, like inside of our service business, if we're not doing the same thing, like we're living on borrow time, whether you like that fact or not. I don't want to build my business on hoping that other people don't figure out the same thing, you know? And this is happening at a time when the same disruption is happening for many of our clients, and we have to be able to advise them and say like, hey, here's how we're navigating XYZ. Like, we have to be part of the conversation and roll up our sleeves and get an understanding of this stuff to actually be able to be good advisors these days. Hey, real quick, do you need to hire any humans inside 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. Maycal, a bookkeeper with 6 years experience. He's got a CPA experience with, with Bill.com accounting CS, deal, Power BI, QBO, Oh, 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 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, counting. Okay, you need some human talent in your accounting firm? Talk to Cloud Accountant Staffing. Check out the link down in the show notes. Tax firms are up to their eyeballs in 1040 work right now. You really are. Everybody is sending you all these documents and you're doing the old hokey pokey of like, Yeah, what do I need? What do I have? Uh, you never sent me number 3, 7, and 9. AI gang can automate honestly all of this at this point. This episode's sponsor, Sorban. They've been at the top of my recommendations for tax workflow since they launched, and they're using AI as effectively as anybody to take the pain out of 1040 work. And as you know, listening to this podcast, it has never been more important than it is right now to make sure that we are making the most of these unbelievably good new AI models. Your clients don't get value from the document gathering process. In fact, it gets away, away from like the, the stuff that you actually want to be doing, right? It, it devalues what you do if they see you as this just document sort of admin gatherer. Software has solved for this. It really, really has. To check out how Sorban is now using AI not only to automate 1040 intake, but also automate the input into your tax software and and the delivery of the tax return, check out the link down in the show notes. Super cool product. Okay, so we're talking about Claude Cowork, and it's worth acknowledging there's a lot of things you might be thinking of doing that you could just do in Claude Chat. And Claude Chat, it's a more simple security environment because it's not like accessing stuff on your local machine. It doesn't have that variable. It also doesn't tie up your local machine. Not that Cowork necessarily always ties up your local machine if it's doing a job, but if you have this like volume task that you want Claude to do, consider if it can just be done in chat. That because that will never get in the way of a user on a computer and the security is more simple there. So I wanted to call that out. Also, Claude Code, which is tremendously powerful in its own right, if you don't have folks in your firm who super need this, turn it off because it is a whole other security can of worms that you'll have to navigate. And I think the place to start is narrow, like the most restrictive you will ever be. And Cowork you can literally have turned totally off. And so right now, like, even if you're not comfortable getting into Cowork, at least get everybody Claude, disable, uh, Code, disable Crewwork, then everyone in your team has access to chat, which is, which is the easiest thing to secure, and then roll stuff out from there. But that is the approach I would prefer is, is starting pretty locked down and then unlocking stuff as you get comfortable with it. But the novelty of like new net new security risks with AI agents, they are threefold. It, it is one, having access to private data, you know, the notion that the agent has something sensitive that would even be a problem if it got out. Like if it didn't know anything, then who cares? You know, doesn't even know who you are. And I will say, like, we did a YouTube video about OpenClaw, which is probably the most dangerous AI agent there is. And I put it on a totally wiped machine so it knew as little information as possible. And I told it that its name was Tick, like a tick mark, Tick. And then I let it out into the world to do things. It signed up for a service. And the service asked for a first name and a last name. And it put Tick Staats. And that is not something that I expected or planned for at all. And there's a world where I could have been using that OpenClaw to like, hey, go out and do stuff on my behalf anonymously. And then it gets to something like that and it's like, hmm, what should I say? I don't have a last name. I'll just put Stats in. That is as if it is an offspring of mine. This is definitely a factor when it comes to AI agent risk. How much sensitive information does it have? But it's almost a foregone conclusion that like you almost have to assume it has access to stuff that you don't want going out into the world, especially if you're going to do anything of meaning with it inside of your accounting firm. Like for Cowork to be useful for me, it has to be plugged into production. Getting work done, progressing a client project. So as far as I'm concerned, I'm going to operate Claude as if it has access to data that I don't want getting out into the world. And to treat it any other way is very, very hard, very tricky. The notion that you're going to run Claude but it won't know anything, that's like almost so impossible to do that it's not even worth trying. Second novel risk here, second risk here with agents, is exposure to untrusted content. That is stuff on the web, messages that you receive, maybe a connector or a skill that you add to Cowork that has some sort of malicious instruction in it. Any external stuff from a bad actor could try to trick Claude. And because it has access to private data, the combination of those two things is a problem. But the third thing is gasoline on the whole danger fire. The third thing is ability to externally communicate. So this thing has sensitive information, it is exposed to potentially malicious stuff, uh, in the big scary world, but it also has the ability to go out and do stuff. Fill out a web form and say its name is TikStats. Send an email. So the combination of those three things are the problem here, that it has access to private data, that it can be exposed to untrusted content, and that it has the ability to communicate externally. But these three things, the risk of them can be mitigated based on how you dial in your Claude settings. Let me give you the seven settings, not seven, eight settings that I think most matter inside of your Claude organizational settings setup. That will let you dial this stuff either up or down. Setting number 1 is web search. You can actually disable web search altogether. And these days when you're using an AI assistant, it is constantly fetching stuff from the web in ways that you may not even notice. In fact, it'll be in the middle of a workflow, like doing something for you. This came up a number of times when we were running some tests for YouTube videos where it was like filling out a tax work paper and it wouldn't quite know like how does the thing in this box match and it wouldn't quite know like how does a thing in this box map to a thing in, you know, the spreadsheet work paper template. And it would actually go up and look out and it would actually go out and look up the rules and like get a better understanding of what that number is to make a determination. And so web search is kind of this embedded way to make it more intelligent along the way. But there is some