Claude Skills for Accountants: How to Make AI Repeatable

All right, we are live. Welcome, welcome, welcome everyone. Looks like we already have a fair amount of people. I see 10 people, 11 already in the session. So welcome everyone. Super exciting today to share the topic that we can see here on the screen, all about Claude skills. An interesting feature that we believe it's probably one of the most underrated ones. It's really the fastest, easiest way to make AI repeatable and to turn some of the one-off wins in your firm into firm-wide workflows. Perfect. Just to warm up, make sure that you all can hear me correct, hear me all right. Let's put in the comments where you are joining us from. I'm here in like sunny and warm Santo Domingo. I know we regularly have some of our Canadian folks and US folks joining us, so just posting in the comments, making sure you all can hear, see us, and we are ready to go. All right, Anna from Florida, Carlos from Puerto Rico. We're not that far, Carlos. I know we've met before. Stephanie from Mississippi. Awesome. I've always found that name so funny. Mississippi. In any case, looks like we are good to go and get going. Agenda for today, we are effectively going to walk through why we are so bullish on skills as being the lowest-hanging fruit AI effort you can do, and we'll talk about tactically how you can build your first few, some examples and use cases so you know what are other— what other firms are doing, and also we'll discuss ways to scale it and how you distribute this across your team in a way that's easy to maintain, that can be updated over time as we have uncovered ourselves over the years. As you might have said on the announcements that we're on this for this webinar workshop, we currently have available An AI accelerator. This is a group offering we are launching to have 6 accounting firms at the same time join us to figure out where AI can have the biggest impact. And we already have 4 firms signed up. Just to give you a sense, like some are as small as 3 people just starting to figure out fraud, so a little bit less sophisticated. as high as firms that have 25 or larger in size, have built custom tools, are a little bit more sophisticated. We kind of have anywhere in between. The doors are closing this Saturday when the 6th firm signs up, and we'll be diving into questions in the Capacity Accelerator towards the end as well for those of you interested. But let's talk about the problem we are solving. I mean, I don't know what more powerful tool than AI we could have possibly been given. And if you're anything like most firms, you might start to realize that even though we've given this incredibly powerful tool that can do so many things, you give this tool to your team and outcomes vary so much, right? Perhaps you as the owner are super excited. You start to learn about it. You have some prompts that work perfectly. You have a ton of skills, schedules, automations in the backend doing a ton of stuff for you, and you're getting the most out of it. Yet perhaps someone else in your team is like doing stuff that's half the way there. They don't feel it's worth it. And perhaps there's other ones in your team like Mary here that they find it unusable. They don't even use it and haven't locked it in. What's the difference? You know, how could it be that We're all provided with this incredibly powerful tool. Some are getting the most out of it and others are barely getting value from it. The answer, at least part of the answer, our perspective, it's the context and the instructions that are provided. And skills are meant to standardize that and enable you to figure something out that works, that works reliably, that works consistently, and being able to distribute it to the rest of your team. We are so bullish on AI skills that actually, as an AI agency, among dozens of things you could possibly do internally with AI, we've chosen specifically actually a rock for this quarter that Quanee, one of our senior client managers internally, is leading to effectively build a set of skills, a firm skills library. As a company goal as it relates to AI. We do believe and are bullish on it being, as I have already mentioned, the fastest, best, and easiest way to turn a one-off win into a firm-wide core club, and we'll be discussing that in a little bit more detail. All right, enough hype. Let's get into the nitty-gritty. 3 things we're going to learn today. Number one, what AI skills are and what they are made of specifically, right? Just definitions, understanding what they actually are can be helpful. Then we're going to be talking how to build one. What are examples that we've built, that we've seen other firms build, and what are the ways in which you can build the first one for yourself? And lastly, we're going to talk about how to scale it once you've figured out a skill that works. How do you put that in the hands of the team in a way that they can actually put to use, in a way that you can track adoption and usage, and in a way that is not a burden to keep up to date? So hope you're all excited as I am. Definitely, as we go through it, drop in some questions. We have some other folks. See, Andre from Ottawa, nice to see you. Gary from Ontario. I knew we had a ton of Canadian folks. I know you always show up, so I appreciate you all. Monica from Florida, good to see you here, Monica. As you have questions, please drop them in, and I'll be pausing as I share and discuss each one of the topics, and I'll come back to make sure I'm answering your questions as we go. All right, let's start. What is an AI skill? I've hinted at it multiple times already, but in summary, an AI skill, and a Claude skill specifically, is the easiest way to teach Claude to do a thing your way every time. Again, the easiest way you can teach Claude to do a thing your way every time. An emphasis there on your way, right? That your way of doing things, your standard. The way things are done in your firm, and also emphasis every time, right? How can you get more consistent results and outputs? So one of the challenges with AI historically has been that even though it's an incredibly powerful tool, sometimes we have challenges with reliability and consistency, right? You get an output one day and then the next day you try something similar, you don't get the same result. This is part of what's going to help you actually get consistent outputs. Out of out of out of water. Now, a good analogy that I've heard some people suggest is like this is like an SOP for an agent, right? Essentially saying, hey, you have a process that's documented steps you need to follow through that you would have traditionally given to a human being to act on, right? And you know there are some there are some challenges with that ensuring that people actually follow the process and they don't skip. 17 of the 20 steps you played out, right? This always had some challenges in terms of getting a human to consistently follow your SOPs. But this is a common analogy. I actually like to think that it's still underselling because I think it's more than that. The analogy I like, and perhaps this resonates with me as an engineer, might not for some of you as accountants, but I like the analogy better of software. And the reason is simple, right? Like, an SOP only documents the work. A feature is expected to perform the work, actually do the work for you. Do the work on your behalf. As we get into things, you'll understand why I think that is a more accurate analogy, but that's the analogy I'd like to propose that a skill is. It's more akin to a software feature. That's simply an SOP for an agent. Now, here are the parts that make up for a skill. As we mentioned, skill, easiest way to teach a bot how to do a thing your way every time. And these are the components that ensure that that's the case. There are 4 components, only one of them required. But as you get more sophisticated and mature with skills, you realize that the other 3 are actually what make the consistency happen. So the first one is a skill.md. .md, for those of you that don't know, it's simply a file type that it represents a text file. So imagine, think big prompt, right? Skill.md, just think a big prompt. In this case, it's a big prompt that has the metadata, right? What's the name of the skill? What's the description of the skill that has to tell Claude when to use it? And then effectively the instructions and the steps for Claude to go through. So that's number one. You have a skill.md file, which is the big main prompt. It is required for everything. And the most basic skills would just have that. Again, as you get sophisticated, you start to learn the power of skills. You realize that it has 3 other potential or optional components. The second we have here is scripts. Effectively, I can get Claude to run something consistently by having executable code within it. Again, hopefully you start to make the connection of why I like to compare this much more closely to a software feature, because if I want to build an Excel workbook consistently, just as an example, and format it in a specific way, I can make sure to put in the skills.md file for Claude not to build it based on the prompt, but based on the script that, by the way, he can write too, right? So don't get intimidated because we're not necessarily talking about you manually coding that script. Claude will still write it, but the knowledge of knowing that, hey, the prompt is a suggestion, the code is a command, right? So we want to make sure that whenever we want something that can be done deterministically and we want it done consistently, that we transfer part of that prompt, which again is guidance, into a script, which is command and consistency. The other 2 parts of a skill are references and assets. And they're a little bit of the same thing from my perspective because they are just files. But references, a good example of reference might be something like a glossary of terminologies, perhaps context about standards, right? You want to provide some static context that the skill might need to refer to, hence reference. And assets, which is also very powerful, and I like to think about assets as what would you provide a team member to increase the chances of succeeding. So in the example of a workbook, I might provide a script for the workbook, but I also might provide a template for that workbook. So it has the highest chances of getting done exactly the way I want. So again, AI skills, easiest way to teach Claude to do a thing consistently, and it's made up of skills and skills that can be filed, but also scripts, references, and assets. And as you get more sophisticated, you learn to ensure to incorporate those things to get better. Now, why are we so bullish on skills and what makes skills What sets it apart from other things that you may have within Claude or within ChatGPT, like projects, artifacts, all of that? 3 things. As we mentioned already, it's the simplest way to automate. And for starters, we believe it is the lowest hanging fruit because, as you saw, it's effectively English and files for the most part, potentially scripts, but also, as I mentioned, not something you have to necessarily write. By him. The other thing I like is it is extremely flexible and insanely portable. Let me walk you through that. A skill can call another skill. A skill can be used within a project. A skill can leverage connectors. A skill can build an artifact. And I can keep going. A skill can be used within Claude Chat, Cloud Core, Cloud Code. A skill can be used within clouding itself. Actually, we have recently built a financial review skill for an accounting firm, and they already have some workbook templates that they use to pull the general ledger, the balance sheet, the P&L, and they didn't want to have to copy that into Excel— sorry, into cloud, run the skill, then bring that context back. was as simple as, hey, pull Claude in Excel, you can use the skill right then and there. It's the case about ChatGPT in Excel as well, by the way. It can also be used in Claude in Chrome. Again, we've seen examples where firms want to scrape a website for analytics. We have this firm that came to us that they were struggling with a specific, like, Airbnb reconciliation because they needed to have the detail that was for each reservation that was not coming to the bank feed. So in this case, they built a skill within that they could trigger within Claude and Chrome to open up Airbnb and then effectively get the data they needed to run the reconciliation. So it is extremely flexible. It can be used pretty much anywhere, and it is extremely portable