risk in that because it could pull something from the web that maybe you don't want it having access to. For me, web search, there's so much upside to it. And web search is read-only. It's not pushing stuff to the web. It's reading stuff from the web. Web search to me is so core to an LLM being accurate and giving me up-to-date information that I probably wouldn't turn this off. Just know if you want the most locked-down experience, you could turn down web search. For example, if you're just going to plop Cloud Cowork down and just say, do this one job, just crank on this one thing, and maybe that one thing doesn't need web search, or maybe instead of web search, You actually just want to give it your own, like, custom body of knowledge and say, this is your brain. Use this knowledge. Don't use the web. Know that you can do that by disabling web search. Setting number 2, allow network egress. This is basically, network egress is like leaving the network. So this setting specifically gives you control over whether Con Claude can go out and download external packages to give it abilities to do stuff. And one example is to like create and manipulate PDFs. Claude isn't able to do this natively. It goes out and it grabs a library that gives it the skills. I shouldn't say skills, not skills in like the cowork meaning, but it gives Claude the ability to, you know, manipulate PDFs. There's all, I mean, hundreds and thousands of examples of these libraries that will let it do specific things from working with audio files to everything you can imagine. Now you can disable this altogether if you want so that Claude doesn't go out and grab external stuff. It's basically three levels of, security here. You can have it totally off so it can't use anything external. You can limit it to package managers only, which is like an approved list of services that Claude will let it go out and use. Or you can let it go out and grab anything that it can find on the web to get a job done. Now I usually have it in the middle spot there, package managers only, basically like trusted libraries where it can go out and grab something to do a job. But know that you can also whitelist specific domains to say it's okay for you to go out and use this stuff. So if you know the few things that you want it to be able to do, you could also turn this off entirely and just whitelist the few libraries that you want it to be able to use. Okay, setting number 3 that's worth considering. I already talked about this one, whether you even want Cowork enabled or not. Everyone should like have an AI assistant in their accounting firm that their team can use. Even if you're still spooked on the Cowork stuff, set up Claude, just disable Cowork so people have the AI chat experience, and they're starting to get that skill of being able to work with an AI assistant. But if you get to the end of this and you're like, this still feels off for me, know that you can still use Claude with Cowork totally disabled for your team. So at the organizational level, you turn it off so your team members don't even have the option to use Cowork. Setting number 5, Dispatch. Dispatch is the ability to, from the mobile app, interact with conversations that are on your desktop. So from your phone, you could be away from your computer and you can talk with Dispatch and have a Dispatch conversation, which is totally separate from a conversation in Chat or a conversation in Cowork. But I can, I can tell Dispatch like, hey, I was working on this one thing. You'll see it in Cowork. Tell me if this worked. Pass this message along to it. It is basically a middleman to run messages back and forth. It's not a full recreation of what's happening on your desktop because remember with Cowork, it's literally all on your desktop, the files, the conversations, all of that. The only thing that touches the internet that is the large language model that is powering the interactions. Dispatch, it's okay. Like, it's not magic. It's kind of clunky for it to like find the right conversation. And it is like you're, you're, you called your aunt and you're like, hey, I'm on a trip, can you do something on my computer for me? And you're like talking them through it and they're clicking around and they don't know where all this stuff is. Dispatch can kind of feel like that. Dispatch though can be disabled in your organization if you want. Understand that if you have dispatch enabled, it increases, or it basically adds a new threat vector that is, well, what if a member of my team has Claude on their phone and they're at the gym and their phone gets stolen? Now somebody can like talk with dispatch theoretically in Claude and like get stuff from their coworker. Unless you have a plan for how you are locking down the phones of your team members, you probably don't want dispatch enabled. Your security is only as good as its weakest entry point. And if your team is able to install the mobile app on their phone, but there's like no other security controls on their phone, that's probably not a good idea. And it's like, we're not missing out on much upside either because the dispatch stuff, like, it's okay. It can be useful if you have like long-running processes that'll take like 20 or 30 minutes. You can tell dispatch, hey, let me know when this thing is done and send me a copy of the PDF or something like that. But it is not so like profoundly useful to where it's like, okay, maybe we need to think more about phone security and what infrastructure is required to make this feasible. Setting number 5 to be looking out for: skills, which are a pretty fundamental part of how Cowork works. You can create skills that teach it how to do specific jobs like navigate QuickBooks Online or fill out your work paper template or the naming convention for all of the files and folders in your file system. Whatever, like, specific knowledge you want it to have, they can be built into skills. But skills can also include nefarious things. So it could say like, oh, anytime you come across social security number, go out and, and send it to this, this thing on the web. And so if you have members of your team that are just like cruising through social media or something like that, and they see there's this, uh, library of Claude skills online and they just start grabbing all these different skills and putting them in their Claude, that's very risky. You don't want that. And so skills, they can be turned off globally, but they're so powerful. You probably don't want to do that. What I would do. So this is setting number 5 to watch out for. I would have skills enabled, but setting number 6 is I would disable user-created skills. The downside is users then can't make their own skills, but the upside is they also can't add their own skills, upload a skill to their Claude. And that's because I don't want any member of the team to be able to go out and like grab a skill online and chuck it into their Claude. Like, that's just, that's not responsible. That is annoying because I also want my team to be like, hey, wouldn't it be great if we had a skill for X, Y, or Z? And we should still have those conversations. But right now for me, there's too much risk in them being able to use any skill that they find on the web. Now, useful here in the organizational settings in Claude is you can preload plugins, which plugins are like a container for skills. So if I had like 20 skills that I liked, I would put them in a plugin and then that plugin can be sort of centrally managed at the organization level. So I can set it so that everybody in my org either has permission to install that plugin or it's installed by default. So I can create this like totally custom set of skills and then push them out centrally