because you can use it in any Claude product. And here's the crazy thing. Because as you saw, if it scripts files and just is still going to be file, it is even so portable that you can then take that skill and put it in another AI tool. What a beauty, right? You can just dump that into ChatGPT if that's the one that you're using. So skills are incredibly, incredibly flexible in that way, and that's why we are so bullish on it. Another thing important to say about skills. And I was just getting out of the Future Firm Community. We had a Q&A literally 30 minutes ago, and they were talking about, hey, I'm struggling to use AI more efficiently because I'm paying $200 a month per user because we're burning so many tokens. And one of the things I told them is like, actually, see if any of the things that you're doing today and you have a project for, you can use Skillful. Instead, because skills are also more token efficient, and here's why. A skill uses something that Anthropic calls progressive disclosure. What does that mean? It actually, it just like imagine a human that has to navigate through, actually I like this analogy better. Imagine you're working on your desk, right? And you have all the files you need. That probably will be the equivalent of a project, right? You have a lot of static context, And instructions that are always loaded. Skills, on the other hand, would be more like a cabinet, right? Of like, okay, I can see the name of the skill and this is what it does on the backend. It reads through to understand which skills you should use. And then once it captures the skill, it essentially reads the skill that we call at that point and then navigates to the various scripts, references, or assets. As needed. So it can be also more tool-efficient in that way. So a lot of things to like about skills: flexible, easy, portable. Really, I don't see any, any downside for the most part for getting it done. And hopefully you are at this point hyped up, excited about skills, and understand what they are and why we're so bullish. So that's number one. Hopefully you all feel good about understanding what SQL are. You've learned something new already. Let me see if we've had any comments or questions come up in the chat. We just— I just see a few more. Hi, Marielle. I think I saw you earlier today. Kyle, etc. Again, if you guys have questions, feel free to drop them in. I'll just pause. as we go to answer them for you. Let's talk about the second topic, the most important one, more likely the most tactical, which is how do you build one to begin with and what are some examples of it? Okay, there's really 4 ways to build a skeleton, and these are almost like the order in which we would recommend. Number 1, It is after the fact. What do I mean by after the fact? If you've already done something challenging, you've done a challenge journal entry going back and forth a ton of times, you've like written the post exactly in your tone, you have, you know, reviewed a set of financials and caught some issues and have to like correct it multiple times, you've built a workbook for a financial model and you've kind of like already discussed with Claude specifically the way you like it, It already has that context, and with as simple as telling it, turn this into a skill, it will do so and you can save it. It's really as simple as that. I'll show you what that actually looks like in a second. The second way you can build a skill is to actually record yourself while you narrate your work, right? So the clicks, the typing, the— Effectively, your thought process as you go through it. It's another way too to build an AI skill that is a feature that came up, I think, probably like 4 to 5 weeks ago, and I've heard a lot of people having a ton of success with it. I myself have used it many times to come up with some skills for our own team and our own processes. If I may just share one example that I did last week, we are introducing a quarterly bonus for our team to incentivize them to be more efficient. And there's all these things I have to do, taking data from our financials, taking data from our project management system to understand production across different team members and ultimately spit out what's the specific bonus for each team member. That's something that I was able to fundamentally run through a skill by doing it once, sharing that skill with my administrative team, but now they'll be able to run those calculations more, much more easily, much more accurately, and now their job is just to confirm. that those numbers aren't correct. As opposed to trying to learn how to manage this workbook and this and that and these formulas, etc. Okay, third way to build a skill is before the fact. If you already have your process documented, you have a recipe of how you do things typically, you can share that and tell, hey, this is the process I usually follow. Can you build a skill from it? I don't like it compared to doing it after the fact because you usually, if you've done it after the fact, you've already like had some back and forth. you have corrected the model several times and it can kind of take some of that context to actually make sure your first version is much better. That's an option as well. And of course, you also have the option to get someone else's skill. So you can install it from somebody else. Although it's worth noting, and here I kind of hint back to my analogy of software architecture, because a skill can have executable code, We want to make sure you vet it just like you would vet software. So because this can technically run, technically you could, you know, send this to someone else's database. There's some data protection protocols you might want to keep in mind. At least just, you may want to drop the skill and ask Claude if there's anything weird in it, or like read it to see if it doesn't have code or not. It's something that might be worth mentioning. So make sure you download skills. From trusted sources. And if you want to go a little bit deeper, that is advisable as well. Now let's get our hands dirty. Come on, let's see what that looks like in Fluent. And for those of you that are getting started, it's actually a piece of advice that you'll share for that Q&A of accounting firm owner that was in, is like, if you don't know where to start, connect your calendar, connect your inbox, and ask Fluent, and it will tell you. And this is what that actually looks like. Let me go ahead and share my screen so we can see what does an actual— what does a scope actually look like in practicality. So I will share my cross-screen right here. This is the desktop app. Highly recommend you install the desktop app. I'm not as critical about continuing to use the browser because Core work code. A lot of the features are coming into the browser environment as well, so you're more comfortable with that. And if you don't want to download a new software into your computer, but a ton of benefits and a ton of features typically coming earlier to the desktop application, and it's the environment we recommend. If I go to customize, this is actually a new interface that they built. I can see not only the software skills that are available. My Dropbox directly. But if I click on yours, this is where I'd see the various skills that either I have created, one, that are part of my organization, which we're going to talk about organizational skills and plugins later. We talk about how to scale this, but also skills that have been shared with me. Okay, so this is almost like my set of skills library. And I have also from Anthropic as a partner. If I were to open up one skill, just as an example, the skill I used to build this presentation, I can see here that it has an overview where it has a description. This is building an Observe-branded PPTX deck, few words per slide, few slides, speaker notes, and use it for any deck that Isaac from Observe will present. Right, that's the description. So it knows what it's doing. to use it. And if I go to content, this is where I'll see, as we talked about earlier, the big file that walks through what are the rules, what are the exact steps I want you to take, like use the option branding. Guess what? That's another skill that I'll have to reference. And then I have references and examples and assets and scripts as needed to make sure it does it consistently every time. Single time. All right, so that's what a skill looks like, where you can find them. If I were to jump into some of those skills that we will be sharing earlier, I should be sharing later. This is an example of what I just said, right? Just to orient ourselves here, this is the sidebar for the chats. This is the main chat that I'm putting it in, and if you're using the In this case, I can see the plan it's made for itself and any skills or context. In this case, there are not skills because I didn't use it. This was just to come up with skills ideas. And here's what I said, hey, propose 2 different tasks, then roll a sample of the best one. And actually, these are some of the things that it suggested to me that actually made a lot of sense. It's like, hey, you have in your calendar Multiple accounting firm official consultations every week, book through Calendly, and then you send an email. Oh, sorry, this was not it. Because I run this in my environment as well, it is suggested almost like this is just in the pre-call brief. The next time it actually suggested like a follow-up draft that I typically do. Also, it's saying, hey, you have all these check-ins with your team, with Alicia, Juan, Natasha, Pablo. Like, like 6 or 17 members, you have regular one-on-ones with them. How about you create like an incredible one-on-one prep guide just to make sure you're not missing anything from the last time? And also stale leads follow-up. You have some leads or partnership opportunities that you haven't followed up with in however long. Like, I could run a flow for you on an ongoing basis to sweep and flag which of those you might want to follow up with. Next, before they go. And this would be as simple as me saying, cool, let's just say I want to do the 101 prep. I'll click on review, it will show me what that skill dynamic file would look like, and I'll just save that skill and then be able to either say slash, this is Journal 101, be able to see that skill available next time. Or worth knowing, big benefit of skills that I didn't mention, if I am telling it, hey, Help me prepare for the one-on-one I have coming up. It will identify, and without me actually triggering it, which matters for adopting it with your team, because I'm talking about something where the skill could be valuable, Claude itself will know to trigger it for me without me using the slash. So it can be triggered both with the slash, but it can be also auto-triggered based on the context. You are on the request you're asking from Claude, assuming it is paired with a skill you already have available on that account. So that is, my friends, what skills look like. And let's talk you through some additional examples that would come across. Example that we have seen from Value, example that we see from— Papers, and that's part of what we do here at Officer for a Living. So there are 4 I want to share, and I'll probably dive deeper into 1 or 2 of them. As you guys read these 4, probably not going to have time to go through all of them in detail, so put in the comment section if there's one in particular you would appreciate I dive deeper into and actually show you an example of what the skill looks like and what the output of the skill Actually, yes. But 4 skills that we shipped recently. Uh, first one is a tax return loading script. This is effectively a skill to allow a firm to drop a tax return into Claude and have it spit out a script for this tax partner and this tax manager to read through, to send a long video to the client, and just make the reporting process a little bit easier on them, and also to cut down Potentially unnecessary meetings for their— during their busiest time of the year. Payroll journal entries is one that we come across so often. It's such a painful process where you have the payroll report with overtime, time off, taxes, benefits, this and that, 401s, and perhaps you even need to allocate this payroll by department and classes, whatever. Like, we've come across with this so Super convoluted payroll reports that ultimately have to be turned and coded into the accounting software. Super, super useful use case for Skill for you to build effectively. These are the reports you need, and this is the output I need to import this journal entry directly into QuickBooks. And for