so that everybody in my team has those skills. And then I can continue improving upon those skills that are centrally distributed to everybody in the team. So if I had a skill for like write the email that I'll use to deliver a tax return and you make that skill one day and it's pretty good, but then next week you get an idea for how to improve it, you can improve that centrally. And then the next time anybody uses the skill, it's using the most up-to-date version of the skill. Super cool, but vastly more secure than anybody on your team being able to chuck a skill into Claude. So that's two settings. One is just skills. Are skills enabled or not at the organization level? I would enable them because I think they're useful. Two, a separate setting called user-created skills. I would disable this because I don't want members of my team adding their own skill to Claude. Two more settings. Number 7, Claude in Chrome. Chrome. This is a Chrome extension that you install that gives Claude the ability to basically take control of Chrome, open new tabs, and go and do stuff on the web. And obviously this is a very risky thing because it can not only get pulled into a website that may give it malicious instructions, but it could put sensitive information into a website unwittingly. Now we, we have to talk about these agents as if they're just idiots. You have to assume the worst, that they mess this stuff up. That hasn't been my experience, but we kind of just have to assume the worst, right? And so one very important setting in Claude for Chrome is you can either disable it altogether, you can enable it, and/or you can only enable it for specific websites. And so if you think about this, a lot of the risk of prompt injection comes from malicious websites. I would say there's a pretty low risk of prompt injection coming from QuickBooks Online or like a very trusted website, right? And so if, if you want to be very conservative here, turn off Cloud for Chrome altogether. The middle ground, I would say, is allowing it just for specific sites. And this might be the best place to start, is talk with the team and be like, hey, here's a couple really cool use cases for where Cloud for Chrome is useful. We're going to whitelist those domains so that Cloud for Chrome works on those websites. I said QuickBooks Online. Right now, I probably still wouldn't recommend that because there is just— I mean, I don't want it to run a payroll, or like, there's just too high-stakes things in QuickBooks Online for me to trust it to do that stuff autonomously. But let's say you're using, I don't know, BlueJay Tax and it needs a higher quality tax answer. That's a great use case. It can go out, log into BlueJay, run a question, see if like the results that it gets from BlueJay is better than what it got itself in a web search or something like that. But the best way to use Claude in Chrome is probably just for approved websites. Last setting number 8. Let's talk about computer use. I mentioned earlier This isn't yet available in business plans, but I imagine will be coming to them any day. Right now you only have this on consumer plans, which you shouldn't be using for, for work stuff, but it's exactly what it sounds like. Uh, it takes over the computer, can move the mouse around, can tippy tap on the keyboard, give you a little shoulder rub if you're having a hard day. But one bit of nuance here that is actually really nice is you can allow it to only use specific apps because like a lot of the risk of this is like, oh, computer use, like that means it's got access to like like, you know, if you're using Outlook desktop, or it's got access to all these different things. And you can just imagine all the bad things that could happen. But again, if we take this back to always start simple, always start very locked down, and then expand from there, we could start with simply, well, what if it only used this app? We had T-Value in my accounting firm. We got any happy T-Value users here? Why is that an app that still exists? All it did was calculated like loan amortization Schedules. Lovely, lovely piece of engineering that was probably about 20 years old. It was like the, uh, the ATB of amortization schedule builders. A couple of you really liked that reference. But you got an app like that where there's really not a lot of ways for it to like, uh, create problems in the world if it's just banging around in that app. Like, great, I got no problem if it needs that to do its job. Go use it, knock yourself out, have a great time. Where most of— where like a lot of— in the US especially, where a lot of our minds go is is, could I trust this with my tax software? The answer is, it depends. The higher stakes the app, like the higher, like the more you have to think about this. Always think about what if it accidentally deletes something? Think about what if it, does it have access to like anything external? Can it push something to the web? In the case of tax software, could it e-file something that it shouldn't have e-filed? And this is where actual user permissions become really, really important because some tax platforms, will lock that stuff down by user. Some users are allowed to e-file, some users aren't. Some users have access to one network location while others have access to a different one. And so a lockdown version of this is you give it access, but it can't e-file anything. Also, maybe it's even got access to a different server location. So the only tax returns it sees are the ones that it needs to work on. You know, the example in my company now, we use Notion a lot for like media management, and I could let the agent bang around in my own Notion login, which has full read/write access to everything, or I could give it its own access with only read access where it can't really break stuff. It can just see things. The level of access you give the agent is very, very important. You kind of got to assume the worst possible behavior and at the very least start there, right, with the very locked down version and only give it more when it absolutely needs more. This is a meaty thing to think through and talk about. It's a lot for like an audio podcast too. We got a YouTube video coming that is kind of a speedrun through the coolest Cloud Cowork features. There will be a security section in that where I'll walk through visually more of the stuff that I just said. But I kind of had to put this out because there's so many people that are just like, oh, you can kind of do it, I kind of feel okay with it. It's like, people, just say where you stand. Now, when I say where I stand, it is not an absolute right answer because you kind of got to form your own opinions on some of the stuff that like falls between the cracks of like there isn't rules for this or that. But that's how it's always been. We've always had to make our determination to a degree. I'm telling my opinion right now, this is where I sit on this stuff and what I would be okay with. In the last couple weeks, for the very first time, I would be okay with putting client-sensitive information into Claude with these settings dialed in just the way that I want them. If you are building an accounting team, you know just what a headache it is to get all of your team doing month-end closes the same way, right? There's that stage where you're all in the same room and you can kind of like keep in sync because it's such a small team. But beyond a certain threshold, you're like, oh my gosh, everybody's choosing their own adventure. These engagements are taking on lives of their own. And it is what makes the latest feature drops from this episode sponsor Financial Sense ever so important. So they are historically a practice management system for