effectively Claude to just do all the work to get you from input meaning report to output, which is like the import-ready file for you just to like hand to QuickBooks and output for you. And if you get a little bit fancy and use our QuickBooks MCP connector, potentially even send the journal entry right there from Claw as well. Financial review, another big area that we come across fairly often from the perspective of ensuring that your team hits catching issues and that perhaps a more senior individual usually would before they call the books done, almost like allowing the team to self-review themselves a little bit more. This is a use case that we see being helpful both for accounting firms or financial statements, for tax firms on the tax return side, and pretty much any deliverable that gets out the door. You can potentially use a review skill. to have your team members build that, you build your standards into it and make sure things get cleaner to you for the reviews at the end. This last one is actually a recent, very recent project. We're halfway through at the moment where we built effectively a process to reconcile bank that was taking a company 30— that would have taken a company 30 to 40 hours. They even stopped doing it because it was taking them so long and convoluted. And we figured out a way to download the specific reports we needed from the bank, the bank statements and the general ledger, to effectively turn that 30 to 40-hour process into a 15 to 30-minute task for that particular firm. So let me see in the comments if I have Anyone suggesting which one they want me to dive into? I don't see any comments related to that. But let me share— let me share the first one at the very least. Stop screen. All right. So we're here back in Claude, and these are skills I've already run. For the interest of time. But as I mentioned, I hinted at, I do want to give you the context of what this is meant to do. For obvious reasons, this is going to be dumb data and not data from clients we've done. But this is a skill that we've developed to run accounting for recently. What's the problem that we're trying to solve? So we talked to them. They had about 500 clients. On the tax side, they had one of the partners and a tax manager that were having literally a meeting, a 45-minute meeting with 85 to 90% of all of those 500 clients. So imagine that at the busiest time of the year, the most expensive resources of this company were taking 45 minutes per client to take the time To walk them through the return, which of course these clients appreciated, and we acknowledge that. We're big fans of doing good for clients, walking them through things. But what we realized, and based on experience working with other firms— shout out to the folks at Admiral Financial. Gareth, I know you're listening in if you're still there. We've seen a lot of firms have the best practice of saying, hey, why don't we send a Loom to the client first, and then only if they want to talk to us, we can still give them that option. And most of the time you'll find that you'll cut down 50, 60, 70, 80% of your meetings. Your clients will continue to be happy with your service. And unless you have a good argument to actually meet with a client, perhaps, you know, you can support them in other ways, they have other needs that you can fulfill, you can potentially upsell them other services. Like sometimes it can really be, let's be honest, a waste— I want to say waste of time, but it can be unnecessarily a burden on tax managers or partners to meet with every single client on the busiest time of the year. And you have like your calendar back to back to back to back on the busiest period. So this is what they want to solve, and this is what that skill looks like. As we mentioned earlier, you should see at any point be able to click on this skill from this bar to see what this skill is made of. As you can see here, it's generating a spoken style script. From a completed text return PDF. That's the purpose of it. And you can see that as we brought up earlier, it reads character, identifies what's there, drafts script, blanks the gaps, and then simply ready. And also, as we mentioned, this is an example because it's a relatively simple skill, but it also has some steps, right? This is one of the things that as we usually learn, clients are like, hey, give us some examples of how you usually would right one. So now it has that as reference examples, perhaps this target as well, so it puts it in your words, in your language. And this is kind of just an example of what that output ultimately looks like. It can be found in this case to say, hey Michael and Sarah, your tax return is finished and I want to talk you through how everything came together. This is your income, this is how much you made, this is your W-2, and then walking them through What are some of the deductions that they have, right? Like, you use the standard deduction, which for a married couple is this, and on top of that you have a qualified business income from this and that. And ultimately walking them through what are all the deductions that you suggested, how much they have to pay, how much they'll get in return, and ultimately walking them through that process and have a simple script that within 5 to 10 minutes you can have that boom, boom, boom, boom, ready to go. Hopefully that helps. Probably the one— the other one I'd like to walk people through. Let me just read other comments. He says we created one for Canadian sales tax. Awesome. We actually have one for Canadian— for a US-based restaurant-focused accounting firm where sales tax is whole mess and helping them, uh, not only understand where are the discrepancies, but why do those discrepancies exist in the first place. Um, and you're saying here it looks at Google Drive, reads invoices, then look at QuickBooks compliance, posted transactions, reports, the IGCS, all the work you need to do. Perfect, that's excellent. Thanks for sharing. All right, uh, you guys want more examples in a little bit more detail? Let's go through the payroll journal entry very quickly. So this is usually what the payroll journal entry usually looks like. It's like you have a report. Let's see if I can pull those reports here. You have a report— ah, not found. But you have a report that usually has— actually, I think there's another one. Let me see. Yeah, I mean, I'm moving these tiles all the time, but effectively the idea is like, hey, you want— you have a report that has like the details of all the employees of a company and their gross and pay and taxes, right? But you also have a separate report usually to like sum up taxes and all of that. And here it's like, hey, I need to run this report and go through and create a journal entry from it. And you can see how it starts to split pay periods, gross earnings, So you can hold these taxes, 401s, etc. Sometimes it might ask some clarifying questions like, hey, how do you want to book this payment? Please upload the general account, the general ledger, and which chart of accounts so I know which are the accounts available. And ultimately, after giving that, it kind of shows you, hey, because I asked it to make a new one by month, it's saying, hey, for the month of March, I had to split it into these. There is credits, makes it a table so you can very easily see that that adds up. This is the rest of the payroll fund that will fall on the following month, and it gives you all the splits as required into the various accounts. I believe that then I asked it, oh, can you now create this account that I have connected? In this case, like the final import-ready file, then pushes into QuickBooks. So hopefully that helps give you like some use cases, some better understanding of what the skills look like. Let me just keep going through our presentation. All right, so hopefully you guys feel good about understanding what AI skills are, why we're so bullish on it, what it is made of, and we've talked about 4 ways to build one and 4 tangible high ROI examples, super time-saving opportunities that we've come across just as examples. Now let's talk about how to scale. Does he— who here has a team? Me. Or it's part of a team? Any, any hands? It's very easy to do like a one-off thing that works very well for you. It's much more challenging to roll that out to your team, to get them to adopt it, to keep this up to date. There's just some challenges We have found as we build some of the AI skills ourselves that become very, very challenging. So Monica says, me, Monica, I know you have a small team there around you. I know you are probably the AI champion, and some of your other team members may not be as excited, so this will be very relevant. So there are 4 ways to create now like a firm-wide library, and in order, these are the ones that kind of require more and more complexity, but from our perspective gives you more and more flexibility on 2 areas. Number one, they give you more flexibility in terms of how you can share with others and ensure that they have those skills turned on by default. Why does that matter? We found that, for example, when you share a skill with sharing feature, which I'll walk you through in a second. For some reason, maybe Cloud fixes this later, this skill would be turned off by default on the other person's account, and then it's a little bit of a back and forth to just ask them to turn it on. A little bit annoying. The other one is like, what if you want to update it? Now if you share the skill, only you can update it, which is also challenging. So that's kind of like the first 2 versions. Maybe it's just you, You use it personally, you don't want to share it, it's just valuable for you, pretty cool. You now have the ability to share with somebody else, but as mentioned, it comes turned off by default for them, and only you as the author of the skill are allowed to update it. So what solves for that is an organizational skill where it would now allow you to actually have this turned on for everyone by default. And in a more— which requires a little bit more setup, but a plugin, which is effectively a set of skills, you can build a plugin for your company that will have that set of skills. You can make sure it's turned on by default, and then you can allow your various team members to push updates as well. Let me show you what that looks like. So if I present Here again, on Claude screen, you should see if I go to customize again, and if I go to our skills, this is where I mentioned, hey, these were created by me. Let's just say the internal one-on-one prep, that might only be valuable for me. I don't want to worry about that. Oh, now I have another manager that have one-on-ones. I want to go ahead and share it. I can click the share button, and I can put their name specifically. That will come up and I'll be able to share with them just like that. I can also make this available for everyone by officer too. This is the feature that we found to be just a little bit finicky, and even though you can share with your team, you will be the author. You can update it for whatever reason. It is turned off by default. So the next best thing we found is if you go to organization settings and you happen to be an active officer for account, if you go to skills, you'll see now something called organizational skills. Benefit of organizational skills, number one, if you put the skill here, everyone in your team will have it by default, which is amazing because now you don't have to like ask them to let's hop on a call, let's share your screen, let me show you how this skill turned off and let me show you how to use it. They'll have it available by default. And as we mentioned earlier, because skills can be auto-triggered, That's going to mean that now they'll be able to run some of the skills that are valuable for the firm just by using Flow. Because now when they say, hey, have me reconcile the bank account, if you have the associated skill, Flow will know and trigger it right there. The challenge we found with organizational skills is that only admins will be able to update it. So you will have to come here and then replace it. Because we don't need to, we're doing incremental improvements with the In an iterative process, it's more likely that your first skill is not going to be perfect and you might want to have updates to it. So that's where it becomes a little bit finicky with organizational skills, at least it became for us, because now in this example, only me and one of our team members are admins. So they have— if a team member on the lower end of the organization wanted to update a skill or update, but still they had