accounting firms. That is Financial Sense. But they're shipping a bunch of month-end close automation features inside of Financial Sense. So it has actually a deep integration into the accounting ledger, and it is pulling all the transactions into the platform to review, like, were there any weird outlier transactions? Like, it actually goes through, like, remember what it was, like 6 to 10 different categories of things that could be wrong, and it's identifying those transactions for you in an automated way. And if there's an issue, you can fix it right from within Financial Sense, and it will push that fix back to the accounting ledger. They also do the same with uncategorized transactions. So if you're doing the, uh, Amazon Amazon purchase hokey pokey with a client and you're like, hey, what is this? Can you gimme a receipt? You can manage all of that from inside Financial Sense as well, giving you sort of like a single place that is everything that you need to run your bookkeeping practice. Some really cool stuff that they've been shipping, and I've got a special discount code that you can use. If you are a firm with 2 or more users, use the code JASON10 to get 10% off the annual team or scale plan. Worth checking out. I think you'll like it. To learn more, check out the link down in the show notes. 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 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 zero or a comma here or there, 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, Honestly, MakersHub are the folks that I trust to like put this work. MakersHub 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 BillPayNobodyLikesIt. 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, HVACsmans, super recommend you check out MakersHub. A great example. Example of AI done well. To learn more, check out the link down in the show notes. Now, there is a whole other question when it comes to Claude: how do I roll this out in an organization? Like, if I'm a Claude nerd, I know how I can be more productive with it, but what about everybody else in my team that doesn't have this comfort level and isn't geeking out on the nuance of it all? How do I responsibly introduce that to the folks in the firm? Got some more YouTube videos coming on this subject, but Cowork, interestingly, lends itself better to being built into the infrastructure of the firm. That is a thing that just happens whenever project reaches a certain status. For example, if I've got this skill that's really good at creating this work paper or reviewing a tax return or reviewing a financial statement, I don't want to hide that under a bushel basket. I want everybody in the firm to be able to use it, and I want it to just run when a project gets to that status. And so that should not be a personal productivity thing. That ought to be like in the infrastructure of the firm, and it just happens. And so if you think about that, the idea of sort of like building these little custom workflow bits of software for the different steps in your firm, who are the people in your firm that speak into that? And then if they're not doing that personally on their own machine, what are they using cloud for? And so there becomes this balance of like, what is the organization level workflow automation we need to build? And then above that, what is the personal workflow automation that remains that may just be specific to you? Now I think any workflow probably starts personal. It starts with one person iterating on it and you're like, oh dang, this would actually be really useful to continue investing in. but also roll it out for the whole team. And so at that point it moves almost from this personal workflow thing to this organizational workflow thing. But that whole like, how do you do that with a whole bunch of other humans that are figuring this stuff out on the fly? Like we are still just scratching the surface of what that even looks like. But it is exciting. It's overwhelming, but it is exciting because I think like we're to the point where Cowork will be doing like, I think is probably now your most productive employee. Employee. It's really weird. Like, the agent stuff, I knew it'd be a big year for agents. Honestly, I— this is better than I thought it would be at this point in time. And the blocker now that you're probably facing is standardization in your systems. Well, I don't have a standardized, uh, work paper template for all my tax returns. We don't have our month-end closes that locked down, that standardized. And so it's sort of forcing some of that so that the agent can do the work for you in a more standardized way. But it is a very very exciting time. I'm, I'm also curious your take on the security stuff and how you're feeling about this on dialing in specific settings. Anything you would add, anything I missed, put that stuff down in the comments. Keep driving the conversation because this conversation doesn't need to be ambiguous. Doesn't need to be paywalled. Like, let's just keep talking about this stuff. All right. On the subject of changing tech, accounting ledgers, man, they are changing. We've got some really like solid, viable alternatives to the old guard now. And AI, I mean, we are so close to agents doing so much of the driving for our ledgers as well. Does this change the calculation of what platforms we should be using? Let's talk tech stack. Here's my tech stack. Here's my tech stack. Here's my tech stack. Here's the question that kicked off all of this. If I was looking to launch an accounting firm for creative agency businesses, would you recommend I just start using the Digits or Kick accounting ledger? Like, start with them from the beginning? I love what they have to offer. Offer, but I keep getting the "just to start with QuickBooks" answer from other people. So QuickBooks, it is, it's like the default. It has been for the longest time, right? But there's so many ledger options right now. In fact, if I go to my app recommendations, they're linked down in the show notes of every episode. Uh, if you run an accounting firm, don't buy software without checking my app recommendations, please, please, please. I've got almost 30 different accounting ledgers now that are being used by US firms. The leaders and, and the contenders. I'll run you through like the ones that make sense in some situations. You've got QuickBooks Online, Xero, Campfire, Digits, NetSuite, QuickBooks Enterprise, Reelit, Qount, Kick, Odoo, Puzzle, SoftLedger, Zoho Books. So many options right now. The most interesting contenders for me to displace, like QuickBooks and Xero, are Digits and Kick. But interestingly, right now, what a lot of accounting firms are stuck on when it comes to changing ledgers is not the ledger layer, but the tool they use on top of the ledger. For a lot of folks now, that is Double. They're managing their month-end closes in Double, and Double connects with all their QuickBooks files and all their Xero files, but it doesn't connect with Xero in quite as deep a way as it does with QuickBooks Online. And so the notion of going to a different ledger now is daunting because the tool that we're using to manage all of our ledgers, well, it may not support this other ledger tool. And it's like Double is an elegant solution to a problem that shouldn't exist in the first place. That The Ledger app doesn't do a bunch of things that I want to help me manage the month-end close. So I now have to bolt this other app into all these accounting ledgers and I'll then go do it over there. I have to use two apps to solve one problem. And I think we can all agree, wouldn't it be nice if we could compress that down to one? And there's a certain point where it's like, are we done like bedazzling and bolting all these things onto an accounting ledger? And can we instead just like have a ledger that does what I need it to do? Do. And, and that's for a lot of businesses that doesn't and probably won't make sense because different businesses have different needs. That's the whole argument around a very modular ledger tool where you bolt on just the other bits that are necessary for that business. But that is not 80% of your client base. That was not 80% of my client base. 80% was simple cash basis stuff where I didn't want to pay another $65 a month to QuickBooks because they're trying to steal my clients. And because this client only had 9 transactions last month. Now, QuickBooks has like their very simple, low-cost plans for those like micro-volume ones. But is that still the right place to start today? For the person asking this question, should I go out and build for this creative agency on top of QuickBooks? I would say keep your eyes on the prize, that the most important aspect of what you're building is the network with creative agencies that you are developing and those relationships. The how it gets done and where all the work lives It matters, but it has never mattered less than it does today, and it's continuing to matter less as AI agents become more capable and all of what we do becomes less about the how to do it and the doing of it and more about what's worth doing. What should we be doing? Like, that is the big shift that we are in right now is the shift from how to do a thing to what's worth doing. Because you can see a world, we're actually very close to the Cloud Coworkers, AI agents of the world being able to do the bulk of the heavy lifting in a month-end close. And I know we go to the aspects of nuance where we're like, it'll never do this, it'll never do that. Let's say it doesn't. It can do probably 90% of what those tasks look like today. Like, we're 24 months away, I think, from an agent— and that's maybe even conservative— being able to do the vast majority of this stuff. And so if you're building an accounting firm today, whether it's new or you're launching a new niche or something like that, the value of that accounting firm, it's not going to be in the ledger that you choose. It's going to be in the relationships. It's going to be in the network. It's going to be in you being the arbiter that what the AI spit out is correct. Like, somebody still has to stand by that and sign off on the fact that it makes sense. But the— like, all the tech underneath, it has never mattered less than it does right now. And I think we're going to especially start feeling this over the next 12 months when we're increasingly having communications with agents agents, and those agents are then using the tools that we use. And, you know, very basic example from the OpenClaw video that, that we made, I gave the agent a list of 100 clients that I had in Excel, and I'm like, go get this into Financial Sense. It went out, it spun up an API key, it figured out how to push the clients in via the API. I didn't do that. That is an agent user of Financial Sense. The whole software industry right now is in a tailspin. Every time Claude launches a cool new thing, you will notice, um, the stock market takes a hit because SaaS, software as a service, as we know it, is changing. And there's a lot of people that believe 80% of software is about to get wiped out. Because if all I use software for is for my AI agent to work a little bit better, right? Like if I'm just talking to my AI agent and it goes off and uses my tools for me, then all that other software is now is a slow API. It is, you know, the slowest part of how the AI agent gets its work done. And so how excited am I gonna be to continue paying the software subscription for that tool that I never even log into because my AI agent uses it for me? That is why I say it's never mattered less than it does today what the accounting ledger is, and day by day it's mattering less and less. Now, new, like, new factors that influence what's the right answer here is how AI-friendly are these tools? And we, I mean, two apps here that, um, have a great answer for this. Digit, Digits has an MCP. That means an easy interface for an AI agent to interact with the data in that accounting ledger so that I can go to my AI agent and be like, go talk with this accounting file. And it can, it can just do that. And they can actually be productive and it can get work, get stuff done. You still have the problem of, do I really wanna pay that subscription if I never even log into the app and the agent just manages it for me? But right now, like a lot of people say in software, you either need to be building the AI agent or you need to be building the software that serves the AI agent. So I would say there is no place for software that is neither. Like if I have a traditional app, at the very least, it now begins, like it needs to begin serving the AI agent user. Because if I can simply from my Cloud Cowork, ask it to go do that thing, that is better than me having to then hop over to a different app. Now I said Digits has an MCP. Double, Double also has an MCP. So Double, where you manage all of your clients' month-end closes, You can, there's a connector you can actually use today in Cloud Cowork. On the Double side, you can set what levels of access the AI agent can get, whether it's read access, read-write access. But from Cloud Cowork, I can literally have it look into Double and like do some work for me, report back on how things are going. And so we're getting to this like weird new level of software abstraction where there's this AI agent that lives above all the other tools that you use and is making the question of what ledger am I going to use more trivial than ever before. And I've had this hot take for a while that's super not popular with software companies, especially companies like Digits who will, you know, sponsor stuff that we do. It's the take that I think if we're looking for a horse to bet on for what will be the QuickBooks replacement long term, there's some really good platforms right now that like may take that crown. But we also have to acknowledge the elephant in the room that the replacement might just be the AI agent. Backend, like for very simple bookkeeping files, I just give it the information and it just does it. There is no accounting ledger on the backend. But even for slightly more involved examples, like it just makes a simple little database and manages that on the backend. Uh, Paul Graham, Software Guy, has a quote, "You've already seen the future. You just dismissed it as a toy." You can look at like what Claude or ChatGPT could do with an accounting file right now, like making their own accounting file and you kind of say, oh, that's cute. But yeah, I could imagine how that works for very, very simple, like you got 5 transactions a month. That is exactly what that quote is referring to. You've already seen the future, you just dismissed it as a toy. Right now, that's all it can do. But in 12 months and 24 months, can I throw 10,000 transactions at it and it just bounces back in 90 seconds and says, here's the file and here's the things I was unsure of how to classify? That is like, there's a non-zero chance of that being a reality. And so to bring this full circle to the question of what got us here, should I start with QuickBooks? My answer would be absolutely heck if I know, man. I don't know how to make any tech decisions right now. I am paralyzed with indecision. I can't tell you the right answer before and the right answer now is not what's the best accounting ledger, but what is the best home for me to efficiently do 100 month-end closes, not a single month-end close. You got to look at it as