to send it to us for us to upload it here. And although this will probably be more than enough for most people, I do want to show you what the next level usually looks like, which is creating a plugin. A plugin for us will solve all the pain points we talked about earlier. We can make it required, but we even have some options like can be installed by default, can be available by default. We can make it required for specific team members. So now if we wanted to create a plugin for departments, we could do so. And it gets a little bit more fancy, but because this is in GitHub, it's actually everyone is able to push updates to that without having to go through, ask the admin if we want to immediately allow it. So this effectively blows up our ability to share skills with other people in our company, have them turn on by default so that they can get benefit from it, but also allowing them to also push updates to us for the people that make sense so that we can approve those updates super easily and have the latest version already available for everyone right away. Those are essentially the 4 ways in which you can scale the Scale a set of skills. So, and that's really it. I hope that was helpful. Let's see if we have more comments here or questions. Bo was just about to ask the question about the control of the data of the skills. Thanks for sharing. Yeah, as I said, because it's a GitHub repo, you have so so much control over who can see it, who can edit it, who can only push a pull request that then you have to approve. Like, there's a lot more control you have. Literally, GitHub, for those of you unfamiliar, it's what companies use to host code bases. So it literally has a ton of features around granular control, who can see who can edit, who can update, and it is incredibly, incredibly powerful. We've found in our experience. So there you have it. You've learned what AI skills are, what they're made of, how to build one, 4 examples, 4 ways to build it, and 4 ways to scale it. So I hope that helps. And as I mentioned earlier, I did want to take the last 10 to 15 minutes either to answer questions, so feel free to comment. And whatever you are, whether you're on LinkedIn, on YouTube, on the StreamYard, I can see that all the comments will make it to me. I'll be more than happy to answer. But as I mentioned earlier, we're excited to launch for the first time ever a effectively a group offering where we'll have 6 firms join us together over the course of 4 weeks, figure out where AI can have the biggest impact in their firm. and ultimately help to start implementing, knowing where to start and knowing how to start getting value from it. If you go to upserve.co/ai-accelerator, let me kind of walk you through briefly what are some of the things that we expect to go through in that cohort. So this is kind of like the promise of like we want to create together a roadmap to save time in your firm without overhauling your system, without risking client data. We'll discuss the topic of security and be able to create more capacity without hiring. One of the things that we have heard many, many times is like our team's at capacity and I'm hearing about AI, but I don't know where to start. And we do believe AI is an incredibly powerful tool to give and create real capacity in your firm. Went down properly. We have over the last 12 months had a one-on-one engagement where we've walked firms through a process and they've paid anywhere between, you know, $4,000 to $7,000 to actually work with us one-on-one to help them figure out what are the exact use cases that are going to add most value to the firm and in what ways and with what automation tools it would make sense to implement them. And what this essentially is, is like doing that exact same process for a much more accessible price point and doing it in a group setting that allows us to do it at a more accessible price. One of the things I have like struggled with, I've been joining and I've joined myself many cloud and AI sessions. Sometimes I feel like they're either too strategic. Right, there's talking about the trends of AI and like use cases potentially, and others are like too tactical, right? It's like, here's where you click, here's the feature. What we're hoping to really have is a more holistic— we're calling AI accelerator that addresses both, right? So it teaches you not only where AI can create real capacity, but what are the tools you can use to start building, right? What is worth automating for your firm specifically? In what order and how do you actually ship it? And the way in which we'll deliver that is we will have over the course of 4 weeks, 2 sessions every week. One will be the more strategic session that we hope that the partners, the directors of operations perhaps join us, anyone that makes decisions around technology and capacity, to learn about AI for leadership, to learn about security and governance and guardrails, and to discuss Together, more importantly, what are the common use cases we've seen that add value, and be able to have that conversation to see what are the most used— the use cases that will add more value to your firm specifically. Simultaneously, we're going to be running sessions every Friday in a much more open-ended technical, technical ask me anything, but you'll be able to just hop on in with me and effectively just talk Creating a skill or having questions about whether I use a project in a skill or it's not working as expected. Like, we kind of walk you through if you're technically stuck, how to effectively be able to bring to life some of the solutions, the use cases that we would have covered up until that point. And just wrapping it up with just a one-on-one to make sure you have a clear plan of execution to either Do this for yourself if you want it, or do it with us. We have some bonuses here as well that you can read through. We will be providing some skills we use internally to facilitate this audit process so that you can run them for yourself. We'll be providing some, like, an AI policy template as well as a switch kit if you're still on ChatGPT or want to move to Claude. We'll also be providing a list of the common use cases we see across bookkeeping, across tax prep, and also across other processes that we've audited and mapped in the past, like bill pay, like payroll, or