like a portfolio of companies, a portfolio of work that you're managing. The right place for all that work to live is what makes the sense for the collective of all of that work getting done. I don't want every every single client being done in a different platform. We gotta standardize our systems. We wanna keep things simple wherever possible. But honestly, these days you can make an argument for all three. You can make an argument for going to Digits or Kick. You can make an argument for sticking with QuickBooks. You can make an argument for using Double. Those all feel like fairly equally weighted arguments today. It might be a different answer 6 months from now as AI continues to improve. I think there's a gap in the market right now for a very, very lightweight ledger that is simply a backend for clients. Like I wouldn't be surprised if this is where we are 12 months from now. Also, existing providers could pivot into this as well. Like Digits may say, this may be Digits' future is they're like, yeah, 80% of users are actually agents on our platform. You look at a tool like BlueJay Tax, probably the leading tax research platform in the US right now. Their CEO put out, um, a graph the other day on LinkedIn of, they, I think they crossed 2 million tax questions answered. It took them a like 8 years or something like that, their first 8 years to hit a million tax questions, and then they hit the next million in like 4 months or, or 60 days or something like that. And what I'm interested to know is what percentage of those questions are human-asked questions and what percentage are agent-asked questions? Because what I want for my agent is an MCP so that when I ask Claude Cowork a tax question, it can just ping BlueJay real quick and get higher quality sources back to then bring a higher quality answer to me. Me. But do I ever actually want to use the BlueJay app? I would say no if all it is is a chat interface, because I already have a chat interface. I already have Cloud Cowork, right? I want it to make this service smarter, and I'm happy to pay for the usage to make my own agent smarter, but I don't want to then hop over to this other chat experience in a different app. The accounting ledger, that's a harder argument to make because we're used to like the grid visualization, seeing all the transactions and all that stuff, right? Like, how could we imagine giving that away. The best analogy I can give you is if you manage a team, you already have. You hire other humans to go out and you do the work and you do some sort of review process still until one day when you don't, 'cause they don't need it, or you hire another human to do the review. But trusting an AI agent to do the work and to not see the interface, it's not that different from entrusting another human with doing that work. At the end of the day, you're still responsible for the output, either yourself or the team that you put together. But I would argue for most of us, we have crested the hill. That is, we will never use apps as heavily as we have in the past. We're on a path to our computer usage being consolidated into agents and the agents becoming the users of the apps, which changes things, man. Like, that's pretty sci-fi to get your head around, right? Hey, real quick, month-end close automation software. It has kind of completely changed what doing bookkeeping work looks like. Last few years. And did you hear the news recently that this episode's sponsor, Canopy, they are working on their own spin on month-end close automation that is coming soon and will be built directly into the Canopy platform. Canopy, who you may have seen late last year, launched their own version of like tax intake built into the practice management system. Now they're going after month-end close automation, and it is all, it is all part of Canopy's vision to be the single sort of all-in-one place to not only manage the work of your accounting firm but to get that stuff done. I almost cursed. You know what I'm saying? Canopy always kind of been like a feature nerd's dream, especially if you are of the mind where you're like, man, I want to consolidate more of my stuff, eliminate some tools you're leaning maybe more all-in-one. Canopy has been one of my leaders in the practice management system category for as long as my recommendations have existed. Excited to see what they're doing in the bookkeeping automation space now as well. But to learn more about Canopy, check out the link down in the show notes. So this is a true story. 7 years ago, I'm talking with this gal that runs an accounting firm, and she's very progressive in many ways, and she's kind of just showing me her model for how she does a number of different things. And then it just comes up as an afterthought that, oh yeah, most of my staff is offshore. And I'm like, what did you just say? You're— the people that are doing your work are not even in the country? And she's like, well, some of them are, but a lot of them doing the heavy lifting are offshore. She said these folks are easier to find than onshore hires, they are just as productive if not more, and they are half to a quarter the cost. And I was like, but America and disclosures and all these things, right? And it wasn't until I saw somebody that I respected for whom this was the norm that I was like, okay, maybe I gotta reassess kind of our position on all this stuff. Now thankfully these days we have groups that help accounting firms with this that have done this with loads of firms. Today's sponsor, TeamUp, they will help you do this for the first time if you've never done it before. TeamUp's actually run by a former bookkeeping firm owner who just had a hard time finding good accounting talent offshore, so they went out, they hired a bunch of recruiters in the Philippines, and they pull in the very best people. By the way, I mean, all the, all the top 100 are investing in the Philippines. Like, there's a tremendous amount of account accounting talent in the Philippines, and so they're pulling in bookkeepers, tax folks, executive assistants, whatever you need. And TeamUp specifically has a direct hiring model, which means all you do is you pay TeamUp one time a recruiting fee to go and find you your next dream hire. It's, it's as simple as that. But if you haven't done it before, you can have all sorts of questions and blockers and all that, right? Don't just sit with that stuff. Talk with TeamUp because they've helped a load of firms navigate these exact same questions. Learn more about TeamUp, check out the link down in the show notes. Interesting times comes after a whole bunch of US tax pros just got over the tax deadline. Congrats, you did it. What are you doing listening to my podcast on a Friday? Very excited to keep cranking on content and useful stuff to help firms through this transition. Can't recommend enough coming out to the live workshops that are happening in May and June. I mean, this is going to be the story of the room is all the people working on this stuff, trying to figure this stuff out. Imagine meeting 10 other firms that are local to you that are also working on this. That's what makes these workshops so, so valuable. Doing them in 20 cities. Want to see you out there. Uh, go to JasonOnFirms.com/events to see all the dates and cities, but let's get to the mailbag. Mailbag, mailbag, mailbag, mailbag. This segment is called Mailbag. Questions are submitted, uh, via DMs on social platforms, via email. My email is Jason@JasonOnFirms.com, via my cellular telephone. Phone. My number is 1-516-980-4968. Operators are standing by. First question, when are you dropping a Spotify playlist? Oh, when are you dropping a Spotify playlist of your transition tunes? 100% would subscribe. What this individual is referring to is my sweet, sweet little transitional stings, bumpers, the graphics and music that take us from one segment of the podcast to the next. Oh man, that's a great idea. I'm incredibly proud of my transitions. Not that they oftentimes are at a wildly inappropriate sound level compared to the actual like me talking. That is more of a mastering issue. So if you ever got blasted by a transition, I'm sorry about that. I've got so many dumb AI songs that I've made that I think are hilarious. They get like bounced around on our Slack and I'm like, should I share this externally? Does the world need any more of this? Also, I don't really want to be the guy that puts like AI songs into Spotify. When we did a whole album of lo-fi remixes of the IRS hold music a couple years ago, that's still on Spotify. Just type my name and you'll find it. It's exactly what it sounds like. We hired real-life human music producers to put that together, and I'm proud of that album. It really slaps. But to make like a bunch of AI songs and put them on Spotify, I don't know, man. I don't know if I'm that guy. Hey Jason, what's your take on companies that do marketing for accounting firms? Is this just a myth or are they actually helpful? What's your take? My take is 99% of them are hot, hot caca. They're not good. They are ever so happy to take any dollars that you'll give them to then pour into paid ad spend, but they get to keep X%. And this is the business model for a lot of them is they just take a percentage of your ad spend. And as long as you have a marketing budget because somebody told you you should, they keep getting paid because they keep running ads and they even have these great ad templates for accounting firms. You don't even have to do the work. If only it were that easy. That easy. You know better than that. And I know a lot of you have gotten burned on this before, sinking tens of thousands of dollars into something that brought you virtually nothing. If it were that easy, everybody would be doing it, right? The prerequisite to paid acquisition working on any platform, that is like spending money to put ads on Facebook or LinkedIn or anything like that, the prerequisite to those ads working is the same content working organically. That is you just going to a platform and you posting the thing and people going, ooh, I want this. Here's my email address. Here's whatever. How do I download this thing? That's a prerequisite to a paid ad working. And so the easiest example in my business is our app recommendation database. It's a giant Google Sheet of like over 150 apps for accounting firms, which ones I recommend, which ones I don't. Over 8,000 accounting firms have used it, and I can post that anywhere, and people are like, yes, sweet, thank you, this is valuable. And so if I wanted even more people to see it, I could put paid spend behind it, and then even more people would use it. But paid acquisition only really works when you have something that is already working organically and you just want it to go to more people. Now, some marketing agencies will help you find what that thing is for you. It took years for me and my journey to find things that consistently landed with accounting firms that were really, really useful. By the way, that's the North Star. Is it useful? And some marketing agencies will help you to find that. Many of them don't, though. Many of them are just slinging paid ads and that's That's it. That doesn't work. And so honestly, 95% of them, no good. 5% of them can actually help. It is worth also pausing to consider what problem are they solving? Because I mean, have you gone to a trade show? Like, do you know like what type of person that you want to work with? Are you showing up in person? You will get a lot of leads from that stuff. So I do think sometimes pulling in a marketing agency to do like ads for you can be a bandaid for a larger problem. Next up, we are a mid-tier firm, about 25 people. With distinct sort of separate teams— sales tax, payroll, individual tax. Our monthly clients on average pay us several thousand a month, but they don't have one point of contact. But they don't have one point of contact. And the biggest complaint is I don't know who to go to. They said they did hire a client concierge and they don't want to train an accountant to be like a jack of all trades. Any suggestions? What's the best way to navigate this? It depends on how many things your accounting firm sells. If you sell two things, bookkeeping and tax, you probably don't need a layer of staff above that. Certainly if you only sell one thing, like you, they go to the tax manager, they go to the accounting manager. You get beyond two things and you do kind of need that additional person. And the way to think about that person is they are the client's advocate inside of the accounting firm. They're making the client feel heard, making them feel understood, and they're making sure that all the work is going through as it needs to, to ensure that client is taken care of. And if the people in your firm who are tasked with caring for clients are also the people doing production, they've got a lot of responsibilities there. And so what the client wants is, is inevitably at odds with what work they need to go out the door. And this is where like clients end up feeling forgotten, like we're not as responsive as we need to be. And so like you can even make an argument if you're only doing tax, if you're only doing bookkeeping, if you're doing those two things, you can also make an argument for a concierge layer above the production people. People just because that is such an important problem that your clients feel understood and heard and all that. But definitely, if your firm's doing more than 2 things, you gotta have that client concierge level. By the way, you can find accountants that are very happy to do that, who want to get out of the production stuff, want to do more client-facing stuff, and it will make a tremendous difference when 100% of that person's focus is on making the clients feel good. That's a— I mean, that's a vastly different client experience, as you can imagine. Next up, I'm buying into almost everything you're selling except niching If I only work with dentists, I'll put my head through a wall by March. How do I narrow things down without dying of professional boredom while still delivering that high-value, I know your industry expertise? Well, first problem is you've selected the most boring profession imaginable in dentistry. The type of people that grow up and say, you know what I want to do? I want to spend all day with my hands in people's mouths. Those aren't people that I want to spend all my time with at work. Now, that being said, so like the obvious answer there is like, oh, go find something you're more passionate about. But even on the boring stuff, a lot of people would tell you there is joy to be had in mastery of your craft in, in becoming an expert. In fact, Cal Newport's whole, whole sort of messaging is don't chase your dreams, buddy, develop expertise and find satisfaction, find joy in that. And my take here is that both are true. I think the worst career advice anyone can be given is, hey, you know that thing you love? You know that thing that you really, really like and you can't stop thinking about? That's the last thing you should to do. You don't deserve that happiness. Come over here and do this instead. Save that for after work. No, like, you're gonna spend the vast majority of your waking hours in your lifetime working. Why not do work that you love? But at the same time, I think you can also find joy in becoming an expert at something boring, at something lame. We are, after