some of my Australian peeps, insolvent activity statements, and other things we have been able to automate for other firms. And as you can see here, these are some of like the price points for both. the one-on-one format as well as the group setting. As I mentioned earlier, we already have 4 out of the 6 firms joining us and signed up already for the group cohort. As I mentioned, they range anywhere from like 3 to 25 people in size, and they also vary in terms of technical sophistication. But we are looking for 2 more firms to join us, and we are starting Next Tuesday. So we are expecting to close the doors Saturday afternoon. So anyone interested is more than welcome. Effectively, if you go through any of the buttons that says reserve your firm's place, you'll be able to put in your name, put in your last name, email address. And if you still see the option for the School of Business Audit or AI Capacity Accelerator, which should be selected by default, You just put in a small deposit to lock in your spot, and we will charge the remaining of the program on this Saturday. If you don't see that option, or it says— it says sold out, you're already covered for all those spots. We also have, worth knowing, 2 spots for one-on-one audits for the month of October. If that is of interest, sometimes I have some people that have presented this. Oh, sorry, I just realized I wasn't showing the form. So let me go ahead and do stuff again. Please share the link again. So link is opsta.co/aiaccelerator. Now let me go ahead and share my screen again, this time with the window settings so I can switch around. Switch around here. So again, these are the 2 options. We have 2 spots available for the group cohort. We already have 4 firms joining us, and we also have 2 spots for the one-on-one audit for the month of October if that is more interesting to you. You want a little bit more attention, you want us to do this for you as opposed to doing it with us. What I was walking you through earlier is like you click this button, you'll be able to put in your first name, last name, email, and company name, then be able to reserve your spot for the group efficiency audit. If you stay— if you see that it says sold out or something like that, or it's not available, because we already have those spots taken. But as I mentioned, there's also an option for the one-on-one audit. If that makes more sense for your firm specifically. With that, I'm happy to open it up for any questions you guys might have about Claude Skills, about Claude, about AI, about accelerator we're launching, or anything of that sort. I'll be more than happy to answer. Otherwise, I'll let you all go. Looking forward to seeing some of you on the other side. I'll just wait for a few more minutes to see if we have anyone signing in, or otherwise we will be wrapping up. Let me first The slides again, just in case it's helpful for others or the people that catch us up on the replay. All right, looks like no more questions. I hope that was helpful. I appreciate you all for joining. And we are about 5 minutes away from the hour, so I'll give you all 5 minutes of your time back. Thanks for any— everyone who joined us here. Love to see some familiar names and familiar faces. And other than that, guys, thanks for joining. Hope you join us on the next one. All right. Looks like we have one more question. Bo says, do you find people use ChatGPT much or just what? Um, Monica says, thank you, Isaac, always helpful information. I appreciate you, Monica. Uh, Bo, do you find people use ChatGPT much or just what? It's funny, we've been— of course ChatGPT like started this revolution, and a lot of other firms we come across are still using ChatGPT primarily. Over the last probably 6 months, or kind of since earlier this year, we've been proponents of Claude and almost like suggesting Claude exclusively just because it's gotten so much better. It is innovating so much quickly. It feels like ChatGPT is sort of like just copying its features, right? Copying skills, copying this, copying that. So it's definitely a company that is innovating much, much faster. And as we see in our experience as well, a lot of people grasp it. It stands around security, it stands around solving moral issues, right? Part of what kind of got brought into the picture of all the way this year was kind of rejecting the Department of Defense thing around weapons and all that, right? But also just so much better, or at least been so much better to this point, especially for coding, by the way, as well. A lot of the things that counterterrorism will benefit from. So we are bullish on Claude overall. We're seeing mixed results, right? Most firms are actually still in ChatGPT for the most part, and a lot of— some of them are trying to transition them in. That said too, I also feel like ChatGPT is still pretty— it's catching up and it's somewhat comparable at this stage. So although 3 months ago I probably would have told everyone, hey, just move to Claude, I think right now it's Hey, I get the job done. And whoa, I know we were talking about like RockBot. Perhaps that's the next thing. Who knows? Like, this is such an evolving race and changing space that it's hard to tell which is the best candidate. All right. Well, with that, thank you all for joining. Appreciate you all. Again, opto.com/ai. DAS Accelerator if you wanted to consider having us join, uh, with not only us but 6 other firms going through this process. Um, Beau, I don't know if this will be relevant for you if you were to consider this, but it's probably more relevant for US, Canadian, and European folks just because of the time we're doing it in. We might consider doing something for Australian— our Australian teams at some point. I was trying today, GPT acts Uh, Brandon says, I had an issue joining. Would you be sending the recording so I can rewatch it? The same link you use, Brandon, where you are right now on the streams should actually be able to become the replay link too. Alternatively, if you go to our YouTube channel, just type Isaac Officer, you should be able to find the live already available. Either one of those should work for you and for anyone that might have had issues joining. All right. Well, other than that, guys, thanks again for joining me, and I'll see you soon. All right. Take care. Bye.