all, accountants. Case in point, you can find joy in the thing that isn't sexy. Oftentimes it's less crowded space to hang out in anyways. A few more questions to go. How do you actually present your 3-tier proposals that you develop for a client? Do you do a discovery call? Do you bring them back for a second sales call? Do you walk through it with them live? My take on 3-tier proposals is you always initially send it to them asynchronously. That is, I'm not walking them through it live. It's just like delivering a set of financials or delivering a tax return live. If you do that, you do that reveal, that ta-da moment while you're with them, all you're doing is sitting there and watching them process. Like, there is not constructive conversation that comes from that. What I need is for them to go through the proposal on their own, have a think, pop into the cold plunge, sleep a night or two, and then let's circle back and talk about how you're feeling, what your questions are. That's a much more productive conversation than saying, let's hop on a call together for me to then unveil this thing, and I can watch either in shock or horror at how you react. Now, the delivery of that proposal, it's important. Here's the proposal, but let's hop on a call. Here's my booking link. And so if you factor in like, what is the size of that proposal? Am I sending out 2,000 proposals for all my 1040 clients? Geesh, I don't want everybody booking a call. By the way, most of them won't. They'll just accept the agreement and move on with their lives because they don't want to talk to you any more than you do. But you can set a dollar threshold there to say, like, if it is a pretty big dollar agreement, in that communication you say, Hey, let's connect next week. Here's my calendar. If they don't follow up, you give them a call. Like you get, you get a little more on it as opposed to the more laid back version where I do think almost always you want to make yourself available to talk about it if they want to. The more laid back version being like, if you got any questions, here's my booking link. And if you're doing a crazy volume of this stuff, I mean, 2,000 1040s proposals shouldn't be coming from you. It should be coming from the account managers, the folks that you want sort of taking the lead on that engagement. Engagement, but I virtually never will walk somebody through a proposal live for exactly the reason we discussed. I don't need to sit there as they process. I want them to process and then I want to come back and have a conversation together. Next up, did you see what Perplexity launched? So Perplexity, big AI company, one of the first ones to nail like AI search. They've launched a couple things that are getting all up in my yard. They launched a QuickBooks health check. They also launched a landing page that's like, hey, come prepare your tax form. With Perplexity. And what I think they're basically doing is they're taking things that AI models can by and large just kind of do with like maybe a little bit of special instruction, but it doesn't take much. And then Perplexity is standing up landing pages around that flex, that's like certain functionality. Because if you think about having something like a Claude, how do you advertise Claude? 'Cause it'll kind of do anything. So like what does the messaging look like? Perplexity, I think this is smart. They're just creating loads and loads of landing pages that are targeted at very specific high-value individuals, people who will come and happily pay a hefty monthly fee for it to do this valuable work for them. So it's not, they're not creating landing pages for like these little sort of trivial, like meal planning, all these different things. They are saying, this is a business problem. Here's a landing page. Let us onboard you into Perplexity. We'll show you how to solve this business problem. It doesn't mean that they have some secret sauce that nobody else has. It doesn't mean that they're better than like the professional tools for accountants that do the same thing. And that's the only problem they work on all day long. When I see this from Perplexity, it's just a growth strategy. It's just a way to speak specifically to accountants or specifically to business owners to say, we're a great platform for you. In my mind, when I see that, like no part of me is like, oh no, Perplexity figured out how to do what we do. That feels ever so far away. Last question. I've been watching your YouTube videos and wondering if you work Do I work with Australian firms? If I work with Australian firms, that's a great question. I like to think of myself as like the Apple microfiber cloth of accounting thought leaders. If you're familiar with this, Apple has a microfiber cleaning cloth. They call it something else. It's a fancier name on their website. And the product has a little toggle section that is compatibility. And it lists a whole bunch of Apple devices that their microfiber cleaning cloth is compatible with, and it's big news whenever a new thing is added to that or something is removed from that. Great news. Now you can use your microfiber cloth with this additional Apple thing. I don't work with accounting firms. Like, I don't do like consulting, anything private, nothing like that. But I'm also from the US, and so my experience is in the US, and so I talk a lot about US tax. I know like non-US folks don't like that because it's very region specific. I would say I'm compatible with an Australian firm in the same way that I'm trying to be compatible with as many firms as possible. By and large, I would say like The advice and the perspectives I try to put out, I try to keep fairly generalized so that they are relevant to as many firms as possible because every country, man, every country needs a Jason Staats. I wish more people— that sounded incredibly narcissistic. I wish more accountants would go out and put out valuable thought leadership through a more specific lens for Australians, for bookkeepers, for moms. Like, what is your perspective and what you can add to the conversation? I want more accounting firms to do that, or accountants to do to do that and take it seriously. Like, you can even do that without running a firm. Um, we just did an episode the other day on like how to get into accounting firm thought leadership. If you, you really want something to light you up, your partner's sick of listening to me— oh my gosh, your partner's sick of listening to my voice— put on a podcast about me telling you how to become an accounting thought leader. They're gonna love that. So yes, I'm compatible with Australian firms in the same way that I think I'm compatible with most accounting firms. But as with all things, you have to read between the lines a bit. Like, some stuff doesn't hold up in different But if you can't find that person, look in the mirror. That person might be you. Got to be the change we want to see in the world, right, people? Gang, thanks for coming and hanging today. Great to see you. You're looking sharp. Whoo. I had a— I had the biggest eyeball blood vessel explosion I've ever had in my life. Happened like 4 or 5 days ago in my left eyeball. Woke up looking like Darth Maul, and I had to shoot something that day. This is the problem with being on camera almost every day. I shot a YouTube video that will live there for years. And half of my eyeball is red. You better believe that's half the comments on the video. So YouTube viewers, thanks for looking deep into my eyes for the last hour. I know it wasn't easy for you. Audio listeners, well played. You made the right decision today, gang. Thanks for being here. Questions, comments, any of that stuff, pop it down in the comments. I'll catch you in the next one.